2012-11-08 18:25:23 +01:00
|
|
|
/*
|
2024-08-14 17:20:36 +01:00
|
|
|
* Copyright (c) 2009-2012, Redis Ltd.
|
2012-11-08 18:25:23 +01:00
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions are met:
|
|
|
|
*
|
|
|
|
* * Redistributions of source code must retain the above copyright notice,
|
|
|
|
* this list of conditions and the following disclaimer.
|
|
|
|
* * Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
* * Neither the name of Redis nor the names of its contributors may be used
|
|
|
|
* to endorse or promote products derived from this software without
|
|
|
|
* specific prior written permission.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
|
|
|
|
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
|
|
|
|
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
|
|
|
|
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
|
|
|
|
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
|
|
|
|
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
|
|
|
|
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
|
|
|
|
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
|
|
|
* POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
2024-04-04 05:20:06 +08:00
|
|
|
#ifndef VALKEY_H
|
|
|
|
#define VALKEY_H
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
#include "fmacros.h"
|
|
|
|
#include "config.h"
|
|
|
|
#include "solarisfixes.h"
|
2016-05-18 11:45:40 +02:00
|
|
|
#include "rio.h"
|
Reimplement cli hints based on command arg docs (#10515)
Now that the command argument specs are available at runtime (#9656), this PR addresses
#8084 by implementing a complete solution for command-line hinting in `redis-cli`.
It correctly handles nearly every case in Redis's complex command argument definitions, including
`BLOCK` and `ONEOF` arguments, reordering of optional arguments, and repeated arguments
(even when followed by mandatory arguments). It also validates numerically-typed arguments.
It may not correctly handle all possible combinations of those, but overall it is quite robust.
Arguments are only matched after the space bar is typed, so partial word matching is not
supported - that proved to be more confusing than helpful. When the user's current input
cannot be matched against the argument specs, hinting is disabled.
Partial support has been implemented for legacy (pre-7.0) servers that do not support
`COMMAND DOCS`, by falling back to a statically-compiled command argument table.
On startup, if the server does not support `COMMAND DOCS`, `redis-cli` will now issue
an `INFO SERVER` command to retrieve the server version (unless `HELLO` has already
been sent, in which case the server version will be extracted from the reply to `HELLO`).
The server version will be used to filter the commands and arguments in the command table,
removing those not supported by that version of the server. However, the static table only
includes core Redis commands, so with a legacy server hinting will not be supported for
module commands. The auto generated help.h and the scripts that generates it are gone.
Command and argument tables for the server and CLI use different structs, due primarily
to the need to support different runtime data. In order to generate code for both, macros
have been added to `commands.def` (previously `commands.c`) to make it possible to
configure the code generation differently for different use cases (one linked with redis-server,
and one with redis-cli).
Also adding a basic testing framework for the command hints based on new (undocumented)
command line options to `redis-cli`: `--test_hint 'INPUT'` prints out the command-line hint for
a given input string, and `--test_hint_file <filename>` runs a suite of test cases for the hinting
mechanism. The test suite is in `tests/assets/test_cli_hint_suite.txt`, and it is run from
`tests/integration/redis-cli.tcl`.
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2023-03-30 19:03:56 +03:00
|
|
|
#include "commands.h"
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
#include <stdio.h>
|
|
|
|
#include <stdlib.h>
|
2022-11-24 21:38:09 +08:00
|
|
|
#include <stddef.h>
|
2024-05-26 17:41:11 +01:00
|
|
|
#include <stdatomic.h>
|
2010-06-22 00:07:48 +02:00
|
|
|
#include <string.h>
|
|
|
|
#include <time.h>
|
|
|
|
#include <limits.h>
|
|
|
|
#include <unistd.h>
|
|
|
|
#include <errno.h>
|
2010-07-01 21:13:38 +02:00
|
|
|
#include <inttypes.h>
|
2010-07-05 20:14:48 +02:00
|
|
|
#include <pthread.h>
|
2010-12-09 11:10:21 -05:00
|
|
|
#include <syslog.h>
|
2011-03-29 17:51:15 +02:00
|
|
|
#include <netinet/in.h>
|
2019-11-11 13:06:08 +00:00
|
|
|
#include <sys/socket.h>
|
2011-04-30 17:46:52 +02:00
|
|
|
#include <lua.h>
|
2012-01-20 12:54:15 +01:00
|
|
|
#include <signal.h>
|
2010-06-22 00:07:48 +02:00
|
|
|
|
Auto-detect and link libsystemd at compile-time
This adds Makefile/build-system support for USE_SYSTEMD=(yes|no|*). This
variable's value determines whether or not libsystemd will be linked at
build-time.
If USE_SYSTEMD is set to "yes", make will use PKG_CONFIG to check for
libsystemd's presence, and fail the build early if it isn't
installed/detected properly.
If USE_SYSTEM is set to "no", libsystemd will *not* be linked, even if
support for it is available on the system redis is being built on.
For any other value that USE_SYSTEM might assume (e.g. "auto"),
PKG_CONFIG will try to determine libsystemd's presence, and set up the
build process to link against it, if it was indicated as being
installed/available.
This approach has a number of repercussions of its own, most importantly
the following: If you build redis on a system that actually has systemd
support, but no libsystemd-dev package(s) installed, you'll end up
*without* support for systemd notification/status reporting support in
redis-server. This changes established runtime behaviour.
I'm not sure if the build system and/or the server binary should
indicate this. I'm also wondering if not actually having
systemd-notify-support, but requesting it via the server's config,
should result in a fatal error now.
2019-05-30 18:44:17 +02:00
|
|
|
#ifdef HAVE_LIBSYSTEMD
|
|
|
|
#include <systemd/sd-daemon.h>
|
|
|
|
#endif
|
|
|
|
|
2022-02-16 13:35:49 -08:00
|
|
|
#ifndef static_assert
|
2024-05-22 23:24:12 -07:00
|
|
|
#define static_assert(expr, lit) extern char __static_assert_failure[(expr) ? 1 : -1]
|
2022-02-16 13:35:49 -08:00
|
|
|
#endif
|
|
|
|
|
2014-07-01 11:30:15 +02:00
|
|
|
typedef long long mstime_t; /* millisecond time type. */
|
2019-11-05 10:14:34 +01:00
|
|
|
typedef long long ustime_t; /* microsecond time type. */
|
2014-07-01 11:30:15 +02:00
|
|
|
|
2024-05-22 23:24:12 -07:00
|
|
|
#include "ae.h" /* Event driven programming library */
|
|
|
|
#include "sds.h" /* Dynamic safe strings */
|
|
|
|
#include "dict.h" /* Hash tables */
|
|
|
|
#include "kvstore.h" /* Slot-based hash table */
|
|
|
|
#include "adlist.h" /* Linked lists */
|
|
|
|
#include "zmalloc.h" /* total memory usage aware version of malloc/free */
|
|
|
|
#include "anet.h" /* Networking the easy way */
|
|
|
|
#include "version.h" /* Version macro */
|
|
|
|
#include "util.h" /* Misc functions useful in many places */
|
|
|
|
#include "latency.h" /* Latency monitor API */
|
|
|
|
#include "sparkline.h" /* ASCII graphs API */
|
2017-03-27 15:26:56 +02:00
|
|
|
#include "quicklist.h" /* Lists are encoded as linked lists of
|
|
|
|
N-elements flat arrays */
|
2024-05-22 23:24:12 -07:00
|
|
|
#include "rax.h" /* Radix tree */
|
2019-09-12 10:56:54 +03:00
|
|
|
#include "connection.h" /* Connection abstraction */
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2024-04-05 16:59:55 -07:00
|
|
|
#define VALKEYMODULE_CORE 1
|
2024-04-04 00:04:51 +07:00
|
|
|
typedef struct serverObject robj;
|
2024-05-22 23:24:12 -07:00
|
|
|
#include "valkeymodule.h" /* Modules API defines. */
|
2019-10-22 10:44:18 +02:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* Following includes allow test functions to be called from main() */
|
2014-11-12 21:58:57 -05:00
|
|
|
#include "zipmap.h"
|
2022-05-22 17:27:54 +08:00
|
|
|
#include "ziplist.h" /* Compact list data structure */
|
2014-11-12 21:58:57 -05:00
|
|
|
#include "sha1.h"
|
|
|
|
#include "endianconv.h"
|
|
|
|
#include "crc64.h"
|
|
|
|
|
2023-04-20 09:49:19 +03:00
|
|
|
struct hdr_histogram;
|
|
|
|
|
2022-11-16 19:24:18 -08:00
|
|
|
/* helpers */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define numElements(x) (sizeof(x) / sizeof((x)[0]))
|
2022-11-16 19:24:18 -08:00
|
|
|
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
/* min/max */
|
2022-05-08 13:12:17 +01:00
|
|
|
#undef min
|
|
|
|
#undef max
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
#define min(a, b) ((a) < (b) ? (a) : (b))
|
|
|
|
#define max(a, b) ((a) > (b) ? (a) : (b))
|
|
|
|
|
optimize unwatchAllKeys() (#11511)
In unwatchAllKeys() function, we traverse all the keys watched by the client,
and for each key we need to remove the client from the list of clients watching that key.
This is implemented by listSearchKey which traverses the list of clients.
If we can reach the node of the list of clients from watchedKey in O(1) time,
then we do not need to call listSearchKey anymore.
Changes in this PR: put the node of the list of clients of each watched key in the
db inside the watchedKey structure. In this way, for every key watched by the client,
we can get the watchedKey structure and then reach the node in the list of clients in
db->watched_keys to remove it from that list.
From the perspective of the list of clients watching the key, the list node is inside a
watchedKey structure, so we can get to the watchedKey struct from the listnode by
struct member offset math. And because of this, node->value is not used, we can point
node->value to the list itself, so that we don't need to fetch the list of clients from the dict.
2022-11-23 23:39:08 +08:00
|
|
|
/* Get the pointer of the outer struct from a member address */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define server_member2struct(struct_name, member_name, member_addr) \
|
|
|
|
((struct_name *)((char *)member_addr - offsetof(struct_name, member_name)))
|
optimize unwatchAllKeys() (#11511)
In unwatchAllKeys() function, we traverse all the keys watched by the client,
and for each key we need to remove the client from the list of clients watching that key.
This is implemented by listSearchKey which traverses the list of clients.
If we can reach the node of the list of clients from watchedKey in O(1) time,
then we do not need to call listSearchKey anymore.
Changes in this PR: put the node of the list of clients of each watched key in the
db inside the watchedKey structure. In this way, for every key watched by the client,
we can get the watchedKey structure and then reach the node in the list of clients in
db->watched_keys to remove it from that list.
From the perspective of the list of clients watching the key, the list node is inside a
watchedKey structure, so we can get to the watchedKey struct from the listnode by
struct member offset math. And because of this, node->value is not used, we can point
node->value to the list itself, so that we don't need to fetch the list of clients from the dict.
2022-11-23 23:39:08 +08:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Error codes */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define C_OK 0
|
|
|
|
#define C_ERR -1
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* Static server configuration */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CONFIG_DEFAULT_HZ 10 /* Time interrupt calls/sec. */
|
|
|
|
#define CONFIG_MIN_HZ 1
|
|
|
|
#define CONFIG_MAX_HZ 500
|
|
|
|
#define MAX_CLIENTS_PER_CLOCK_TICK 200 /* HZ is adapted based on that. */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define CRON_DBS_PER_CALL 16
|
Optimize resizing hash table to resize not only non-empty dicts. (#12819)
The function `tryResizeHashTables` only attempts to shrink the dicts
that has keys (change from #11695), this was a serious problem until the
change in #12850 since it meant if all keys are deleted, we won't shrink
the dick.
But still, both dictShrink and dictExpand may be blocked by a fork child
process, therefore, the cron job needs to perform both dictShrink and
dictExpand, for not just non-empty dicts, but all dicts in DBs.
What this PR does:
1. Try to resize all dicts in DBs (not just non-empty ones, as it was
since #12850)
2. handle both shrink and expand (not just shrink, as it was since
forever)
3. Refactor some APIs about dict resizing (get rid of `htNeedsShrink`
`htNeedsShrink` `dictShrinkToFit`, and expose `dictShrinkIfNeeded`
`dictExpandIfNeeded` which already contains all the code of those
functions we get rid of, to make APIs more neat)
4. In the `Don't rehash if redis has child process` test, now that cron
would do resizing, we no longer need to write to DB after the child
process got killed, and can wait for the cron to expand the hash table.
2024-01-30 03:02:07 +08:00
|
|
|
#define CRON_DICTS_PER_DB 16
|
2024-05-22 23:24:12 -07:00
|
|
|
#define NET_MAX_WRITES_PER_EVENT (1024 * 64)
|
2015-07-27 09:41:48 +02:00
|
|
|
#define PROTO_SHARED_SELECT_CMDS 10
|
|
|
|
#define OBJ_SHARED_INTEGERS 10000
|
|
|
|
#define OBJ_SHARED_BULKHDR_LEN 32
|
2022-02-23 16:15:12 +00:00
|
|
|
#define OBJ_SHARED_HDR_STRLEN(_len_) (((_len_) < 10) ? 4 : 5) /* see shared.mbulkhdr etc. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define LOG_MAX_LEN 1024 /* Default maximum length of syslog messages.*/
|
2015-07-27 09:41:48 +02:00
|
|
|
#define AOF_REWRITE_ITEMS_PER_CMD 64
|
2021-10-25 18:08:34 +08:00
|
|
|
#define AOF_ANNOTATION_LINE_MAX_LEN 1024
|
2015-07-27 09:41:48 +02:00
|
|
|
#define CONFIG_RUN_ID_SIZE 40
|
|
|
|
#define RDB_EOF_MARK_SIZE 40
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CONFIG_REPL_BACKLOG_MIN_SIZE (1024 * 16) /* 16k */
|
|
|
|
#define CONFIG_BGSAVE_RETRY_DELAY 5 /* Wait a few secs before trying again. */
|
2024-04-30 12:45:08 -04:00
|
|
|
#define CONFIG_DEFAULT_PID_FILE "/var/run/valkey.pid"
|
2021-06-22 12:50:17 +03:00
|
|
|
#define CONFIG_DEFAULT_BINDADDR_COUNT 2
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CONFIG_DEFAULT_BINDADDR {"*", "-::*"}
|
|
|
|
#define NET_HOST_STR_LEN 256 /* Longest valid hostname */
|
|
|
|
#define NET_IP_STR_LEN 46 /* INET6_ADDRSTRLEN is 46, but we need to be sure */
|
|
|
|
#define NET_ADDR_STR_LEN (NET_IP_STR_LEN + 32) /* Must be enough for ip:port */
|
|
|
|
#define NET_HOST_PORT_STR_LEN (NET_HOST_STR_LEN + 32) /* Must be enough for hostname:port */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define CONFIG_BINDADDR_MAX 16
|
|
|
|
#define CONFIG_MIN_RESERVED_FDS 32
|
2021-01-28 18:17:39 +02:00
|
|
|
#define CONFIG_DEFAULT_PROC_TITLE_TEMPLATE "{title} {listen-addr} {server-mode}"
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
#define DEFAULT_WAIT_BEFORE_RDB_CLIENT_FREE \
|
|
|
|
60 /* Grace period in seconds for replica main \
|
|
|
|
channel to establish psync. */
|
Unified db rehash method for both standalone and cluster (#12848)
After #11695, we added two functions `rehashingStarted` and
`rehashingCompleted` to the dict structure. We also registered two
handlers for the main database's dict and expire structures. This allows
the main database to record the dict in `rehashing` list when rehashing
starts. Later, in `serverCron`, the `incrementallyRehash` function is
continuously called to perform the rehashing operation. However,
currently, when rehashing is completed, `rehashingCompleted` does not
remove the dict from the `rehashing` list. This results in the
`rehashing` list containing many invalid dicts. Although subsequent cron
checks and removes dicts that don't require rehashing, it is still
inefficient.
This PR implements the functionality to remove the dict from the
`rehashing` list in `rehashingCompleted`. This is achieved by adding
`metadata` to the dict structure, which keeps track of its position in
the `rehashing` list, allowing for quick removal. This approach avoids
storing duplicate dicts in the `rehashing` list.
Additionally, there are other modifications:
1. Whether in standalone or cluster mode, the dict in database is
inserted into the rehashing linked list when rehashing starts. This
eliminates the need to distinguish between standalone and cluster mode
in `incrementallyRehash`. The function only needs to focus on the dicts
in the `rehashing` list that require rehashing.
2. `rehashing` list is moved from per-database to Redis server level.
This decouples `incrementallyRehash` from the database ID, and in
standalone mode, there is no need to iterate over all databases,
avoiding unnecessary access to databases that do not require rehashing.
In the future, even if unsharded-cluster mode supports multiple
databases, there will be no risk involved.
3. The insertion and removal operations of dict structures in the
`rehashing` list are decoupled from `activerehashing` config.
`activerehashing` only controls whether `incrementallyRehash` is
executed in serverCron. There is no need for additional steps when
modifying the `activerehashing` switch, as in #12705.
2023-12-15 10:42:53 +08:00
|
|
|
#define INCREMENTAL_REHASHING_THRESHOLD_US 1000
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
#define LOADING_PROCESS_EVENTS_INTERVAL_DEFAULT 100 /* Default: 0.1 seconds */
|
2012-03-08 10:08:44 +01:00
|
|
|
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
/* Bucket sizes for client eviction pools. Each bucket stores clients with
|
|
|
|
* memory usage of up to twice the size of the bucket below it. */
|
|
|
|
#define CLIENT_MEM_USAGE_BUCKET_MIN_LOG 15 /* Bucket sizes start at up to 32KB (2^15) */
|
|
|
|
#define CLIENT_MEM_USAGE_BUCKET_MAX_LOG 33 /* Bucket for largest clients: sizes above 4GB (2^32) */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CLIENT_MEM_USAGE_BUCKETS (1 + CLIENT_MEM_USAGE_BUCKET_MAX_LOG - CLIENT_MEM_USAGE_BUCKET_MIN_LOG)
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
|
2013-08-06 12:55:49 +02:00
|
|
|
#define ACTIVE_EXPIRE_CYCLE_SLOW 0
|
|
|
|
#define ACTIVE_EXPIRE_CYCLE_FAST 1
|
|
|
|
|
2019-09-27 11:39:40 +02:00
|
|
|
/* Children process will exit with this status code to signal that the
|
|
|
|
* process terminated without an error: this is useful in order to kill
|
|
|
|
* a saving child (RDB or AOF one), without triggering in the parent the
|
|
|
|
* write protection that is normally turned on on write errors.
|
|
|
|
* Usually children that are terminated with SIGUSR1 will exit with this
|
|
|
|
* special code. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define SERVER_CHILD_NOERROR_RETVAL 255
|
2019-09-27 11:39:40 +02:00
|
|
|
|
2021-03-22 13:25:58 +02:00
|
|
|
/* Reading copy-on-write info is sometimes expensive and may slow down child
|
|
|
|
* processes that report it continuously. We measure the cost of obtaining it
|
|
|
|
* and hold back additional reading based on this factor. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CHILD_COW_DUTY_CYCLE 100
|
2021-03-22 13:25:58 +02:00
|
|
|
|
2024-07-24 18:23:06 +03:00
|
|
|
/* When child process is performing write to connset it iterates on the set
|
|
|
|
* writing a chunk of the available data to send on each connection.
|
|
|
|
* This constant defines the maximal size of the chunk to use. */
|
|
|
|
#define RIO_CONNSET_WRITE_MAX_CHUNK_SIZE 16384
|
|
|
|
|
2014-12-03 12:06:54 +01:00
|
|
|
/* Instantaneous metrics tracking. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define STATS_METRIC_SAMPLES 16 /* Number of samples per metric. */
|
|
|
|
#define STATS_METRIC_COMMAND 0 /* Number of commands executed. */
|
|
|
|
#define STATS_METRIC_NET_INPUT 1 /* Bytes read to network. */
|
|
|
|
#define STATS_METRIC_NET_OUTPUT 2 /* Bytes written to network. */
|
|
|
|
#define STATS_METRIC_NET_INPUT_REPLICATION 3 /* Bytes read to network during replication. */
|
|
|
|
#define STATS_METRIC_NET_OUTPUT_REPLICATION 4 /* Bytes written to network during replication. */
|
|
|
|
#define STATS_METRIC_EL_CYCLE 5 /* Number of eventloop cycled. */
|
|
|
|
#define STATS_METRIC_EL_DURATION 6 /* Eventloop duration. */
|
2023-05-13 01:13:15 +08:00
|
|
|
#define STATS_METRIC_COUNT 7
|
2014-12-03 12:06:54 +01:00
|
|
|
|
2011-12-31 15:37:33 +01:00
|
|
|
/* Protocol and I/O related defines */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define PROTO_IOBUF_LEN (1024 * 16) /* Generic I/O buffer size */
|
|
|
|
#define PROTO_REPLY_CHUNK_BYTES (16 * 1024) /* 16k output buffer */
|
|
|
|
#define PROTO_INLINE_MAX_SIZE (1024 * 64) /* Max size of inline reads */
|
|
|
|
#define PROTO_MBULK_BIG_ARG (1024 * 32)
|
|
|
|
#define PROTO_RESIZE_THRESHOLD (1024 * 32) /* Threshold for determining whether to resize query buffer */
|
|
|
|
#define PROTO_REPLY_MIN_BYTES (1024) /* the lower limit on reply buffer size */
|
|
|
|
#define REDIS_AUTOSYNC_BYTES (1024 * 1024 * 4) /* Sync file every 4MB. */
|
2015-07-27 09:41:48 +02:00
|
|
|
|
2022-02-22 11:19:38 +02:00
|
|
|
#define REPLY_BUFFER_DEFAULT_PEAK_RESET_TIME 5000 /* 5 seconds */
|
|
|
|
|
2015-07-27 09:41:48 +02:00
|
|
|
/* When configuring the server eventloop, we setup it so that the total number
|
|
|
|
* of file descriptors we can handle are server.maxclients + RESERVED_FDS +
|
|
|
|
* a few more to stay safe. Since RESERVED_FDS defaults to 32, we add 96
|
|
|
|
* in order to make sure of not over provisioning more than 128 fds. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CONFIG_FDSET_INCR (CONFIG_MIN_RESERVED_FDS + 96)
|
2011-10-31 11:13:28 +01:00
|
|
|
|
2020-08-12 17:58:56 +03:00
|
|
|
/* OOM Score Adjustment classes. */
|
2024-06-07 14:21:33 -07:00
|
|
|
#define CONFIG_OOM_PRIMARY 0
|
2020-08-12 17:58:56 +03:00
|
|
|
#define CONFIG_OOM_REPLICA 1
|
|
|
|
#define CONFIG_OOM_BGCHILD 2
|
|
|
|
#define CONFIG_OOM_COUNT 3
|
|
|
|
|
|
|
|
extern int configOOMScoreAdjValuesDefaults[CONFIG_OOM_COUNT];
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Hash table parameters */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define HASHTABLE_MAX_LOAD_FACTOR 1.618 /* Maximum hash table load factor. */
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2024-04-03 23:54:33 +07:00
|
|
|
/* Command flags. Please check the definition of struct serverCommand in this file
|
2011-09-26 15:40:39 +02:00
|
|
|
* for more information about the meaning of every flag. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CMD_WRITE (1ULL << 0)
|
|
|
|
#define CMD_READONLY (1ULL << 1)
|
|
|
|
#define CMD_DENYOOM (1ULL << 2)
|
|
|
|
#define CMD_MODULE (1ULL << 3) /* Command exported by module. */
|
|
|
|
#define CMD_ADMIN (1ULL << 4)
|
|
|
|
#define CMD_PUBSUB (1ULL << 5)
|
|
|
|
#define CMD_NOSCRIPT (1ULL << 6)
|
|
|
|
#define CMD_BLOCKING (1ULL << 8) /* Has potential to block. */
|
|
|
|
#define CMD_LOADING (1ULL << 9)
|
|
|
|
#define CMD_STALE (1ULL << 10)
|
|
|
|
#define CMD_SKIP_MONITOR (1ULL << 11)
|
|
|
|
#define CMD_SKIP_SLOWLOG (1ULL << 12)
|
|
|
|
#define CMD_ASKING (1ULL << 13)
|
|
|
|
#define CMD_FAST (1ULL << 14)
|
|
|
|
#define CMD_NO_AUTH (1ULL << 15)
|
|
|
|
#define CMD_MAY_REPLICATE (1ULL << 16)
|
|
|
|
#define CMD_SENTINEL (1ULL << 17)
|
|
|
|
#define CMD_ONLY_SENTINEL (1ULL << 18)
|
|
|
|
#define CMD_NO_MANDATORY_KEYS (1ULL << 19)
|
|
|
|
#define CMD_PROTECTED (1ULL << 20)
|
|
|
|
#define CMD_MODULE_GETKEYS (1ULL << 21) /* Use the modules getkeys interface. */
|
|
|
|
#define CMD_MODULE_NO_CLUSTER (1ULL << 22) /* Deny on Cluster. */
|
|
|
|
#define CMD_NO_ASYNC_LOADING (1ULL << 23)
|
|
|
|
#define CMD_NO_MULTI (1ULL << 24)
|
|
|
|
#define CMD_MOVABLE_KEYS \
|
|
|
|
(1ULL << 25) /* The legacy range spec doesn't cover all keys. \
|
|
|
|
* Populated by populateCommandLegacyRangeSpec. */
|
|
|
|
#define CMD_ALLOW_BUSY ((1ULL << 26))
|
|
|
|
#define CMD_MODULE_GETCHANNELS (1ULL << 27) /* Use the modules getchannels interface. */
|
|
|
|
#define CMD_TOUCHES_ARBITRARY_KEYS (1ULL << 28)
|
2019-01-23 16:47:29 +01:00
|
|
|
|
|
|
|
/* Command flags that describe ACLs categories. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define ACL_CATEGORY_KEYSPACE (1ULL << 0)
|
|
|
|
#define ACL_CATEGORY_READ (1ULL << 1)
|
|
|
|
#define ACL_CATEGORY_WRITE (1ULL << 2)
|
|
|
|
#define ACL_CATEGORY_SET (1ULL << 3)
|
|
|
|
#define ACL_CATEGORY_SORTEDSET (1ULL << 4)
|
|
|
|
#define ACL_CATEGORY_LIST (1ULL << 5)
|
|
|
|
#define ACL_CATEGORY_HASH (1ULL << 6)
|
|
|
|
#define ACL_CATEGORY_STRING (1ULL << 7)
|
|
|
|
#define ACL_CATEGORY_BITMAP (1ULL << 8)
|
|
|
|
#define ACL_CATEGORY_HYPERLOGLOG (1ULL << 9)
|
|
|
|
#define ACL_CATEGORY_GEO (1ULL << 10)
|
|
|
|
#define ACL_CATEGORY_STREAM (1ULL << 11)
|
|
|
|
#define ACL_CATEGORY_PUBSUB (1ULL << 12)
|
|
|
|
#define ACL_CATEGORY_ADMIN (1ULL << 13)
|
|
|
|
#define ACL_CATEGORY_FAST (1ULL << 14)
|
|
|
|
#define ACL_CATEGORY_SLOW (1ULL << 15)
|
|
|
|
#define ACL_CATEGORY_BLOCKING (1ULL << 16)
|
|
|
|
#define ACL_CATEGORY_DANGEROUS (1ULL << 17)
|
|
|
|
#define ACL_CATEGORY_CONNECTION (1ULL << 18)
|
|
|
|
#define ACL_CATEGORY_TRANSACTION (1ULL << 19)
|
|
|
|
#define ACL_CATEGORY_SCRIPTING (1ULL << 20)
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
|
New detailed key-spec flags (RO, RW, OW, RM, ACCESS, UPDATE, INSERT, DELETE) (#10122)
The new ACL key based permissions in #9974 require the key-specs (#8324) to have more
explicit flags rather than just READ and WRITE. See discussion in #10040
This PR defines two groups of flags:
One about how redis internally handles the key (mutually-exclusive).
The other is about the logical operation done from the user's point of view (3 mutually exclusive
write flags, and one read flag, all optional).
In both groups, if we can't explicitly flag something as explicit read-only, delete-only, or
insert-only, we flag it as `RW` or `UPDATE`.
here's the definition from the code:
```
/* Key-spec flags *
* -------------- */
/* The following refer what the command actually does with the value or metadata
* of the key, and not necessarily the user data or how it affects it.
* Each key-spec may must have exaclty one of these. Any operation that's not
* distinctly deletion, overwrite or read-only would be marked as RW. */
#define CMD_KEY_RO (1ULL<<0) /* Read-Only - Reads the value of the key, but
* doesn't necessarily returns it. */
#define CMD_KEY_RW (1ULL<<1) /* Read-Write - Modifies the data stored in the
* value of the key or its metadata. */
#define CMD_KEY_OW (1ULL<<2) /* Overwrite - Overwrites the data stored in
* the value of the key. */
#define CMD_KEY_RM (1ULL<<3) /* Deletes the key. */
/* The follwing refer to user data inside the value of the key, not the metadata
* like LRU, type, cardinality. It refers to the logical operation on the user's
* data (actual input strings / TTL), being used / returned / copied / changed,
* It doesn't refer to modification or returning of metadata (like type, count,
* presence of data). Any write that's not INSERT or DELETE, would be an UPADTE.
* Each key-spec may have one of the writes with or without access, or none: */
#define CMD_KEY_ACCESS (1ULL<<4) /* Returns, copies or uses the user data from
* the value of the key. */
#define CMD_KEY_UPDATE (1ULL<<5) /* Updates data to the value, new value may
* depend on the old value. */
#define CMD_KEY_INSERT (1ULL<<6) /* Adds data to the value with no chance of,
* modification or deletion of existing data. */
#define CMD_KEY_DELETE (1ULL<<7) /* Explicitly deletes some content
* from the value of the key. */
```
Unrelated changes:
- generate-command-code.py is only compatible with python3 (modified the shabang)
- generate-command-code.py print file on json parsing error
- rename `shard_channel` key-spec flag to just `channel`.
- add INCOMPLETE flag in input spec of SORT and SORT_RO
2022-01-18 16:00:00 +02:00
|
|
|
/* Key-spec flags *
|
|
|
|
* -------------- */
|
|
|
|
/* The following refer what the command actually does with the value or metadata
|
|
|
|
* of the key, and not necessarily the user data or how it affects it.
|
|
|
|
* Each key-spec may must have exactly one of these. Any operation that's not
|
|
|
|
* distinctly deletion, overwrite or read-only would be marked as RW. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CMD_KEY_RO \
|
|
|
|
(1ULL << 0) /* Read-Only - Reads the value of the key, but \
|
|
|
|
* doesn't necessarily returns it. */
|
|
|
|
#define CMD_KEY_RW \
|
|
|
|
(1ULL << 1) /* Read-Write - Modifies the data stored in the \
|
|
|
|
* value of the key or its metadata. */
|
|
|
|
#define CMD_KEY_OW \
|
|
|
|
(1ULL << 2) /* Overwrite - Overwrites the data stored in \
|
|
|
|
* the value of the key. */
|
|
|
|
#define CMD_KEY_RM (1ULL << 3) /* Deletes the key. */
|
New detailed key-spec flags (RO, RW, OW, RM, ACCESS, UPDATE, INSERT, DELETE) (#10122)
The new ACL key based permissions in #9974 require the key-specs (#8324) to have more
explicit flags rather than just READ and WRITE. See discussion in #10040
This PR defines two groups of flags:
One about how redis internally handles the key (mutually-exclusive).
The other is about the logical operation done from the user's point of view (3 mutually exclusive
write flags, and one read flag, all optional).
In both groups, if we can't explicitly flag something as explicit read-only, delete-only, or
insert-only, we flag it as `RW` or `UPDATE`.
here's the definition from the code:
```
/* Key-spec flags *
* -------------- */
/* The following refer what the command actually does with the value or metadata
* of the key, and not necessarily the user data or how it affects it.
* Each key-spec may must have exaclty one of these. Any operation that's not
* distinctly deletion, overwrite or read-only would be marked as RW. */
#define CMD_KEY_RO (1ULL<<0) /* Read-Only - Reads the value of the key, but
* doesn't necessarily returns it. */
#define CMD_KEY_RW (1ULL<<1) /* Read-Write - Modifies the data stored in the
* value of the key or its metadata. */
#define CMD_KEY_OW (1ULL<<2) /* Overwrite - Overwrites the data stored in
* the value of the key. */
#define CMD_KEY_RM (1ULL<<3) /* Deletes the key. */
/* The follwing refer to user data inside the value of the key, not the metadata
* like LRU, type, cardinality. It refers to the logical operation on the user's
* data (actual input strings / TTL), being used / returned / copied / changed,
* It doesn't refer to modification or returning of metadata (like type, count,
* presence of data). Any write that's not INSERT or DELETE, would be an UPADTE.
* Each key-spec may have one of the writes with or without access, or none: */
#define CMD_KEY_ACCESS (1ULL<<4) /* Returns, copies or uses the user data from
* the value of the key. */
#define CMD_KEY_UPDATE (1ULL<<5) /* Updates data to the value, new value may
* depend on the old value. */
#define CMD_KEY_INSERT (1ULL<<6) /* Adds data to the value with no chance of,
* modification or deletion of existing data. */
#define CMD_KEY_DELETE (1ULL<<7) /* Explicitly deletes some content
* from the value of the key. */
```
Unrelated changes:
- generate-command-code.py is only compatible with python3 (modified the shabang)
- generate-command-code.py print file on json parsing error
- rename `shard_channel` key-spec flag to just `channel`.
- add INCOMPLETE flag in input spec of SORT and SORT_RO
2022-01-18 16:00:00 +02:00
|
|
|
/* The following refer to user data inside the value of the key, not the metadata
|
|
|
|
* like LRU, type, cardinality. It refers to the logical operation on the user's
|
|
|
|
* data (actual input strings / TTL), being used / returned / copied / changed,
|
|
|
|
* It doesn't refer to modification or returning of metadata (like type, count,
|
|
|
|
* presence of data). Any write that's not INSERT or DELETE, would be an UPDATE.
|
|
|
|
* Each key-spec may have one of the writes with or without access, or none: */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CMD_KEY_ACCESS \
|
|
|
|
(1ULL << 4) /* Returns, copies or uses the user data from \
|
|
|
|
* the value of the key. */
|
|
|
|
#define CMD_KEY_UPDATE \
|
|
|
|
(1ULL << 5) /* Updates data to the value, new value may \
|
|
|
|
* depend on the old value. */
|
|
|
|
#define CMD_KEY_INSERT \
|
|
|
|
(1ULL << 6) /* Adds data to the value with no chance of \
|
|
|
|
* modification or deletion of existing data. */
|
|
|
|
#define CMD_KEY_DELETE \
|
|
|
|
(1ULL << 7) /* Explicitly deletes some content \
|
|
|
|
* from the value of the key. */
|
New detailed key-spec flags (RO, RW, OW, RM, ACCESS, UPDATE, INSERT, DELETE) (#10122)
The new ACL key based permissions in #9974 require the key-specs (#8324) to have more
explicit flags rather than just READ and WRITE. See discussion in #10040
This PR defines two groups of flags:
One about how redis internally handles the key (mutually-exclusive).
The other is about the logical operation done from the user's point of view (3 mutually exclusive
write flags, and one read flag, all optional).
In both groups, if we can't explicitly flag something as explicit read-only, delete-only, or
insert-only, we flag it as `RW` or `UPDATE`.
here's the definition from the code:
```
/* Key-spec flags *
* -------------- */
/* The following refer what the command actually does with the value or metadata
* of the key, and not necessarily the user data or how it affects it.
* Each key-spec may must have exaclty one of these. Any operation that's not
* distinctly deletion, overwrite or read-only would be marked as RW. */
#define CMD_KEY_RO (1ULL<<0) /* Read-Only - Reads the value of the key, but
* doesn't necessarily returns it. */
#define CMD_KEY_RW (1ULL<<1) /* Read-Write - Modifies the data stored in the
* value of the key or its metadata. */
#define CMD_KEY_OW (1ULL<<2) /* Overwrite - Overwrites the data stored in
* the value of the key. */
#define CMD_KEY_RM (1ULL<<3) /* Deletes the key. */
/* The follwing refer to user data inside the value of the key, not the metadata
* like LRU, type, cardinality. It refers to the logical operation on the user's
* data (actual input strings / TTL), being used / returned / copied / changed,
* It doesn't refer to modification or returning of metadata (like type, count,
* presence of data). Any write that's not INSERT or DELETE, would be an UPADTE.
* Each key-spec may have one of the writes with or without access, or none: */
#define CMD_KEY_ACCESS (1ULL<<4) /* Returns, copies or uses the user data from
* the value of the key. */
#define CMD_KEY_UPDATE (1ULL<<5) /* Updates data to the value, new value may
* depend on the old value. */
#define CMD_KEY_INSERT (1ULL<<6) /* Adds data to the value with no chance of,
* modification or deletion of existing data. */
#define CMD_KEY_DELETE (1ULL<<7) /* Explicitly deletes some content
* from the value of the key. */
```
Unrelated changes:
- generate-command-code.py is only compatible with python3 (modified the shabang)
- generate-command-code.py print file on json parsing error
- rename `shard_channel` key-spec flag to just `channel`.
- add INCOMPLETE flag in input spec of SORT and SORT_RO
2022-01-18 16:00:00 +02:00
|
|
|
/* Other flags: */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CMD_KEY_NOT_KEY \
|
|
|
|
(1ULL << 8) /* A 'fake' key that should be routed \
|
|
|
|
* like a key in cluster mode but is \
|
|
|
|
* excluded from other key checks. */
|
|
|
|
#define CMD_KEY_INCOMPLETE \
|
|
|
|
(1ULL << 9) /* Means that the keyspec might not point \
|
|
|
|
* out to all keys it should cover */
|
|
|
|
#define CMD_KEY_VARIABLE_FLAGS \
|
|
|
|
(1ULL << 10) /* Means that some keys might have \
|
|
|
|
* different flags depending on arguments */
|
Treat subcommands as commands (#9504)
## Intro
The purpose is to allow having different flags/ACL categories for
subcommands (Example: CONFIG GET is ok-loading but CONFIG SET isn't)
We create a small command table for every command that has subcommands
and each subcommand has its own flags, etc. (same as a "regular" command)
This commit also unites the Redis and the Sentinel command tables
## Affected commands
CONFIG
Used to have "admin ok-loading ok-stale no-script"
Changes:
1. Dropped "ok-loading" in all except GET (this doesn't change behavior since
there were checks in the code doing that)
XINFO
Used to have "read-only random"
Changes:
1. Dropped "random" in all except CONSUMERS
XGROUP
Used to have "write use-memory"
Changes:
1. Dropped "use-memory" in all except CREATE and CREATECONSUMER
COMMAND
No changes.
MEMORY
Used to have "random read-only"
Changes:
1. Dropped "random" in PURGE and USAGE
ACL
Used to have "admin no-script ok-loading ok-stale"
Changes:
1. Dropped "admin" in WHOAMI, GENPASS, and CAT
LATENCY
No changes.
MODULE
No changes.
SLOWLOG
Used to have "admin random ok-loading ok-stale"
Changes:
1. Dropped "random" in RESET
OBJECT
Used to have "read-only random"
Changes:
1. Dropped "random" in ENCODING and REFCOUNT
SCRIPT
Used to have "may-replicate no-script"
Changes:
1. Dropped "may-replicate" in all except FLUSH and LOAD
CLIENT
Used to have "admin no-script random ok-loading ok-stale"
Changes:
1. Dropped "random" in all except INFO and LIST
2. Dropped "admin" in ID, TRACKING, CACHING, GETREDIR, INFO, SETNAME, GETNAME, and REPLY
STRALGO
No changes.
PUBSUB
No changes.
CLUSTER
Changes:
1. Dropped "admin in countkeysinslots, getkeysinslot, info, nodes, keyslot, myid, and slots
SENTINEL
No changes.
(note that DEBUG also fits, but we decided not to convert it since it's for
debugging and anyway undocumented)
## New sub-command
This commit adds another element to the per-command output of COMMAND,
describing the list of subcommands, if any (in the same structure as "regular" commands)
Also, it adds a new subcommand:
```
COMMAND LIST [FILTERBY (MODULE <module-name>|ACLCAT <cat>|PATTERN <pattern>)]
```
which returns a set of all commands (unless filters), but excluding subcommands.
## Module API
A new module API, RM_CreateSubcommand, was added, in order to allow
module writer to define subcommands
## ACL changes:
1. Now, that each subcommand is actually a command, each has its own ACL id.
2. The old mechanism of allowed_subcommands is redundant
(blocking/allowing a subcommand is the same as blocking/allowing a regular command),
but we had to keep it, to support the widespread usage of allowed_subcommands
to block commands with certain args, that aren't subcommands (e.g. "-select +select|0").
3. I have renamed allowed_subcommands to allowed_firstargs to emphasize the difference.
4. Because subcommands are commands in ACL too, you can now use "-" to block subcommands
(e.g. "+client -client|kill"), which wasn't possible in the past.
5. It is also possible to use the allowed_firstargs mechanism with subcommand.
For example: `+config -config|set +config|set|loglevel` will block all CONFIG SET except
for setting the log level.
6. All of the ACL changes above required some amount of refactoring.
## Misc
1. There are two approaches: Either each subcommand has its own function or all
subcommands use the same function, determining what to do according to argv[0].
For now, I took the former approaches only with CONFIG and COMMAND,
while other commands use the latter approach (for smaller blamelog diff).
2. Deleted memoryGetKeys: It is no longer needed because MEMORY USAGE now uses the "range" key spec.
4. Bugfix: GETNAME was missing from CLIENT's help message.
5. Sentinel and Redis now use the same table, with the same function pointer.
Some commands have a different implementation in Sentinel, so we redirect
them (these are ROLE, PUBLISH, and INFO).
6. Command stats now show the stats per subcommand (e.g. instead of stats just
for "config" you will have stats for "config|set", "config|get", etc.)
7. It is now possible to use COMMAND directly on subcommands:
COMMAND INFO CONFIG|GET (The pipeline syntax was inspired from ACL, and
can be used in functions lookupCommandBySds and lookupCommandByCString)
8. STRALGO is now a container command (has "help")
## Breaking changes:
1. Command stats now show the stats per subcommand (see (5) above)
2021-10-20 10:52:57 +02:00
|
|
|
|
2022-02-08 10:01:35 +02:00
|
|
|
/* Key flags for when access type is unknown */
|
|
|
|
#define CMD_KEY_FULL_ACCESS (CMD_KEY_RW | CMD_KEY_ACCESS | CMD_KEY_UPDATE)
|
|
|
|
|
2022-11-30 17:56:36 +08:00
|
|
|
/* Key flags for how key is removed */
|
|
|
|
#define DB_FLAG_KEY_NONE 0
|
2024-05-22 23:24:12 -07:00
|
|
|
#define DB_FLAG_KEY_DELETED (1ULL << 0)
|
|
|
|
#define DB_FLAG_KEY_EXPIRED (1ULL << 1)
|
|
|
|
#define DB_FLAG_KEY_EVICTED (1ULL << 2)
|
|
|
|
#define DB_FLAG_KEY_OVERWRITE (1ULL << 3)
|
2022-11-30 17:56:36 +08:00
|
|
|
|
2022-02-22 01:00:03 -08:00
|
|
|
/* Channel flags share the same flag space as the key flags */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CMD_CHANNEL_PATTERN (1ULL << 11) /* The argument is a channel pattern */
|
|
|
|
#define CMD_CHANNEL_SUBSCRIBE (1ULL << 12) /* The command subscribes to channels */
|
|
|
|
#define CMD_CHANNEL_UNSUBSCRIBE (1ULL << 13) /* The command unsubscribes to channels */
|
|
|
|
#define CMD_CHANNEL_PUBLISH (1ULL << 14) /* The command publishes to channels. */
|
2022-02-22 01:00:03 -08:00
|
|
|
|
2011-12-21 10:05:32 +01:00
|
|
|
/* AOF states */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define AOF_OFF 0 /* AOF is off */
|
|
|
|
#define AOF_ON 1 /* AOF is on */
|
|
|
|
#define AOF_WAIT_REWRITE 2 /* AOF waits rewrite to start appending */
|
2011-12-21 10:05:32 +01:00
|
|
|
|
2022-05-29 18:33:16 +08:00
|
|
|
/* AOF return values for loadAppendOnlyFiles() and loadSingleAppendOnlyFile() */
|
2021-06-14 10:38:08 +03:00
|
|
|
#define AOF_OK 0
|
|
|
|
#define AOF_NOT_EXIST 1
|
|
|
|
#define AOF_EMPTY 2
|
|
|
|
#define AOF_OPEN_ERR 3
|
|
|
|
#define AOF_FAILED 4
|
Implement Multi Part AOF mechanism to avoid AOFRW overheads. (#9788)
Implement Multi-Part AOF mechanism to avoid overheads during AOFRW.
Introducing a folder with multiple AOF files tracked by a manifest file.
The main issues with the the original AOFRW mechanism are:
* buffering of commands that are processed during rewrite (consuming a lot of RAM)
* freezes of the main process when the AOFRW completes to drain the remaining part of the buffer and fsync it.
* double disk IO for the data that arrives during AOFRW (had to be written to both the old and new AOF files)
The main modifications of this PR:
1. Remove the AOF rewrite buffer and related code.
2. Divide the AOF into multiple files, they are classified as two types, one is the the `BASE` type,
it represents the full amount of data (Maybe AOF or RDB format) after each AOFRW, there is only
one `BASE` file at most. The second is `INCR` type, may have more than one. They represent the
incremental commands since the last AOFRW.
3. Use a AOF manifest file to record and manage these AOF files mentioned above.
4. The original configuration of `appendfilename` will be the base part of the new file name, for example:
`appendonly.aof.1.base.rdb` and `appendonly.aof.2.incr.aof`
5. Add manifest-related TCL tests, and modified some existing tests that depend on the `appendfilename`
6. Remove the `aof_rewrite_buffer_length` field in info.
7. Add `aof-disable-auto-gc` configuration. By default we're automatically deleting HISTORY type AOFs.
It also gives users the opportunity to preserve the history AOFs. just for testing use now.
8. Add AOFRW limiting measure. When the AOFRW failures reaches the threshold (3 times now),
we will delay the execution of the next AOFRW by 1 minute. If the next AOFRW also fails, it will be
delayed by 2 minutes. The next is 4, 8, 16, the maximum delay is 60 minutes (1 hour). During the limit
period, we can still use the 'bgrewriteaof' command to execute AOFRW immediately.
9. Support upgrade (load) data from old version redis.
10. Add `appenddirname` configuration, as the directory name of the append only files. All AOF files and
manifest file will be placed in this directory.
11. Only the last AOF file (BASE or INCR) can be truncated. Otherwise redis will exit even if
`aof-load-truncated` is enabled.
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-04 01:14:13 +08:00
|
|
|
#define AOF_TRUNCATED 5
|
2021-06-14 10:38:08 +03:00
|
|
|
|
2022-07-26 20:13:13 +08:00
|
|
|
/* RDB return values for rdbLoad. */
|
|
|
|
#define RDB_OK 0
|
|
|
|
#define RDB_NOT_EXIST 1 /* RDB file doesn't exist. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define RDB_FAILED 2 /* Failed to load the RDB file. */
|
2022-07-26 20:13:13 +08:00
|
|
|
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
/* Command doc flags */
|
|
|
|
#define CMD_DOC_NONE 0
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CMD_DOC_DEPRECATED (1 << 0) /* Command is deprecated */
|
|
|
|
#define CMD_DOC_SYSCMD (1 << 1) /* System (internal) command */
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
|
2024-06-27 19:00:45 +08:00
|
|
|
/* Client capabilities */
|
|
|
|
#define CLIENT_CAPA_REDIRECT (1 << 0) /* Indicate that the client can handle redirection */
|
|
|
|
|
2013-12-03 17:43:53 +01:00
|
|
|
/* Client block type (btype field in client structure)
|
2015-07-27 09:41:48 +02:00
|
|
|
* if CLIENT_BLOCKED flag is set. */
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
typedef enum blocking_type {
|
2024-05-30 23:45:47 -07:00
|
|
|
BLOCKED_NONE, /* Not blocked, no CLIENT_BLOCKED flag set. */
|
|
|
|
BLOCKED_LIST, /* BLPOP & co. */
|
|
|
|
BLOCKED_WAIT, /* WAIT for synchronous replication. */
|
|
|
|
BLOCKED_MODULE, /* Blocked by a loadable module. */
|
|
|
|
BLOCKED_STREAM, /* XREAD. */
|
|
|
|
BLOCKED_ZSET, /* BZPOP et al. */
|
|
|
|
BLOCKED_POSTPONE, /* Blocked by processCommand, re-try processing later. */
|
|
|
|
BLOCKED_SHUTDOWN, /* SHUTDOWN. */
|
|
|
|
BLOCKED_NUM, /* Number of blocked states. */
|
|
|
|
BLOCKED_END /* End of enumeration */
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
} blocking_type;
|
2013-12-03 17:43:53 +01:00
|
|
|
|
2010-10-15 15:40:25 +02:00
|
|
|
/* Client request types */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define PROTO_REQ_INLINE 1
|
|
|
|
#define PROTO_REQ_MULTIBULK 2
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2012-01-17 12:43:01 +01:00
|
|
|
/* Client classes for client limits, currently used only for
|
|
|
|
* the max-client-output-buffer limit implementation. */
|
2024-06-07 14:21:33 -07:00
|
|
|
#define CLIENT_TYPE_NORMAL 0 /* Normal req-reply clients + MONITORs */
|
|
|
|
#define CLIENT_TYPE_REPLICA 1 /* Replicas. */
|
|
|
|
#define CLIENT_TYPE_PUBSUB 2 /* Clients subscribed to PubSub channels. */
|
|
|
|
#define CLIENT_TYPE_PRIMARY 3 /* Primary. */
|
|
|
|
#define CLIENT_TYPE_COUNT 4 /* Total number of client types. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CLIENT_TYPE_OBUF_COUNT \
|
|
|
|
3 /* Number of clients to expose to output \
|
|
|
|
buffer configuration. Just the first \
|
2024-06-07 14:21:33 -07:00
|
|
|
three: normal, replica, pubsub. */
|
2015-07-27 09:41:48 +02:00
|
|
|
|
2024-06-07 14:21:33 -07:00
|
|
|
/* Replica replication state. Used in server.repl_state for replicas to remember
|
2015-07-27 09:41:48 +02:00
|
|
|
* what to do next. */
|
2020-12-18 22:10:31 +02:00
|
|
|
typedef enum {
|
2024-05-22 23:24:12 -07:00
|
|
|
REPL_STATE_NONE = 0, /* No active replication */
|
2024-06-07 14:21:33 -07:00
|
|
|
REPL_STATE_CONNECT, /* Must connect to primary */
|
|
|
|
REPL_STATE_CONNECTING, /* Connecting to primary */
|
2020-12-18 22:10:31 +02:00
|
|
|
/* --- Handshake states, must be ordered --- */
|
2024-05-27 23:03:34 +02:00
|
|
|
REPL_STATE_RECEIVE_PING_REPLY, /* Wait for PING reply */
|
2024-06-07 14:21:33 -07:00
|
|
|
REPL_STATE_SEND_HANDSHAKE, /* Send handshake sequence to primary */
|
2024-05-27 23:03:34 +02:00
|
|
|
REPL_STATE_RECEIVE_AUTH_REPLY, /* Wait for AUTH reply */
|
|
|
|
REPL_STATE_RECEIVE_PORT_REPLY, /* Wait for REPLCONF reply */
|
|
|
|
REPL_STATE_RECEIVE_IP_REPLY, /* Wait for REPLCONF reply */
|
|
|
|
REPL_STATE_RECEIVE_CAPA_REPLY, /* Wait for REPLCONF reply */
|
|
|
|
REPL_STATE_RECEIVE_VERSION_REPLY, /* Wait for REPLCONF reply */
|
|
|
|
REPL_STATE_SEND_PSYNC, /* Send PSYNC */
|
|
|
|
REPL_STATE_RECEIVE_PSYNC_REPLY, /* Wait for PSYNC reply */
|
2020-12-18 22:10:31 +02:00
|
|
|
/* --- End of handshake states --- */
|
2024-06-07 14:21:33 -07:00
|
|
|
REPL_STATE_TRANSFER, /* Receiving .rdb from primary */
|
|
|
|
REPL_STATE_CONNECTED, /* Connected to primary */
|
2020-12-18 22:10:31 +02:00
|
|
|
} repl_state;
|
2015-07-27 09:41:48 +02:00
|
|
|
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
/* Replica rdb-channel replication state. Used in server.repl_rdb_channel_state for
|
|
|
|
* replicas to remember what to do next. */
|
|
|
|
typedef enum {
|
|
|
|
REPL_DUAL_CHANNEL_STATE_NONE = 0, /* No active rdb channel sync */
|
|
|
|
REPL_DUAL_CHANNEL_SEND_HANDSHAKE, /* Send handshake sequence to primary */
|
|
|
|
REPL_DUAL_CHANNEL_RECEIVE_AUTH_REPLY, /* Wait for AUTH reply */
|
|
|
|
REPL_DUAL_CHANNEL_RECEIVE_REPLCONF_REPLY, /* Wait for REPLCONF reply */
|
|
|
|
REPL_DUAL_CHANNEL_RECEIVE_ENDOFF, /* Wait for $ENDOFF reply */
|
|
|
|
REPL_DUAL_CHANNEL_RDB_LOAD, /* Loading rdb using rdb channel */
|
|
|
|
REPL_DUAL_CHANNEL_RDB_LOADED,
|
|
|
|
} repl_rdb_channel_state;
|
|
|
|
|
2021-01-28 13:18:05 -08:00
|
|
|
/* The state of an in progress coordinated failover */
|
|
|
|
typedef enum {
|
|
|
|
NO_FAILOVER = 0, /* No failover in progress */
|
|
|
|
FAILOVER_WAIT_FOR_SYNC, /* Waiting for target replica to catch up */
|
|
|
|
FAILOVER_IN_PROGRESS /* Waiting for target replica to accept
|
|
|
|
* PSYNC FAILOVER request. */
|
|
|
|
} failover_state;
|
|
|
|
|
2024-06-07 14:21:33 -07:00
|
|
|
/* State of replicas from the POV of the primary. Used in client->replstate.
|
|
|
|
* In SEND_BULK and ONLINE state the replica receives new updates
|
2015-07-27 09:41:48 +02:00
|
|
|
* in its output queue. In the WAIT_BGSAVE states instead the server is waiting
|
2010-06-22 00:07:48 +02:00
|
|
|
* to start the next background saving in order to send updates to it. */
|
2024-06-07 14:21:33 -07:00
|
|
|
#define REPLICA_STATE_WAIT_BGSAVE_START 6 /* We need to produce a new RDB file. */
|
|
|
|
#define REPLICA_STATE_WAIT_BGSAVE_END 7 /* Waiting RDB file creation to finish. */
|
|
|
|
#define REPLICA_STATE_SEND_BULK 8 /* Sending RDB file to replica. */
|
|
|
|
#define REPLICA_STATE_ONLINE 9 /* RDB file transmitted, sending just updates. */
|
|
|
|
#define REPLICA_STATE_RDB_TRANSMITTED \
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
10 /* RDB file transmitted - This state is used only for \
|
|
|
|
* a replica that only wants RDB without replication buffer */
|
|
|
|
#define REPLICA_STATE_BG_RDB_LOAD 11 /* Main channel of a replica which uses dual channel replication. */
|
2012-11-03 11:56:28 +01:00
|
|
|
|
2024-06-07 14:21:33 -07:00
|
|
|
/* Replica capabilities. */
|
|
|
|
#define REPLICA_CAPA_NONE 0
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
#define REPLICA_CAPA_EOF (1 << 0) /* Can parse the RDB EOF streaming format. */
|
|
|
|
#define REPLICA_CAPA_PSYNC2 (1 << 1) /* Supports PSYNC2 protocol. */
|
|
|
|
#define REPLICA_CAPA_DUAL_CHANNEL (1 << 2) /* Supports dual channel replication sync */
|
2015-08-06 09:23:23 +02:00
|
|
|
|
2024-06-07 14:21:33 -07:00
|
|
|
/* Replica requirements */
|
|
|
|
#define REPLICA_REQ_NONE 0
|
|
|
|
#define REPLICA_REQ_RDB_EXCLUDE_DATA (1 << 0) /* Exclude data from RDB */
|
|
|
|
#define REPLICA_REQ_RDB_EXCLUDE_FUNCTIONS (1 << 1) /* Exclude functions from RDB */
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
#define REPLICA_REQ_RDB_CHANNEL (1 << 2) /* Use dual-channel-replication */
|
2024-06-07 14:21:33 -07:00
|
|
|
/* Mask of all bits in the replica requirements bitfield that represent non-standard (filtered) RDB requirements */
|
|
|
|
#define REPLICA_REQ_RDB_MASK (REPLICA_REQ_RDB_EXCLUDE_DATA | REPLICA_REQ_RDB_EXCLUDE_FUNCTIONS)
|
2022-01-02 09:39:01 +02:00
|
|
|
|
2024-06-07 14:21:33 -07:00
|
|
|
/* Synchronous read timeout - replica side */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define CONFIG_REPL_SYNCIO_TIMEOUT 5
|
2010-06-22 00:07:48 +02:00
|
|
|
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
/* The default number of replication backlog blocks to trim per call. */
|
|
|
|
#define REPL_BACKLOG_TRIM_BLOCKS_PER_CALL 64
|
|
|
|
|
|
|
|
/* In order to quickly find the requested offset for PSYNC requests,
|
|
|
|
* we index some nodes in the replication buffer linked list into a rax. */
|
|
|
|
#define REPL_BACKLOG_INDEX_PER_BLOCKS 64
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* List related stuff */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define LIST_HEAD 0
|
|
|
|
#define LIST_TAIL 1
|
2018-05-11 17:31:46 +02:00
|
|
|
#define ZSET_MIN 0
|
|
|
|
#define ZSET_MAX 1
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* Sort operations */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define SORT_OP_GET 0
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* Log levels */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define LL_DEBUG 0
|
|
|
|
#define LL_VERBOSE 1
|
|
|
|
#define LL_NOTICE 2
|
|
|
|
#define LL_WARNING 3
|
2023-05-23 23:30:44 +08:00
|
|
|
#define LL_NOTHING 4
|
2024-05-22 23:24:12 -07:00
|
|
|
#define LL_RAW (1 << 10) /* Modifier to log without timestamp */
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2015-01-08 15:22:33 -05:00
|
|
|
/* Supervision options */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define SUPERVISED_NONE 0
|
|
|
|
#define SUPERVISED_AUTODETECT 1
|
|
|
|
#define SUPERVISED_SYSTEMD 2
|
|
|
|
#define SUPERVISED_UPSTART 3
|
2015-01-08 15:22:33 -05:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Anti-warning macro... */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define UNUSED(V) ((void)V)
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2020-02-02 02:48:00 -06:00
|
|
|
#define ZSKIPLIST_MAXLEVEL 32 /* Should be enough for 2^64 elements */
|
2010-06-22 00:07:48 +02:00
|
|
|
#define ZSKIPLIST_P 0.25 /* Skiplist P = 1/4 */
|
Optimize ZRANGE offset location from linear search to skiplist jump. (#12450)
ZRANGE BYSCORE/BYLEX with [LIMIT offset count] option was
using every level in skiplist to jump to the first/last node in range,
but only use level[0] in skiplist to locate the node at offset, resulting
in sub-optimal performance using LIMIT:
```
while (ln && offset--) {
if (reverse) {
ln = ln->backward;
} else {
ln = ln->level[0].forward;
}
}
```
It could be slow when offset is very big. We can get the total rank of
the offset location and use skiplist to jump to it. It is an improvement
from O(offset) to O(log rank).
Below shows how this is implemented (if the offset is positve):
Use the skiplist to seach for the first element in the range, record its
rank `rank_0`, so we can have the rank of the target node `rank_t`.
Meanwhile we record the last node we visited which has zsl->level-1
levels and its rank `rank_1`. Then we start from the zsl->level-1 node,
use skiplist to go forward `rank_t-rank_1` nodes to reach the target node.
It is very similiar when the offset is reversed.
Note that if `rank_t` is very close to `rank_0`, we just start from the first
element in range and go node by node, this for the case when zsl->level-1
node is to far away and it is quicker to reach the target node by node.
Here is a test using a random generated zset including 10000 elements
(with different positive scores), doing a bench mark which compares how
fast the `ZRANGE` command is exucuted before and after the optimization.
The start score is set to 0 and the count is set to 1 to make sure that
most of the time is spent on locating the offset.
```
memtier_benchmark -h 127.0.0.1 -p 6379 --command="zrange test 0 +inf byscore limit <offset> 1"
```
| offset | QPS(unstable) | QPS(optimized) |
|--------|--------|--------|
| 10 | 73386.02 | 74819.82 |
| 1000 | 48084.96 | 73177.73 |
| 2000 | 31156.79 | 72805.83 |
| 5000 | 10954.83 | 71218.21 |
With the result above, we can see that the original code is greatly
slowed down when offset gets bigger, and with the optimization the
speed is almost not affected.
Similiar results are generated when testing reversed offset:
```
memtier_benchmark -h 127.0.0.1 -p 6379 --command="zrange test +inf 0 byscore rev limit <offset> 1"
```
| offset | QPS(unstable) | QPS(optimized) |
|--------|--------|--------|
| 10 | 74505.14 | 71653.67 |
| 1000 | 46829.25 | 72842.75 |
| 2000 | 28985.48 | 73669.01 |
| 5000 | 11066.22 | 73963.45 |
And the same conclusion is drawn from the tests of ZRANGE BYLEX.
2023-08-31 19:42:08 +08:00
|
|
|
#define ZSKIPLIST_MAX_SEARCH 10
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* Append only defines */
|
2011-12-21 11:58:42 +01:00
|
|
|
#define AOF_FSYNC_NO 0
|
|
|
|
#define AOF_FSYNC_ALWAYS 1
|
|
|
|
#define AOF_FSYNC_EVERYSEC 2
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2019-07-01 15:22:29 +03:00
|
|
|
/* Replication diskless load defines */
|
|
|
|
#define REPL_DISKLESS_LOAD_DISABLED 0
|
|
|
|
#define REPL_DISKLESS_LOAD_WHEN_DB_EMPTY 1
|
|
|
|
#define REPL_DISKLESS_LOAD_SWAPDB 2
|
2014-04-15 17:46:51 +02:00
|
|
|
|
2020-07-28 15:45:21 +08:00
|
|
|
/* TLS Client Authentication */
|
|
|
|
#define TLS_CLIENT_AUTH_NO 0
|
|
|
|
#define TLS_CLIENT_AUTH_YES 1
|
|
|
|
#define TLS_CLIENT_AUTH_OPTIONAL 2
|
|
|
|
|
Sanitize dump payload: ziplist, listpack, zipmap, intset, stream
When loading an encoded payload we will at least do a shallow validation to
check that the size that's encoded in the payload matches the size of the
allocation.
This let's us later use this encoded size to make sure the various offsets
inside encoded payload don't reach outside the allocation, if they do, we'll
assert/panic, but at least we won't segfault or smear memory.
We can also do 'deep' validation which runs on all the records of the encoded
payload and validates that they don't contain invalid offsets. This lets us
detect corruptions early and reject a RESTORE command rather than accepting
it and asserting (crashing) later when accessing that payload via some command.
configuration:
- adding ACL flag skip-sanitize-payload
- adding config sanitize-dump-payload [yes/no/clients]
For now, we don't have a good way to ensure MIGRATE in cluster resharding isn't
being slowed down by these sanitation, so i'm setting the default value to `no`,
but later on it should be set to `clients` by default.
changes:
- changing rdbReportError not to `exit` in RESTORE command
- adding a new stat to be able to later check if cluster MIGRATE isn't being
slowed down by sanitation.
2020-08-13 16:41:05 +03:00
|
|
|
/* Sanitize dump payload */
|
|
|
|
#define SANITIZE_DUMP_NO 0
|
|
|
|
#define SANITIZE_DUMP_YES 1
|
|
|
|
#define SANITIZE_DUMP_CLIENTS 2
|
|
|
|
|
Protected configs and sensitive commands (#9920)
Block sensitive configs and commands by default.
* `enable-protected-configs` - block modification of configs with the new `PROTECTED_CONFIG` flag.
Currently we add this flag to `dbfilename`, and `dir` configs,
all of which are non-mutable configs that can set a file redis will write to.
* `enable-debug-command` - block the `DEBUG` command
* `enable-module-command` - block the `MODULE` command
These have a default value set to `no`, so that these features are not
exposed by default to client connections, and can only be set by modifying the config file.
Users can change each of these to either `yes` (allow all access), or `local` (allow access from
local TCP connections and unix domain connections)
Note that this is a **breaking change** (specifically the part about MODULE command being disabled by default).
I.e. we don't consider DEBUG command being blocked as an issue (people shouldn't have been using it),
and the few configs we protected are unlikely to have been set at runtime anyway.
On the other hand, it's likely to assume some users who use modules, load them from the config file anyway.
Note that's the whole point of this PR, for redis to be more secure by default and reduce the attack surface on
innocent users, so secure defaults will necessarily mean a breaking change.
2021-12-19 10:46:16 +02:00
|
|
|
/* Enable protected config/command */
|
|
|
|
#define PROTECTED_ACTION_ALLOWED_NO 0
|
|
|
|
#define PROTECTED_ACTION_ALLOWED_YES 1
|
|
|
|
#define PROTECTED_ACTION_ALLOWED_LOCAL 2
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Sets operations codes */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define SET_OP_UNION 0
|
|
|
|
#define SET_OP_DIFF 1
|
|
|
|
#define SET_OP_INTER 2
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2020-11-22 13:57:56 +02:00
|
|
|
/* oom-score-adj defines */
|
|
|
|
#define OOM_SCORE_ADJ_NO 0
|
|
|
|
#define OOM_SCORE_RELATIVE 1
|
|
|
|
#define OOM_SCORE_ADJ_ABSOLUTE 2
|
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* Server maxmemory strategies. Instead of using just incremental number
|
2016-07-15 12:12:52 +02:00
|
|
|
* for this defines, we use a set of flags so that testing for certain
|
|
|
|
* properties common to multiple policies is faster. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define MAXMEMORY_FLAG_LRU (1 << 0)
|
|
|
|
#define MAXMEMORY_FLAG_LFU (1 << 1)
|
|
|
|
#define MAXMEMORY_FLAG_ALLKEYS (1 << 2)
|
|
|
|
#define MAXMEMORY_FLAG_NO_SHARED_INTEGERS (MAXMEMORY_FLAG_LRU | MAXMEMORY_FLAG_LFU)
|
|
|
|
|
|
|
|
#define MAXMEMORY_VOLATILE_LRU ((0 << 8) | MAXMEMORY_FLAG_LRU)
|
|
|
|
#define MAXMEMORY_VOLATILE_LFU ((1 << 8) | MAXMEMORY_FLAG_LFU)
|
|
|
|
#define MAXMEMORY_VOLATILE_TTL (2 << 8)
|
|
|
|
#define MAXMEMORY_VOLATILE_RANDOM (3 << 8)
|
|
|
|
#define MAXMEMORY_ALLKEYS_LRU ((4 << 8) | MAXMEMORY_FLAG_LRU | MAXMEMORY_FLAG_ALLKEYS)
|
|
|
|
#define MAXMEMORY_ALLKEYS_LFU ((5 << 8) | MAXMEMORY_FLAG_LFU | MAXMEMORY_FLAG_ALLKEYS)
|
|
|
|
#define MAXMEMORY_ALLKEYS_RANDOM ((6 << 8) | MAXMEMORY_FLAG_ALLKEYS)
|
|
|
|
#define MAXMEMORY_NO_EVICTION (7 << 8)
|
2016-07-15 12:12:52 +02:00
|
|
|
|
2011-11-10 17:52:02 +01:00
|
|
|
/* Units */
|
|
|
|
#define UNIT_SECONDS 0
|
|
|
|
#define UNIT_MILLISECONDS 1
|
|
|
|
|
2011-11-18 14:10:48 +01:00
|
|
|
/* SHUTDOWN flags */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define SHUTDOWN_NOFLAGS 0 /* No flags. */
|
|
|
|
#define SHUTDOWN_SAVE \
|
|
|
|
1 /* Force SAVE on SHUTDOWN even if no save \
|
|
|
|
points are configured. */
|
|
|
|
#define SHUTDOWN_NOSAVE 2 /* Don't SAVE on SHUTDOWN. */
|
|
|
|
#define SHUTDOWN_NOW 4 /* Don't wait for replicas to catch up. */
|
|
|
|
#define SHUTDOWN_FORCE 8 /* Don't let errors prevent shutdown. */
|
2011-11-18 14:10:48 +01:00
|
|
|
|
2012-02-02 16:30:52 +01:00
|
|
|
/* Command call flags, see call() function */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define CMD_CALL_NONE 0
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CMD_CALL_PROPAGATE_AOF (1 << 0)
|
|
|
|
#define CMD_CALL_PROPAGATE_REPL (1 << 1)
|
|
|
|
#define CMD_CALL_REPROCESSING (1 << 2)
|
|
|
|
#define CMD_CALL_FROM_MODULE (1 << 3) /* From RM_Call */
|
|
|
|
#define CMD_CALL_PROPAGATE (CMD_CALL_PROPAGATE_AOF | CMD_CALL_PROPAGATE_REPL)
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
#define CMD_CALL_FULL (CMD_CALL_PROPAGATE)
|
2012-02-02 16:30:52 +01:00
|
|
|
|
Sort out mess around propagation and MULTI/EXEC (#9890)
The mess:
Some parts use alsoPropagate for late propagation, others using an immediate one (propagate()),
causing edge cases, ugly/hacky code, and the tendency for bugs
The basic idea is that all commands are propagated via alsoPropagate (i.e. added to a list) and the
top-most call() is responsible for going over that list and actually propagating them (and wrapping
them in MULTI/EXEC if there's more than one command). This is done in the new function,
propagatePendingCommands.
Callers to propagatePendingCommands:
1. top-most call() (we want all nested call()s to add to the also_propagate array and just the top-most
one to propagate them) - via `afterCommand`
2. handleClientsBlockedOnKeys: it is out of call() context and it may propagate stuff - via `afterCommand`.
3. handleClientsBlockedOnKeys edge case: if the looked-up key is already expired, we will propagate the
expire but will not unblock any client so `afterCommand` isn't called. in that case, we have to propagate
the deletion explicitly.
4. cron stuff: active-expire and eviction may also propagate stuff
5. modules: the module API allows to propagate stuff from just about anywhere (timers, keyspace notifications,
threads). I could have tried to catch all the out-of-call-context places but it seemed easier to handle it in one
place: when we free the context. in the spirit of what was done in call(), only the top-most freeing of a module
context may cause propagation.
6. modules: when using a thread-safe ctx it's not clear when/if the ctx will be freed. we do know that the module
must lock the GIL before calling RM_Replicate/RM_Call so we propagate the pending commands when
releasing the GIL.
A "known limitation", which were actually a bug, was fixed because of this commit (see propagate.tcl):
When using a mix of RM_Call with `!` and RM_Replicate, the command would propagate out-of-order:
first all the commands from RM_Call, and then the ones from RM_Replicate
Another thing worth mentioning is that if, in the past, a client would issue a MULTI/EXEC with just one
write command the server would blindly propagate the MULTI/EXEC too, even though it's redundant.
not anymore.
This commit renames propagate() to propagateNow() in order to cause conflicts in pending PRs.
propagatePendingCommands is the only caller of propagateNow, which is now a static, internal helper function.
Optimizations:
1. alsoPropagate will not add stuff to also_propagate if there's no AOF and replicas
2. alsoPropagate reallocs also_propagagte exponentially, to save calls to memmove
Bugfixes:
1. CONFIG SET can create evictions, sending notifications which can cause to dirty++ with modules.
we need to prevent it from propagating to AOF/replicas
2. We need to set current_client in RM_Call. buggy scenario:
- CONFIG SET maxmemory, eviction notifications, module hook calls RM_Call
- assertion in lookupKey crashes, because current_client has CONFIG SET, which isn't CMD_WRITE
3. minor: in eviction, call propagateDeletion after notification, like active-expire and all commands
(we always send a notification before propagating the command)
2021-12-22 23:03:48 +01:00
|
|
|
/* Command propagation flags, see propagateNow() function */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define PROPAGATE_NONE 0
|
|
|
|
#define PROPAGATE_AOF 1
|
|
|
|
#define PROPAGATE_REPL 2
|
2012-02-28 16:17:00 +01:00
|
|
|
|
2022-10-27 11:57:04 +03:00
|
|
|
/* Actions pause types */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define PAUSE_ACTION_CLIENT_WRITE (1 << 0)
|
|
|
|
#define PAUSE_ACTION_CLIENT_ALL (1 << 1) /* must be bigger than PAUSE_ACTION_CLIENT_WRITE */
|
|
|
|
#define PAUSE_ACTION_EXPIRE (1 << 2)
|
|
|
|
#define PAUSE_ACTION_EVICT (1 << 3)
|
|
|
|
#define PAUSE_ACTION_REPLICA (1 << 4) /* pause replica traffic */
|
2022-10-27 11:57:04 +03:00
|
|
|
|
|
|
|
/* common sets of actions to pause/unpause */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define PAUSE_ACTIONS_CLIENT_WRITE_SET \
|
|
|
|
(PAUSE_ACTION_CLIENT_WRITE | PAUSE_ACTION_EXPIRE | PAUSE_ACTION_EVICT | PAUSE_ACTION_REPLICA)
|
|
|
|
#define PAUSE_ACTIONS_CLIENT_ALL_SET \
|
|
|
|
(PAUSE_ACTION_CLIENT_ALL | PAUSE_ACTION_EXPIRE | PAUSE_ACTION_EVICT | PAUSE_ACTION_REPLICA)
|
2021-01-07 23:36:54 -08:00
|
|
|
|
2022-01-02 08:50:15 +01:00
|
|
|
/* Client pause purposes. Each purpose has its own end time and pause type. */
|
|
|
|
typedef enum {
|
|
|
|
PAUSE_BY_CLIENT_COMMAND = 0,
|
|
|
|
PAUSE_DURING_SHUTDOWN,
|
|
|
|
PAUSE_DURING_FAILOVER,
|
|
|
|
NUM_PAUSE_PURPOSES /* This value is the number of purposes above. */
|
|
|
|
} pause_purpose;
|
|
|
|
|
|
|
|
typedef struct {
|
2022-10-27 11:57:04 +03:00
|
|
|
uint32_t paused_actions; /* Bitmask of actions */
|
2022-01-02 08:50:15 +01:00
|
|
|
mstime_t end;
|
|
|
|
} pause_event;
|
|
|
|
|
2022-01-02 19:48:29 -08:00
|
|
|
/* Ways that a clusters endpoint can be described */
|
|
|
|
typedef enum {
|
|
|
|
CLUSTER_ENDPOINT_TYPE_IP = 0, /* Show IP address */
|
|
|
|
CLUSTER_ENDPOINT_TYPE_HOSTNAME, /* Show hostname */
|
|
|
|
CLUSTER_ENDPOINT_TYPE_UNKNOWN_ENDPOINT /* Show NULL or empty */
|
|
|
|
} cluster_endpoint_type;
|
|
|
|
|
2014-10-08 09:09:01 +02:00
|
|
|
/* RDB active child save type. */
|
2015-07-27 09:41:48 +02:00
|
|
|
#define RDB_CHILD_TYPE_NONE 0
|
2024-05-22 23:24:12 -07:00
|
|
|
#define RDB_CHILD_TYPE_DISK 1 /* RDB is written to disk. */
|
2024-06-07 14:21:33 -07:00
|
|
|
#define RDB_CHILD_TYPE_SOCKET 2 /* RDB is written to replica socket. */
|
2014-10-08 09:09:01 +02:00
|
|
|
|
2013-01-25 13:19:08 +01:00
|
|
|
/* Keyspace changes notification classes. Every class is associated with a
|
|
|
|
* character for configuration purposes. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define NOTIFY_KEYSPACE (1 << 0) /* K */
|
|
|
|
#define NOTIFY_KEYEVENT (1 << 1) /* E */
|
|
|
|
#define NOTIFY_GENERIC (1 << 2) /* g */
|
|
|
|
#define NOTIFY_STRING (1 << 3) /* $ */
|
|
|
|
#define NOTIFY_LIST (1 << 4) /* l */
|
|
|
|
#define NOTIFY_SET (1 << 5) /* s */
|
|
|
|
#define NOTIFY_HASH (1 << 6) /* h */
|
|
|
|
#define NOTIFY_ZSET (1 << 7) /* z */
|
|
|
|
#define NOTIFY_EXPIRED (1 << 8) /* x */
|
|
|
|
#define NOTIFY_EVICTED (1 << 9) /* e */
|
|
|
|
#define NOTIFY_STREAM (1 << 10) /* t */
|
|
|
|
#define NOTIFY_KEY_MISS (1 << 11) /* m (Note: This one is excluded from NOTIFY_ALL on purpose) */
|
|
|
|
#define NOTIFY_LOADED (1 << 12) /* module only key space notification, indicate a key loaded from rdb */
|
|
|
|
#define NOTIFY_MODULE (1 << 13) /* d, module key space notification */
|
|
|
|
#define NOTIFY_NEW (1 << 14) /* n, new key notification */
|
|
|
|
#define NOTIFY_ALL \
|
|
|
|
(NOTIFY_GENERIC | NOTIFY_STRING | NOTIFY_LIST | NOTIFY_SET | NOTIFY_HASH | NOTIFY_ZSET | NOTIFY_EXPIRED | \
|
|
|
|
NOTIFY_EVICTED | NOTIFY_STREAM | NOTIFY_MODULE) /* A flag */
|
2013-01-25 13:19:08 +01:00
|
|
|
|
2012-07-23 12:54:52 +02:00
|
|
|
/* Using the following macro you can run code inside serverCron() with the
|
|
|
|
* specified period, specified in milliseconds.
|
2012-12-14 17:10:40 +01:00
|
|
|
* The actual resolution depends on server.hz. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define run_with_period(_ms_) if (((_ms_) <= 1000 / server.hz) || !(server.cronloops % ((_ms_) / (1000 / server.hz))))
|
2012-07-23 12:54:52 +02:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* We can print the stacktrace, so our assert is defined this way: */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define serverAssertWithInfo(_c, _o, _e) \
|
|
|
|
(likely(_e) ? (void)0 : (_serverAssertWithInfo(_c, _o, #_e, __FILE__, __LINE__), valkey_unreachable()))
|
|
|
|
#define serverAssert(_e) (likely(_e) ? (void)0 : (_serverAssert(#_e, __FILE__, __LINE__), valkey_unreachable()))
|
|
|
|
#define serverPanic(...) _serverPanic(__FILE__, __LINE__, __VA_ARGS__), valkey_unreachable()
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2024-06-02 13:15:08 -07:00
|
|
|
/* The following macros provide a conditional assertion that is only executed
|
2024-05-31 22:50:08 -07:00
|
|
|
* when the server config 'enable-debug-assert' is true. This is useful for adding
|
|
|
|
* assertions that are too computationally expensive or risky to run in normal
|
|
|
|
* operation, but are valuable for debugging or testing. */
|
2024-06-02 13:15:08 -07:00
|
|
|
#define debugServerAssert(...) (server.enable_debug_assert ? serverAssert(__VA_ARGS__) : (void)0)
|
2024-05-31 22:50:08 -07:00
|
|
|
#define debugServerAssertWithInfo(...) (server.enable_debug_assert ? serverAssertWithInfo(__VA_ARGS__) : (void)0)
|
2023-11-12 04:31:34 +00:00
|
|
|
|
Added INFO LATENCYSTATS section: latency by percentile distribution/latency by cumulative distribution of latencies (#9462)
# Short description
The Redis extended latency stats track per command latencies and enables:
- exporting the per-command percentile distribution via the `INFO LATENCYSTATS` command.
**( percentile distribution is not mergeable between cluster nodes ).**
- exporting the per-command cumulative latency distributions via the `LATENCY HISTOGRAM` command.
Using the cumulative distribution of latencies we can merge several stats from different cluster nodes
to calculate aggregate metrics .
By default, the extended latency monitoring is enabled since the overhead of keeping track of the
command latency is very small.
If you don't want to track extended latency metrics, you can easily disable it at runtime using the command:
- `CONFIG SET latency-tracking no`
By default, the exported latency percentiles are the p50, p99, and p999.
You can alter them at runtime using the command:
- `CONFIG SET latency-tracking-info-percentiles "0.0 50.0 100.0"`
## Some details:
- The total size per histogram should sit around 40 KiB. We only allocate those 40KiB when a command
was called for the first time.
- With regards to the WRITE overhead As seen below, there is no measurable overhead on the achievable
ops/sec or full latency spectrum on the client. Including also the measured redis-benchmark for unstable
vs this branch.
- We track from 1 nanosecond to 1 second ( everything above 1 second is considered +Inf )
## `INFO LATENCYSTATS` exposition format
- Format: `latency_percentiles_usec_<CMDNAME>:p0=XX,p50....`
## `LATENCY HISTOGRAM [command ...]` exposition format
Return a cumulative distribution of latencies in the format of a histogram for the specified command names.
The histogram is composed of a map of time buckets:
- Each representing a latency range, between 1 nanosecond and roughly 1 second.
- Each bucket covers twice the previous bucket's range.
- Empty buckets are not printed.
- Everything above 1 sec is considered +Inf.
- At max there will be log2(1000000000)=30 buckets
We reply a map for each command in the format:
`<command name> : { `calls`: <total command calls> , `histogram` : { <bucket 1> : latency , < bucket 2> : latency, ... } }`
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-05 12:01:05 +00:00
|
|
|
/* latency histogram per command init settings */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define LATENCY_HISTOGRAM_MIN_VALUE 1L /* >= 1 nanosec */
|
|
|
|
#define LATENCY_HISTOGRAM_MAX_VALUE 1000000000L /* <= 1 secs */
|
|
|
|
#define LATENCY_HISTOGRAM_PRECISION \
|
|
|
|
2 /* Maintain a value precision of 2 significant digits across LATENCY_HISTOGRAM_MIN_VALUE and \
|
2024-05-28 09:27:51 -07:00
|
|
|
* LATENCY_HISTOGRAM_MAX_VALUE range. Value quantization within the range will thus be no larger than 1/100th \
|
2024-05-22 23:24:12 -07:00
|
|
|
* (or 1%) of any value. The total size per histogram should sit around 40 KiB Bytes. */
|
Added INFO LATENCYSTATS section: latency by percentile distribution/latency by cumulative distribution of latencies (#9462)
# Short description
The Redis extended latency stats track per command latencies and enables:
- exporting the per-command percentile distribution via the `INFO LATENCYSTATS` command.
**( percentile distribution is not mergeable between cluster nodes ).**
- exporting the per-command cumulative latency distributions via the `LATENCY HISTOGRAM` command.
Using the cumulative distribution of latencies we can merge several stats from different cluster nodes
to calculate aggregate metrics .
By default, the extended latency monitoring is enabled since the overhead of keeping track of the
command latency is very small.
If you don't want to track extended latency metrics, you can easily disable it at runtime using the command:
- `CONFIG SET latency-tracking no`
By default, the exported latency percentiles are the p50, p99, and p999.
You can alter them at runtime using the command:
- `CONFIG SET latency-tracking-info-percentiles "0.0 50.0 100.0"`
## Some details:
- The total size per histogram should sit around 40 KiB. We only allocate those 40KiB when a command
was called for the first time.
- With regards to the WRITE overhead As seen below, there is no measurable overhead on the achievable
ops/sec or full latency spectrum on the client. Including also the measured redis-benchmark for unstable
vs this branch.
- We track from 1 nanosecond to 1 second ( everything above 1 second is considered +Inf )
## `INFO LATENCYSTATS` exposition format
- Format: `latency_percentiles_usec_<CMDNAME>:p0=XX,p50....`
## `LATENCY HISTOGRAM [command ...]` exposition format
Return a cumulative distribution of latencies in the format of a histogram for the specified command names.
The histogram is composed of a map of time buckets:
- Each representing a latency range, between 1 nanosecond and roughly 1 second.
- Each bucket covers twice the previous bucket's range.
- Empty buckets are not printed.
- Everything above 1 sec is considered +Inf.
- At max there will be log2(1000000000)=30 buckets
We reply a map for each command in the format:
`<command name> : { `calls`: <total command calls> , `histogram` : { <bucket 1> : latency , < bucket 2> : latency, ... } }`
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-05 12:01:05 +00:00
|
|
|
|
2022-01-20 09:05:53 +02:00
|
|
|
/* Busy module flags, see busy_module_yield_flags */
|
|
|
|
#define BUSY_MODULE_YIELD_NONE (0)
|
2024-05-22 23:24:12 -07:00
|
|
|
#define BUSY_MODULE_YIELD_EVENTS (1 << 0)
|
|
|
|
#define BUSY_MODULE_YIELD_CLIENTS (1 << 1)
|
2022-01-20 09:05:53 +02:00
|
|
|
|
Io thread work offload (#763)
### IO-Threads Work Offloading
This PR is the 2nd of 3 PRs intended to achieve the goal of 1M requests
per second.
(1st PR: https://github.com/valkey-io/valkey/pull/758)
This PR offloads additional work to the I/O threads, beyond the current
read-parse/write operations, to better utilize the I/O threads and
reduce the load on the main thread.
It contains the following 3 commits:
### Poll Offload
Currently, the main thread is responsible for executing the poll-wait
system call, while the IO threads wait for tasks from the main thread.
The poll-wait operation is expensive and can consume up to 30% of the
main thread's time. We could have let the IO threads do the poll-wait by
themselves, with each thread listening to some of the clients and
notifying the main thread when a client's command is ready to execute.
However, the current approach, where the main thread listens for events
from the network, has several benefits. The main thread remains in
charge, allowing it to know the state of each client
(idle/read/write/close) at any given time. Additionally, it makes the
threads flexible, enabling us to drain an IO thread's job queue and stop
a thread when the load is light without modifying the event loop and
moving its clients to a different IO thread. Furthermore, with this
approach, the IO threads don't need to wait for both messages from the
network and from the main thread instead, the threads wait only for
tasks from the main thread.
To enjoy the benefits of both the main thread remaining in charge and
the poll being offloaded, we propose offloading the poll-wait as a
single-time, non-blocking job to one of the IO threads. The IO thread
will perform a poll-wait non-blocking call while the main thread
processes the client commands. Later, in `aeProcessEvents`, instead of
sleeping on the poll, we check for the IO thread's poll-wait results.
The poll-wait will be offloaded in `beforeSleep` only when there are
ready events for the main thread to process. If no events are pending,
the main thread will revert to the current behavior and sleep on the
poll by itself.
**Implementation Details**
A new call back `custompoll` was added to the `aeEventLoop` when not set
to `NULL` the ae will call the `custompoll` callback instead of the
`aeApiPoll`.
When the poll is offloaded we will set the `custompoll` to
`getIOThreadPollResults` and send a poll-job to the thread. the thread
will take a mutex, call a non-blocking (with timeout 0) to `aePoll`
which will populate the fired events array. the IO thread will set the
`server.io_fired_events` to the number of the returning `numevents`,
later the main-thread in `custompoll` will return the
`server.io_fired_events` and will set the `customPoll` back to `NULL`.
To ensure thread safety when accessing server.el, all functions that
modify the eventloop events were wrapped with a mutex to ensure mutual
exclusion when modifying the events.
### Command Lookup Offload
As the IO thread parses the command from the client's Querybuf, it can
perform a command lookup in the commands dictionary, which can consume
up to ~5% of the main-thread runtime.
**Implementation details**
The IO thread will store the looked-up command in the client's new field
`io_parsed_cmd` field. We can't use `c->cmd` for that since we use
`c->cmd `to check if a command was reprocessed or not.
To ensure thread safety when accessing the command dictionary, we make
sure the main thread isn't changing the dictionary while IO threads are
accessing it. This is accomplished by introducing a new flag called
`no_incremental_rehash` for the `dictType` commands. When performing
`dictResize`, we will rehash the entire dictionary in place rather than
deferring the process.
### Free Offload
Since the command arguments are allocated by the I/O thread, it would be
beneficial if they were also freed by the same thread. If the main
thread frees objects allocated by the I/O thread, two issues arise:
1. During the freeing process, the main thread needs to access the SDS
pointed to by the object to get its length.
2. With Jemalloc, each thread manages thread local pool (`tcache`) of
buffers for quick reallocation without accessing the arena. If the main
thread constantly frees objects allocated by other threads, those
threads will have to frequently access the shared arena to obtain new
memory allocations
**Implementation Details**
When freeing the client's argv, we will send the argv array to the
thread that allocated it. The thread will be identified by the client
ID. When freeing an object during `dbOverwrite`, we will offload the
object free as well. We will extend this to offload the free during
`dbDelete` in a future PR, as its effects on defrag/memory evictions
need to be studied.
---------
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-19 05:21:45 +03:00
|
|
|
/* IO poll */
|
|
|
|
typedef enum { AE_IO_STATE_NONE, AE_IO_STATE_POLL, AE_IO_STATE_DONE } AeIoState;
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/*-----------------------------------------------------------------------------
|
|
|
|
* Data types
|
|
|
|
*----------------------------------------------------------------------------*/
|
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* An Object, that is a type able to hold a string / list / set */
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* The actual Object */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define OBJ_STRING 0 /* String object. */
|
|
|
|
#define OBJ_LIST 1 /* List object. */
|
|
|
|
#define OBJ_SET 2 /* Set object. */
|
|
|
|
#define OBJ_ZSET 3 /* Sorted set object. */
|
|
|
|
#define OBJ_HASH 4 /* Hash object. */
|
2015-07-30 11:46:31 +02:00
|
|
|
|
2016-05-18 11:45:40 +02:00
|
|
|
/* The "module" object type is a special one that signals that the object
|
2024-04-09 01:24:03 -07:00
|
|
|
* is one directly managed by a module. In this case the value points
|
2016-05-18 11:45:40 +02:00
|
|
|
* to a moduleValue struct, which contains the object value (which is only
|
2024-04-05 16:59:55 -07:00
|
|
|
* handled by the module itself) and the ValkeyModuleType struct which lists
|
2016-05-18 11:45:40 +02:00
|
|
|
* function pointers in order to serialize, deserialize, AOF-rewrite and
|
|
|
|
* free the object.
|
|
|
|
*
|
|
|
|
* Inside the RDB file, module types are encoded as OBJ_MODULE followed
|
|
|
|
* by a 64 bit module type ID, which has a 54 bits module-specific signature
|
|
|
|
* in order to dispatch the loading to the right module, plus a 10 bits
|
|
|
|
* encoding version. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define OBJ_MODULE 5 /* Module object. */
|
|
|
|
#define OBJ_STREAM 6 /* Stream object. */
|
|
|
|
#define OBJ_TYPE_MAX 7 /* Maximum number of object types */
|
2016-05-18 11:45:40 +02:00
|
|
|
|
|
|
|
/* Extract encver / signature from a module type ID. */
|
2024-04-05 16:59:55 -07:00
|
|
|
#define VALKEYMODULE_TYPE_ENCVER_BITS 10
|
2024-05-22 23:24:12 -07:00
|
|
|
#define VALKEYMODULE_TYPE_ENCVER_MASK ((1 << VALKEYMODULE_TYPE_ENCVER_BITS) - 1)
|
2024-04-05 16:59:55 -07:00
|
|
|
#define VALKEYMODULE_TYPE_ENCVER(id) ((id) & VALKEYMODULE_TYPE_ENCVER_MASK)
|
2024-05-22 23:24:12 -07:00
|
|
|
#define VALKEYMODULE_TYPE_SIGN(id) \
|
|
|
|
(((id) & ~((uint64_t)VALKEYMODULE_TYPE_ENCVER_MASK)) >> VALKEYMODULE_TYPE_ENCVER_BITS)
|
2016-05-18 11:45:40 +02:00
|
|
|
|
2019-07-21 17:41:03 +03:00
|
|
|
/* Bit flags for moduleTypeAuxSaveFunc */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define VALKEYMODULE_AUX_BEFORE_RDB (1 << 0)
|
|
|
|
#define VALKEYMODULE_AUX_AFTER_RDB (1 << 1)
|
2019-07-21 17:41:03 +03:00
|
|
|
|
2024-04-05 16:59:55 -07:00
|
|
|
struct ValkeyModule;
|
|
|
|
struct ValkeyModuleIO;
|
|
|
|
struct ValkeyModuleDigest;
|
|
|
|
struct ValkeyModuleCtx;
|
2018-04-20 21:01:21 +08:00
|
|
|
struct moduleLoadQueueEntry;
|
2024-04-05 16:59:55 -07:00
|
|
|
struct ValkeyModuleKeyOptCtx;
|
|
|
|
struct ValkeyModuleCommand;
|
2023-10-30 12:38:43 +02:00
|
|
|
struct clusterState;
|
2016-05-18 11:45:40 +02:00
|
|
|
|
|
|
|
/* Each module type implementation should export a set of methods in order
|
|
|
|
* to serialize and deserialize the value in the RDB file, rewrite the AOF
|
|
|
|
* log, create the digest for "DEBUG DIGEST", and free the value when a key
|
|
|
|
* is deleted. */
|
2024-04-05 16:59:55 -07:00
|
|
|
typedef void *(*moduleTypeLoadFunc)(struct ValkeyModuleIO *io, int encver);
|
|
|
|
typedef void (*moduleTypeSaveFunc)(struct ValkeyModuleIO *io, void *value);
|
|
|
|
typedef int (*moduleTypeAuxLoadFunc)(struct ValkeyModuleIO *rdb, int encver, int when);
|
|
|
|
typedef void (*moduleTypeAuxSaveFunc)(struct ValkeyModuleIO *rdb, int when);
|
|
|
|
typedef void (*moduleTypeRewriteFunc)(struct ValkeyModuleIO *io, struct serverObject *key, void *value);
|
|
|
|
typedef void (*moduleTypeDigestFunc)(struct ValkeyModuleDigest *digest, void *value);
|
2017-01-12 12:47:46 +01:00
|
|
|
typedef size_t (*moduleTypeMemUsageFunc)(const void *value);
|
2016-05-18 11:45:40 +02:00
|
|
|
typedef void (*moduleTypeFreeFunc)(void *value);
|
2024-04-04 00:04:51 +07:00
|
|
|
typedef size_t (*moduleTypeFreeEffortFunc)(struct serverObject *key, const void *value);
|
|
|
|
typedef void (*moduleTypeUnlinkFunc)(struct serverObject *key, void *value);
|
|
|
|
typedef void *(*moduleTypeCopyFunc)(struct serverObject *fromkey, struct serverObject *tokey, const void *value);
|
2024-04-05 16:59:55 -07:00
|
|
|
typedef int (*moduleTypeDefragFunc)(struct ValkeyModuleDefragCtx *ctx, struct serverObject *key, void **value);
|
|
|
|
typedef size_t (*moduleTypeMemUsageFunc2)(struct ValkeyModuleKeyOptCtx *ctx, const void *value, size_t sample_size);
|
|
|
|
typedef void (*moduleTypeFreeFunc2)(struct ValkeyModuleKeyOptCtx *ctx, void *value);
|
|
|
|
typedef size_t (*moduleTypeFreeEffortFunc2)(struct ValkeyModuleKeyOptCtx *ctx, const void *value);
|
|
|
|
typedef void (*moduleTypeUnlinkFunc2)(struct ValkeyModuleKeyOptCtx *ctx, void *value);
|
|
|
|
typedef void *(*moduleTypeCopyFunc2)(struct ValkeyModuleKeyOptCtx *ctx, const void *value);
|
|
|
|
typedef int (*moduleTypeAuthCallback)(struct ValkeyModuleCtx *ctx, void *username, void *password, const char **err);
|
2016-05-18 11:45:40 +02:00
|
|
|
|
2019-02-26 01:23:11 +00:00
|
|
|
|
2016-05-18 11:45:40 +02:00
|
|
|
/* The module type, which is referenced in each value of a given type, defines
|
|
|
|
* the methods and links to the module exporting the type. */
|
2024-04-05 16:59:55 -07:00
|
|
|
typedef struct ValkeyModuleType {
|
2016-05-18 11:45:40 +02:00
|
|
|
uint64_t id; /* Higher 54 bits of type ID + 10 lower bits of encoding ver. */
|
2024-04-05 16:59:55 -07:00
|
|
|
struct ValkeyModule *module;
|
2016-05-18 11:45:40 +02:00
|
|
|
moduleTypeLoadFunc rdb_load;
|
|
|
|
moduleTypeSaveFunc rdb_save;
|
|
|
|
moduleTypeRewriteFunc aof_rewrite;
|
2016-11-30 11:13:55 +01:00
|
|
|
moduleTypeMemUsageFunc mem_usage;
|
2016-05-18 11:45:40 +02:00
|
|
|
moduleTypeDigestFunc digest;
|
|
|
|
moduleTypeFreeFunc free;
|
2020-11-16 16:34:04 +08:00
|
|
|
moduleTypeFreeEffortFunc free_effort;
|
|
|
|
moduleTypeUnlinkFunc unlink;
|
2020-12-09 20:22:45 +02:00
|
|
|
moduleTypeCopyFunc copy;
|
2020-12-13 09:56:01 +02:00
|
|
|
moduleTypeDefragFunc defrag;
|
2019-07-21 17:41:03 +03:00
|
|
|
moduleTypeAuxLoadFunc aux_load;
|
|
|
|
moduleTypeAuxSaveFunc aux_save;
|
2021-06-16 14:45:49 +08:00
|
|
|
moduleTypeMemUsageFunc2 mem_usage2;
|
|
|
|
moduleTypeFreeEffortFunc2 free_effort2;
|
|
|
|
moduleTypeUnlinkFunc2 unlink2;
|
|
|
|
moduleTypeCopyFunc2 copy2;
|
Avoid saving module aux on RDB if no aux data was saved by the module. (#11374)
### Background
The issue is that when saving an RDB with module AUX data, the module AUX metadata
(moduleid, when, ...) is saved to the RDB even though the module did not saved any actual data.
This prevent loading the RDB in the absence of the module (although there is no actual data in
the RDB that requires the module to be loaded).
### Solution
The solution suggested in this PR is that module AUX will be saved on the RDB only if the module
actually saved something during `aux_save` function.
To support backward compatibility, we introduce `aux_save2` callback that acts the same as
`aux_save` with the tiny change of avoid saving the aux field if no data was actually saved by
the module. Modules can use the new API to make sure that if they have no data to save,
then it will be possible to load the created RDB even without the module.
### Concerns
A module may register for the aux load and save hooks just in order to be notified when
saving or loading starts or completed (there are better ways to do that, but it still possible
that someone used it).
However, if a module didn't save a single field in the save callback, it means it's not allowed
to read in the read callback, since it has no way to distinguish between empty and non-empty
payloads. furthermore, it means that if the module did that, it must never change it, since it'll
break compatibility with it's old RDB files, so this is really not a valid use case.
Since some modules (ones who currently save one field indicating an empty payload), need
to know if saving an empty payload is valid, and if Redis is gonna ignore an empty payload
or store it, we opted to add a new API (rather than change behavior of an existing API and
expect modules to check the redis version)
### Technical Details
To avoid saving AUX data on RDB, we change the code to first save the AUX metadata
(moduleid, when, ...) into a temporary buffer. The buffer is then flushed to the rio at the first
time the module makes a write operation inside the `aux_save` function. If the module saves
nothing (and `aux_save2` was used), the entire temporary buffer is simply dropped and no
data about this AUX field is saved to the RDB. This make it possible to load the RDB even in
the absence of the module.
Test was added to verify the fix.
2022-10-18 19:45:46 +03:00
|
|
|
moduleTypeAuxSaveFunc aux_save2;
|
2019-07-21 17:41:03 +03:00
|
|
|
int aux_save_triggers;
|
2016-05-18 11:45:40 +02:00
|
|
|
char name[10]; /* 9 bytes name + null term. Charset: A-Z a-z 0-9 _- */
|
|
|
|
} moduleType;
|
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* In Object 'robj' structures of type OBJ_MODULE, the value pointer
|
2016-05-18 11:45:40 +02:00
|
|
|
* is set to the following structure, referencing the moduleType structure
|
|
|
|
* in order to work with the value, and at the same time providing a raw
|
|
|
|
* pointer to the value, as created by the module commands operating with
|
|
|
|
* the module type.
|
|
|
|
*
|
|
|
|
* So for example in order to free such a value, it is possible to use
|
|
|
|
* the following code:
|
|
|
|
*
|
|
|
|
* if (robj->type == OBJ_MODULE) {
|
|
|
|
* moduleValue *mt = robj->ptr;
|
|
|
|
* mt->type->free(mt->value);
|
|
|
|
* zfree(mt); // We need to release this in-the-middle struct as well.
|
|
|
|
* }
|
|
|
|
*/
|
|
|
|
typedef struct moduleValue {
|
|
|
|
moduleType *type;
|
|
|
|
void *value;
|
|
|
|
} moduleValue;
|
|
|
|
|
2018-04-20 21:01:21 +08:00
|
|
|
/* This structure represents a module inside the system. */
|
2024-04-05 16:59:55 -07:00
|
|
|
struct ValkeyModule {
|
2024-05-22 23:24:12 -07:00
|
|
|
void *handle; /* Module dlopen() handle. */
|
|
|
|
char *name; /* Module name. */
|
|
|
|
int ver; /* Module version. We use just progressive integers. */
|
|
|
|
int apiver; /* Module API version as requested during initialization.*/
|
|
|
|
list *types; /* Module data types. */
|
|
|
|
list *usedby; /* List of modules using APIs from this one. */
|
|
|
|
list *using; /* List of modules we use some APIs of. */
|
|
|
|
list *filters; /* List of filters the module has registered. */
|
|
|
|
list *module_configs; /* List of configurations the module has registered */
|
|
|
|
int configs_initialized; /* Have the module configurations been initialized? */
|
|
|
|
int in_call; /* RM_Call() nesting level */
|
|
|
|
int in_hook; /* Hooks callback nesting level for this module (0 or 1). */
|
|
|
|
int options; /* Module options and capabilities. */
|
|
|
|
int blocked_clients; /* Count of ValkeyModuleBlockedClient in this module. */
|
|
|
|
ValkeyModuleInfoFunc info_cb; /* Callback for module to add INFO fields. */
|
|
|
|
ValkeyModuleDefragFunc defrag_cb; /* Callback for global data defrag. */
|
2018-04-20 21:01:21 +08:00
|
|
|
struct moduleLoadQueueEntry *loadmod; /* Module load arguments for config rewrite. */
|
2023-03-21 10:07:11 -07:00
|
|
|
int num_commands_with_acl_categories; /* Number of commands in this module included in acl categories */
|
2024-05-22 23:24:12 -07:00
|
|
|
int onload; /* Flag to identify if the call is being made from Onload (0 or 1) */
|
|
|
|
size_t num_acl_categories_added; /* Number of acl categories added by this module. */
|
2018-04-20 21:01:21 +08:00
|
|
|
};
|
2024-04-05 16:59:55 -07:00
|
|
|
typedef struct ValkeyModule ValkeyModule;
|
2018-04-20 21:01:21 +08:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* This is a wrapper for the 'rio' streams used inside rdb.c in the server, so that
|
2016-05-18 11:45:40 +02:00
|
|
|
* the user does not have to take the total count of the written bytes nor
|
|
|
|
* to care about error conditions. */
|
2024-04-05 16:59:55 -07:00
|
|
|
struct ValkeyModuleIO {
|
2024-05-22 23:24:12 -07:00
|
|
|
size_t bytes; /* Bytes read / written so far. */
|
|
|
|
rio *rio; /* Rio stream. */
|
|
|
|
moduleType *type; /* Module type doing the operation. */
|
|
|
|
int error; /* True if error condition happened. */
|
2024-04-05 16:59:55 -07:00
|
|
|
struct ValkeyModuleCtx *ctx; /* Optional context, see RM_GetContextFromIO()*/
|
2024-04-04 00:04:51 +07:00
|
|
|
struct serverObject *key; /* Optional name of key processed */
|
2024-05-22 23:24:12 -07:00
|
|
|
int dbid; /* The dbid of the key being processed, -1 when unknown. */
|
|
|
|
sds pre_flush_buffer; /* A buffer that should be flushed before next write operation
|
|
|
|
* See rdbSaveSingleModuleAux for more details */
|
Build TLS as a loadable module
* Support BUILD_TLS=module to be loaded as a module via config file or
command line. e.g. redis-server --loadmodule redis-tls.so
* Updates to redismodule.h to allow it to be used side by side with
server.h by defining REDISMODULE_CORE_MODULE
* Changes to server.h, redismodule.h and module.c to avoid repeated
type declarations (gcc 4.8 doesn't like these)
* Add a mechanism for non-ABI neutral modules (ones who include
server.h) to refuse loading if they detect not being built together with
redis (release.c)
* Fix wrong signature of RedisModuleDefragFunc, this could break
compilation of a module, but not the ABI
* Move initialization of listeners in server.c to be after loading
the modules
* Config TLS after initialization of listeners
* Init cluster after initialization of listeners
* Add TLS module to CI
* Fix a test suite race conditions:
Now that the listeners are initialized later, it's not sufficient to
wait for the PID message in the log, we need to wait for the "Server
Initialized" message.
* Fix issues with moduleconfigs test as a result from start_server
waiting for "Server Initialized"
* Fix issues with modules/infra test as a result of an additional module
present
Notes about Sentinel:
Sentinel can't really rely on the tls module, since it uses hiredis to
initiate connections and depends on OpenSSL (won't be able to use any
other connection modules for that), so it was decided that when TLS is
built as a module, sentinel does not support TLS at all.
This means that it keeps using redis_tls_ctx and redis_tls_client_ctx directly.
Example code of config in redis-tls.so(may be use in the future):
RedisModuleString *tls_cfg = NULL;
void tlsInfo(RedisModuleInfoCtx *ctx, int for_crash_report) {
UNUSED(for_crash_report);
RedisModule_InfoAddSection(ctx, "");
RedisModule_InfoAddFieldLongLong(ctx, "var", 42);
}
int tlsCommand(RedisModuleCtx *ctx, RedisModuleString **argv, int argc)
{
if (argc != 2) return RedisModule_WrongArity(ctx);
return RedisModule_ReplyWithString(ctx, argv[1]);
}
RedisModuleString *getStringConfigCommand(const char *name, void *privdata) {
REDISMODULE_NOT_USED(name);
REDISMODULE_NOT_USED(privdata);
return tls_cfg;
}
int setStringConfigCommand(const char *name, RedisModuleString *new, void *privdata, RedisModuleString **err) {
REDISMODULE_NOT_USED(name);
REDISMODULE_NOT_USED(err);
REDISMODULE_NOT_USED(privdata);
if (tls_cfg) RedisModule_FreeString(NULL, tls_cfg);
RedisModule_RetainString(NULL, new);
tls_cfg = new;
return REDISMODULE_OK;
}
int RedisModule_OnLoad(void *ctx, RedisModuleString **argv, int argc)
{
....
if (RedisModule_CreateCommand(ctx,"tls",tlsCommand,"",0,0,0) == REDISMODULE_ERR)
return REDISMODULE_ERR;
if (RedisModule_RegisterStringConfig(ctx, "cfg", "", REDISMODULE_CONFIG_DEFAULT, getStringConfigCommand, setStringConfigCommand, NULL, NULL) == REDISMODULE_ERR)
return REDISMODULE_ERR;
if (RedisModule_LoadConfigs(ctx) == REDISMODULE_ERR) {
if (tls_cfg) {
RedisModule_FreeString(ctx, tls_cfg);
tls_cfg = NULL;
}
return REDISMODULE_ERR;
}
...
}
Co-authored-by: zhenwei pi <pizhenwei@bytedance.com>
Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
2022-08-22 15:53:56 +08:00
|
|
|
};
|
2016-05-18 11:45:40 +02:00
|
|
|
|
RDB modules values serialization format version 2.
The original RDB serialization format was not parsable without the
module loaded, becuase the structure was managed only by the module
itself. Moreover RDB is a streaming protocol in the sense that it is
both produce di an append-only fashion, and is also sometimes directly
sent to the socket (in the case of diskless replication).
The fact that modules values cannot be parsed without the relevant
module loaded is a problem in many ways: RDB checking tools must have
loaded modules even for doing things not involving the value at all,
like splitting an RDB into N RDBs by key or alike, or just checking the
RDB for sanity.
In theory module values could be just a blob of data with a prefixed
length in order for us to be able to skip it. However prefixing the values
with a length would mean one of the following:
1. To be able to write some data at a previous offset. This breaks
stremaing.
2. To bufferize values before outputting them. This breaks performances.
3. To have some chunked RDB output format. This breaks simplicity.
Moreover, the above solution, still makes module values a totally opaque
matter, with the fowllowing problems:
1. The RDB check tool can just skip the value without being able to at
least check the general structure. For datasets composed mostly of
modules values this means to just check the outer level of the RDB not
actually doing any checko on most of the data itself.
2. It is not possible to do any recovering or processing of data for which a
module no longer exists in the future, or is unknown.
So this commit implements a different solution. The modules RDB
serialization API is composed if well defined calls to store integers,
floats, doubles or strings. After this commit, the parts generated by
the module API have a one-byte prefix for each of the above emitted
parts, and there is a final EOF byte as well. So even if we don't know
exactly how to interpret a module value, we can always parse it at an
high level, check the overall structure, understand the types used to
store the information, and easily skip the whole value.
The change is backward compatible: older RDB files can be still loaded
since the new encoding has a new RDB type: MODULE_2 (of value 7).
The commit also implements the ability to check RDB files for sanity
taking advantage of the new feature.
2017-06-27 13:09:33 +02:00
|
|
|
/* Macro to initialize an IO context. Note that the 'ver' field is populated
|
|
|
|
* inside rdb.c according to the version of the value to load. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define moduleInitIOContext(iovar, mtype, rioptr, keyptr, db) \
|
|
|
|
do { \
|
|
|
|
iovar.rio = rioptr; \
|
|
|
|
iovar.type = mtype; \
|
|
|
|
iovar.bytes = 0; \
|
|
|
|
iovar.error = 0; \
|
|
|
|
iovar.key = keyptr; \
|
|
|
|
iovar.dbid = db; \
|
|
|
|
iovar.ctx = NULL; \
|
|
|
|
iovar.pre_flush_buffer = NULL; \
|
|
|
|
} while (0)
|
2016-05-18 11:45:40 +02:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* This is a structure used to export DEBUG DIGEST capabilities to
|
2017-07-06 10:29:19 +02:00
|
|
|
* modules. We want to capture both the ordered and unordered elements of
|
|
|
|
* a data structure, so that a digest can be created in a way that correctly
|
|
|
|
* reflects the values. See the DEBUG DIGEST command implementation for more
|
|
|
|
* background. */
|
2024-04-05 16:59:55 -07:00
|
|
|
struct ValkeyModuleDigest {
|
2024-05-22 23:24:12 -07:00
|
|
|
unsigned char o[20]; /* Ordered elements. */
|
|
|
|
unsigned char x[20]; /* Xored elements. */
|
2024-04-04 00:04:51 +07:00
|
|
|
struct serverObject *key; /* Optional name of key processed */
|
2024-05-22 23:24:12 -07:00
|
|
|
int dbid; /* The dbid of the key being processed */
|
Build TLS as a loadable module
* Support BUILD_TLS=module to be loaded as a module via config file or
command line. e.g. redis-server --loadmodule redis-tls.so
* Updates to redismodule.h to allow it to be used side by side with
server.h by defining REDISMODULE_CORE_MODULE
* Changes to server.h, redismodule.h and module.c to avoid repeated
type declarations (gcc 4.8 doesn't like these)
* Add a mechanism for non-ABI neutral modules (ones who include
server.h) to refuse loading if they detect not being built together with
redis (release.c)
* Fix wrong signature of RedisModuleDefragFunc, this could break
compilation of a module, but not the ABI
* Move initialization of listeners in server.c to be after loading
the modules
* Config TLS after initialization of listeners
* Init cluster after initialization of listeners
* Add TLS module to CI
* Fix a test suite race conditions:
Now that the listeners are initialized later, it's not sufficient to
wait for the PID message in the log, we need to wait for the "Server
Initialized" message.
* Fix issues with moduleconfigs test as a result from start_server
waiting for "Server Initialized"
* Fix issues with modules/infra test as a result of an additional module
present
Notes about Sentinel:
Sentinel can't really rely on the tls module, since it uses hiredis to
initiate connections and depends on OpenSSL (won't be able to use any
other connection modules for that), so it was decided that when TLS is
built as a module, sentinel does not support TLS at all.
This means that it keeps using redis_tls_ctx and redis_tls_client_ctx directly.
Example code of config in redis-tls.so(may be use in the future):
RedisModuleString *tls_cfg = NULL;
void tlsInfo(RedisModuleInfoCtx *ctx, int for_crash_report) {
UNUSED(for_crash_report);
RedisModule_InfoAddSection(ctx, "");
RedisModule_InfoAddFieldLongLong(ctx, "var", 42);
}
int tlsCommand(RedisModuleCtx *ctx, RedisModuleString **argv, int argc)
{
if (argc != 2) return RedisModule_WrongArity(ctx);
return RedisModule_ReplyWithString(ctx, argv[1]);
}
RedisModuleString *getStringConfigCommand(const char *name, void *privdata) {
REDISMODULE_NOT_USED(name);
REDISMODULE_NOT_USED(privdata);
return tls_cfg;
}
int setStringConfigCommand(const char *name, RedisModuleString *new, void *privdata, RedisModuleString **err) {
REDISMODULE_NOT_USED(name);
REDISMODULE_NOT_USED(err);
REDISMODULE_NOT_USED(privdata);
if (tls_cfg) RedisModule_FreeString(NULL, tls_cfg);
RedisModule_RetainString(NULL, new);
tls_cfg = new;
return REDISMODULE_OK;
}
int RedisModule_OnLoad(void *ctx, RedisModuleString **argv, int argc)
{
....
if (RedisModule_CreateCommand(ctx,"tls",tlsCommand,"",0,0,0) == REDISMODULE_ERR)
return REDISMODULE_ERR;
if (RedisModule_RegisterStringConfig(ctx, "cfg", "", REDISMODULE_CONFIG_DEFAULT, getStringConfigCommand, setStringConfigCommand, NULL, NULL) == REDISMODULE_ERR)
return REDISMODULE_ERR;
if (RedisModule_LoadConfigs(ctx) == REDISMODULE_ERR) {
if (tls_cfg) {
RedisModule_FreeString(ctx, tls_cfg);
tls_cfg = NULL;
}
return REDISMODULE_ERR;
}
...
}
Co-authored-by: zhenwei pi <pizhenwei@bytedance.com>
Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
2022-08-22 15:53:56 +08:00
|
|
|
};
|
2017-07-06 10:29:19 +02:00
|
|
|
|
|
|
|
/* Just start with a digest composed of all zero bytes. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define moduleInitDigestContext(mdvar) \
|
|
|
|
do { \
|
|
|
|
memset(mdvar.o, 0, sizeof(mdvar.o)); \
|
|
|
|
memset(mdvar.x, 0, sizeof(mdvar.x)); \
|
|
|
|
} while (0)
|
2017-07-06 10:29:19 +02:00
|
|
|
|
Custom authentication for Modules (#11659)
This change adds new module callbacks that can override the default password based authentication associated with ACLs. With this, Modules can register auth callbacks through which they can implement their own Authentication logic. When `AUTH` and `HELLO AUTH ...` commands are used, Module based authentication is attempted and then normal password based authentication is attempted if needed.
The new Module APIs added in this PR are - `RM_RegisterCustomAuthCallback` and `RM_BlockClientOnAuth` and `RedisModule_ACLAddLogEntryByUserName `.
Module based authentication will be attempted for all Redis users (created through the ACL SETUSER cmd or through Module APIs) even if the Redis user does not exist at the time of the command. This gives a chance for the Module to create the RedisModule user and then authenticate via the RedisModule API - from the custom auth callback.
For the AUTH command, we will support both variations - `AUTH <username> <password>` and `AUTH <password>`. In case of the `AUTH <password>` variation, the custom auth callbacks are triggered with “default” as the username and password as what is provided.
### RedisModule_RegisterCustomAuthCallback
```
void RM_RegisterCustomAuthCallback(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback cb) {
```
This API registers a callback to execute to prior to normal password based authentication. Multiple callbacks can be registered across different modules. These callbacks are responsible for either handling the authentication, each authenticating the user or explicitly denying, or deferring it to other authentication mechanisms. Callbacks are triggered in the order they were registered. When a Module is unloaded, all the auth callbacks registered by it are unregistered. The callbacks are attempted, in the order of most recently registered callbacks, when the AUTH/HELLO (with AUTH field is provided) commands are called. The callbacks will be called with a module context along with a username and a password, and are expected to take one of the following actions:
(1) Authenticate - Use the RM_Authenticate* API successfully and return `REDISMODULE_AUTH_HANDLED`. This will immediately end the auth chain as successful and add the OK reply.
(2) Block a client on authentication - Use the `RM_BlockClientOnAuth` API and return `REDISMODULE_AUTH_HANDLED`. Here, the client will be blocked until the `RM_UnblockClient `API is used which will trigger the auth reply callback (provided earlier through the `RM_BlockClientOnAuth`). In this reply callback, the Module should authenticate, deny or skip handling authentication.
(3) Deny Authentication - Return `REDISMODULE_AUTH_HANDLED` without authenticating or blocking the client. Optionally, `err` can be set to a custom error message. This will immediately end the auth chain as unsuccessful and add the ERR reply.
(4) Skip handling Authentication - Return `REDISMODULE_AUTH_NOT_HANDLED` without blocking the client. This will allow the engine to attempt the next custom auth callback.
If none of the callbacks authenticate or deny auth, then password based auth is attempted and will authenticate or add failure logs and reply to the clients accordingly.
### RedisModule_BlockClientOnAuth
```
RedisModuleBlockedClient *RM_BlockClientOnAuth(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback reply_callback,
void (*free_privdata)(RedisModuleCtx*,void*))
```
This API can only be used from a Module from the custom auth callback. If a client is not in the middle of custom module based authentication, ERROR is returned. Otherwise, the client is blocked and the `RedisModule_BlockedClient` is returned similar to the `RedisModule_BlockClient` API.
### RedisModule_ACLAddLogEntryByUserName
```
int RM_ACLAddLogEntryByUserName(RedisModuleCtx *ctx, RedisModuleString *username, RedisModuleString *object, RedisModuleACLLogEntryReason reason)
```
Adds a new entry in the ACL log with the `username` RedisModuleString provided. This simplifies the Module usage because now, developers do not need to create a Module User just to add an error ACL Log entry. Aside from accepting username (RedisModuleString) instead of a RedisModuleUser, it is the same as the existing `RedisModule_ACLAddLogEntry` API.
### Breaking changes
- HELLO command - Clients can now only set the client name and RESP protocol from the `HELLO` command if they are authenticated. Also, we now finish command arg validation first and return early with a ERR reply if any arg is invalid. This is to avoid mutating the client name / RESP from a command that would have failed on invalid arguments.
### Notable behaviors
- Module unblocking - Now, we will not allow Modules to block the client from inside the context of a reply callback (triggered from the Module unblock flow `moduleHandleBlockedClients`).
---------
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
2023-03-15 15:18:42 -07:00
|
|
|
/* Macro to check if the client is in the middle of module based authentication. */
|
|
|
|
#define clientHasModuleAuthInProgress(c) ((c)->module_auth_ctx != NULL)
|
|
|
|
|
2015-07-30 11:46:31 +02:00
|
|
|
/* Objects encoding. Some kind of objects like Strings and Hashes can be
|
|
|
|
* internally represented in multiple ways. The 'encoding' field of the object
|
|
|
|
* is set to one of this fields for this object. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define OBJ_ENCODING_RAW 0 /* Raw representation */
|
|
|
|
#define OBJ_ENCODING_INT 1 /* Encoded as integer */
|
|
|
|
#define OBJ_ENCODING_HT 2 /* Encoded as hash table */
|
|
|
|
#define OBJ_ENCODING_ZIPMAP 3 /* No longer used: old hash encoding. */
|
2016-09-13 10:26:36 +02:00
|
|
|
#define OBJ_ENCODING_LINKEDLIST 4 /* No longer used: old list encoding. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define OBJ_ENCODING_ZIPLIST 5 /* No longer used: old list/hash/zset encoding. */
|
|
|
|
#define OBJ_ENCODING_INTSET 6 /* Encoded as intset */
|
|
|
|
#define OBJ_ENCODING_SKIPLIST 7 /* Encoded as skiplist */
|
|
|
|
#define OBJ_ENCODING_EMBSTR 8 /* Embedded sds string encoding */
|
|
|
|
#define OBJ_ENCODING_QUICKLIST 9 /* Encoded as linked list of listpacks */
|
|
|
|
#define OBJ_ENCODING_STREAM 10 /* Encoded as a radix tree of listpacks */
|
|
|
|
#define OBJ_ENCODING_LISTPACK 11 /* Encoded as a listpack */
|
2015-07-30 11:46:31 +02:00
|
|
|
|
2015-07-27 09:41:48 +02:00
|
|
|
#define LRU_BITS 24
|
2024-05-22 23:24:12 -07:00
|
|
|
#define LRU_CLOCK_MAX ((1 << LRU_BITS) - 1) /* Max value of obj->lru */
|
|
|
|
#define LRU_CLOCK_RESOLUTION 1000 /* LRU clock resolution in ms */
|
2015-07-30 11:46:31 +02:00
|
|
|
|
2024-05-22 23:24:12 -07:00
|
|
|
#define OBJ_SHARED_REFCOUNT INT_MAX /* Global object never destroyed. */
|
|
|
|
#define OBJ_STATIC_REFCOUNT (INT_MAX - 1) /* Object allocated in the stack. */
|
2020-04-09 16:20:41 +02:00
|
|
|
#define OBJ_FIRST_SPECIAL_REFCOUNT OBJ_STATIC_REFCOUNT
|
2024-04-04 00:04:51 +07:00
|
|
|
struct serverObject {
|
2024-05-22 23:24:12 -07:00
|
|
|
unsigned type : 4;
|
|
|
|
unsigned encoding : 4;
|
|
|
|
unsigned lru : LRU_BITS; /* LRU time (relative to global lru_clock) or
|
|
|
|
* LFU data (least significant 8 bits frequency
|
|
|
|
* and most significant 16 bits access time). */
|
2010-06-22 00:07:48 +02:00
|
|
|
int refcount;
|
|
|
|
void *ptr;
|
Build TLS as a loadable module
* Support BUILD_TLS=module to be loaded as a module via config file or
command line. e.g. redis-server --loadmodule redis-tls.so
* Updates to redismodule.h to allow it to be used side by side with
server.h by defining REDISMODULE_CORE_MODULE
* Changes to server.h, redismodule.h and module.c to avoid repeated
type declarations (gcc 4.8 doesn't like these)
* Add a mechanism for non-ABI neutral modules (ones who include
server.h) to refuse loading if they detect not being built together with
redis (release.c)
* Fix wrong signature of RedisModuleDefragFunc, this could break
compilation of a module, but not the ABI
* Move initialization of listeners in server.c to be after loading
the modules
* Config TLS after initialization of listeners
* Init cluster after initialization of listeners
* Add TLS module to CI
* Fix a test suite race conditions:
Now that the listeners are initialized later, it's not sufficient to
wait for the PID message in the log, we need to wait for the "Server
Initialized" message.
* Fix issues with moduleconfigs test as a result from start_server
waiting for "Server Initialized"
* Fix issues with modules/infra test as a result of an additional module
present
Notes about Sentinel:
Sentinel can't really rely on the tls module, since it uses hiredis to
initiate connections and depends on OpenSSL (won't be able to use any
other connection modules for that), so it was decided that when TLS is
built as a module, sentinel does not support TLS at all.
This means that it keeps using redis_tls_ctx and redis_tls_client_ctx directly.
Example code of config in redis-tls.so(may be use in the future):
RedisModuleString *tls_cfg = NULL;
void tlsInfo(RedisModuleInfoCtx *ctx, int for_crash_report) {
UNUSED(for_crash_report);
RedisModule_InfoAddSection(ctx, "");
RedisModule_InfoAddFieldLongLong(ctx, "var", 42);
}
int tlsCommand(RedisModuleCtx *ctx, RedisModuleString **argv, int argc)
{
if (argc != 2) return RedisModule_WrongArity(ctx);
return RedisModule_ReplyWithString(ctx, argv[1]);
}
RedisModuleString *getStringConfigCommand(const char *name, void *privdata) {
REDISMODULE_NOT_USED(name);
REDISMODULE_NOT_USED(privdata);
return tls_cfg;
}
int setStringConfigCommand(const char *name, RedisModuleString *new, void *privdata, RedisModuleString **err) {
REDISMODULE_NOT_USED(name);
REDISMODULE_NOT_USED(err);
REDISMODULE_NOT_USED(privdata);
if (tls_cfg) RedisModule_FreeString(NULL, tls_cfg);
RedisModule_RetainString(NULL, new);
tls_cfg = new;
return REDISMODULE_OK;
}
int RedisModule_OnLoad(void *ctx, RedisModuleString **argv, int argc)
{
....
if (RedisModule_CreateCommand(ctx,"tls",tlsCommand,"",0,0,0) == REDISMODULE_ERR)
return REDISMODULE_ERR;
if (RedisModule_RegisterStringConfig(ctx, "cfg", "", REDISMODULE_CONFIG_DEFAULT, getStringConfigCommand, setStringConfigCommand, NULL, NULL) == REDISMODULE_ERR)
return REDISMODULE_ERR;
if (RedisModule_LoadConfigs(ctx) == REDISMODULE_ERR) {
if (tls_cfg) {
RedisModule_FreeString(ctx, tls_cfg);
tls_cfg = NULL;
}
return REDISMODULE_ERR;
}
...
}
Co-authored-by: zhenwei pi <pizhenwei@bytedance.com>
Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
2022-08-22 15:53:56 +08:00
|
|
|
};
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2023-03-16 14:50:32 +08:00
|
|
|
/* The string name for an object's type as listed above
|
2019-07-08 11:04:37 +01:00
|
|
|
* Native types are checked against the OBJ_STRING, OBJ_LIST, OBJ_* defines,
|
|
|
|
* and Module types have their registered name returned. */
|
2024-05-22 23:24:12 -07:00
|
|
|
char *getObjectTypeName(robj *);
|
2019-06-10 17:41:44 +01:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* Macro used to initialize an Object allocated on the stack.
|
2010-06-22 00:07:48 +02:00
|
|
|
* Note that this macro is taken near the structure definition to make sure
|
|
|
|
* we'll update it when the structure is changed, to avoid bugs like
|
|
|
|
* bug #85 introduced exactly in this way. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define initStaticStringObject(_var, _ptr) \
|
|
|
|
do { \
|
|
|
|
_var.refcount = OBJ_STATIC_REFCOUNT; \
|
|
|
|
_var.type = OBJ_STRING; \
|
|
|
|
_var.encoding = OBJ_ENCODING_RAW; \
|
|
|
|
_var.ptr = _ptr; \
|
|
|
|
} while (0)
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2016-07-07 15:04:25 +02:00
|
|
|
struct evictionPoolEntry; /* Defined in evict.c */
|
2014-03-20 11:57:29 +01:00
|
|
|
|
2018-07-16 17:56:54 +02:00
|
|
|
/* This structure is used in order to represent the output buffer of a client,
|
|
|
|
* which is actually a linked list of blocks like that, that is: client->reply. */
|
2018-02-21 20:18:34 +02:00
|
|
|
typedef struct clientReplyBlock {
|
|
|
|
size_t size, used;
|
|
|
|
char buf[];
|
|
|
|
} clientReplyBlock;
|
|
|
|
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
/* Replication buffer blocks is the list of replBufBlock.
|
|
|
|
*
|
|
|
|
* +--------------+ +--------------+ +--------------+
|
|
|
|
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
|
|
|
|
* +--------------+ +--------------+ +--------------+
|
|
|
|
* | / \
|
|
|
|
* | / \
|
|
|
|
* | / \
|
2023-03-13 22:12:29 +08:00
|
|
|
* Repl Backlog Replica_A Replica_B
|
2023-03-16 14:50:32 +08:00
|
|
|
*
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
* Each replica or replication backlog increments only the refcount of the
|
|
|
|
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
|
|
|
|
* node, it should first increase the next node's refcount, and when we trim
|
|
|
|
* the replication buffer nodes, we remove node always from the head node which
|
|
|
|
* refcount is 0. If the refcount of the head node is not 0, we must stop
|
|
|
|
* trimming and never iterate the next node. */
|
|
|
|
|
|
|
|
/* Similar with 'clientReplyBlock', it is used for shared buffers between
|
|
|
|
* all replica clients and replication backlog. */
|
|
|
|
typedef struct replBufBlock {
|
2024-05-22 23:24:12 -07:00
|
|
|
int refcount; /* Number of replicas or repl backlog using. */
|
|
|
|
long long id; /* The unique incremental number. */
|
|
|
|
long long repl_offset; /* Start replication offset of the block. */
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
size_t size, used;
|
|
|
|
char buf[];
|
|
|
|
} replBufBlock;
|
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* Database representation. There are multiple databases identified
|
2014-03-20 11:57:29 +01:00
|
|
|
* by integers from 0 (the default database) up to the max configured
|
|
|
|
* database. The database number is the 'id' field in the structure. */
|
2024-04-03 10:02:43 +07:00
|
|
|
typedef struct serverDb {
|
2024-05-22 23:24:12 -07:00
|
|
|
kvstore *keys; /* The keyspace for this DB */
|
|
|
|
kvstore *expires; /* Timeout of keys with a timeout set */
|
|
|
|
dict *blocking_keys; /* Keys with clients waiting for data (BLPOP)*/
|
|
|
|
dict *blocking_keys_unblock_on_nokey; /* Keys with clients waiting for
|
|
|
|
* data, and should be unblocked if key is deleted (XREADEDGROUP).
|
|
|
|
* This is a subset of blocking_keys*/
|
|
|
|
dict *ready_keys; /* Blocked keys that received a PUSH */
|
|
|
|
dict *watched_keys; /* WATCHED keys for MULTI/EXEC CAS */
|
|
|
|
int id; /* Database ID */
|
|
|
|
long long avg_ttl; /* Average TTL, just for stats */
|
|
|
|
unsigned long expires_cursor; /* Cursor of the active expire cycle. */
|
|
|
|
list *defrag_later; /* List of key names to attempt to defrag one by one, gradually. */
|
2024-04-03 10:02:43 +07:00
|
|
|
} serverDb;
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2021-10-07 14:41:26 +03:00
|
|
|
/* forward declaration for functions ctx */
|
Redis Function Libraries (#10004)
# Redis Function Libraries
This PR implements Redis Functions Libraries as describe on: https://github.com/redis/redis/issues/9906.
Libraries purpose is to provide a better code sharing between functions by allowing to create multiple
functions in a single command. Functions that were created together can safely share code between
each other without worrying about compatibility issues and versioning.
Creating a new library is done using 'FUNCTION LOAD' command (full API is described below)
This PR introduces a new struct called libraryInfo, libraryInfo holds information about a library:
* name - name of the library
* engine - engine used to create the library
* code - library code
* description - library description
* functions - the functions exposed by the library
When Redis gets the `FUNCTION LOAD` command it creates a new empty libraryInfo.
Redis passes the `CODE` to the relevant engine alongside the empty libraryInfo.
As a result, the engine will create one or more functions by calling 'libraryCreateFunction'.
The new funcion will be added to the newly created libraryInfo. So far Everything is happening
locally on the libraryInfo so it is easy to abort the operation (in case of an error) by simply
freeing the libraryInfo. After the library info is fully constructed we start the joining phase by
which we will join the new library to the other libraries currently exist on Redis.
The joining phase make sure there is no function collision and add the library to the
librariesCtx (renamed from functionCtx). LibrariesCtx is used all around the code in the exact
same way as functionCtx was used (with respect to RDB loading, replicatio, ...).
The only difference is that apart from function dictionary (maps function name to functionInfo
object), the librariesCtx contains also a libraries dictionary that maps library name to libraryInfo object.
## New API
### FUNCTION LOAD
`FUNCTION LOAD <ENGINE> <LIBRARY NAME> [REPLACE] [DESCRIPTION <DESCRIPTION>] <CODE>`
Create a new library with the given parameters:
* ENGINE - REPLACE Engine name to use to create the library.
* LIBRARY NAME - The new library name.
* REPLACE - If the library already exists, replace it.
* DESCRIPTION - Library description.
* CODE - Library code.
Return "OK" on success, or error on the following cases:
* Library name already taken and REPLACE was not used
* Name collision with another existing library (even if replace was uses)
* Library registration failed by the engine (usually compilation error)
## Changed API
### FUNCTION LIST
`FUNCTION LIST [LIBRARYNAME <LIBRARY NAME PATTERN>] [WITHCODE]`
Command was modified to also allow getting libraries code (so `FUNCTION INFO` command is no longer
needed and removed). In addition the command gets an option argument, `LIBRARYNAME` allows you to
only get libraries that match the given `LIBRARYNAME` pattern. By default, it returns all libraries.
### INFO MEMORY
Added number of libraries to `INFO MEMORY`
### Commands flags
`DENYOOM` flag was set on `FUNCTION LOAD` and `FUNCTION RESTORE`. We consider those commands
as commands that add new data to the dateset (functions are data) and so we want to disallows
to run those commands on OOM.
## Removed API
* FUNCTION CREATE - Decided on https://github.com/redis/redis/issues/9906
* FUNCTION INFO - Decided on https://github.com/redis/redis/issues/9899
## Lua engine changes
When the Lua engine gets the code given on `FUNCTION LOAD` command, it immediately runs it, we call
this run the loading run. Loading run is not a usual script run, it is not possible to invoke any
Redis command from within the load run.
Instead there is a new API provided by `library` object. The new API's:
* `redis.log` - behave the same as `redis.log`
* `redis.register_function` - register a new function to the library
The loading run purpose is to register functions using the new `redis.register_function` API.
Any attempt to use any other API will result in an error. In addition, the load run is has a time
limit of 500ms, error is raise on timeout and the entire operation is aborted.
### `redis.register_function`
`redis.register_function(<function_name>, <callback>, [<description>])`
This new API allows users to register a new function that will be linked to the newly created library.
This API can only be called during the load run (see definition above). Any attempt to use it outside
of the load run will result in an error.
The parameters pass to the API are:
* function_name - Function name (must be a Lua string)
* callback - Lua function object that will be called when the function is invokes using fcall/fcall_ro
* description - Function description, optional (must be a Lua string).
### Example
The following example creates a library called `lib` with 2 functions, `f1` and `f1`, returns 1 and 2 respectively:
```
local function f1(keys, args)
return 1
end
local function f2(keys, args)
return 2
end
redis.register_function('f1', f1)
redis.register_function('f2', f2)
```
Notice: Unlike `eval`, functions inside a library get the KEYS and ARGV as arguments to the
functions and not as global.
### Technical Details
On the load run we only want the user to be able to call a white list on API's. This way, in
the future, if new API's will be added, the new API's will not be available to the load run
unless specifically added to this white list. We put the while list on the `library` object and
make sure the `library` object is only available to the load run by using [lua_setfenv](https://www.lua.org/manual/5.1/manual.html#lua_setfenv) API. This API allows us to set
the `globals` of a function (and all the function it creates). Before starting the load run we
create a new fresh Lua table (call it `g`) that only contains the `library` API (we make sure
to set global protection on this table just like the general global protection already exists
today), then we use [lua_setfenv](https://www.lua.org/manual/5.1/manual.html#lua_setfenv)
to set `g` as the global table of the load run. After the load run finished we update `g`
metatable and set `__index` and `__newindex` functions to be `_G` (Lua default globals),
we also pop out the `library` object as we do not need it anymore.
This way, any function that was created on the load run (and will be invoke using `fcall`) will
see the default globals as it expected to see them and will not have the `library` API anymore.
An important outcome of this new approach is that now we can achieve a distinct global table
for each library (it is not yet like that but it is very easy to achieve it now). In the future we can
decide to remove global protection because global on different libraries will not collide or we
can chose to give different API to different libraries base on some configuration or input.
Notice that this technique was meant to prevent errors and was not meant to prevent malicious
user from exploit it. For example, the load run can still save the `library` object on some local
variable and then using in `fcall` context. To prevent such a malicious use, the C code also make
sure it is running in the right context and if not raise an error.
2022-01-06 13:39:38 +02:00
|
|
|
typedef struct functionsLibCtx functionsLibCtx;
|
2021-10-07 14:41:26 +03:00
|
|
|
|
|
|
|
/* Holding object that need to be populated during
|
|
|
|
* rdb loading. On loading end it is possible to decide
|
|
|
|
* whether not to set those objects on their rightful place.
|
|
|
|
* For example: dbarray need to be set as main database on
|
|
|
|
* successful loading and dropped on failure. */
|
|
|
|
typedef struct rdbLoadingCtx {
|
2024-05-22 23:24:12 -07:00
|
|
|
serverDb *dbarray;
|
|
|
|
functionsLibCtx *functions_lib_ctx;
|
|
|
|
} rdbLoadingCtx;
|
2021-10-07 14:41:26 +03:00
|
|
|
|
2024-06-07 20:32:27 -07:00
|
|
|
typedef sds (*rdbAuxFieldEncoder)(int flags);
|
|
|
|
typedef int (*rdbAuxFieldDecoder)(int flags, sds s);
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Client MULTI/EXEC state */
|
|
|
|
typedef struct multiCmd {
|
|
|
|
robj **argv;
|
2021-10-06 11:39:09 +03:00
|
|
|
int argv_len;
|
2010-06-22 00:07:48 +02:00
|
|
|
int argc;
|
2024-04-03 23:54:33 +07:00
|
|
|
struct serverCommand *cmd;
|
2010-06-22 00:07:48 +02:00
|
|
|
} multiCmd;
|
|
|
|
|
|
|
|
typedef struct multiState {
|
2024-05-22 23:24:12 -07:00
|
|
|
multiCmd *commands; /* Array of MULTI commands */
|
|
|
|
int count; /* Total number of MULTI commands */
|
|
|
|
int cmd_flags; /* The accumulated command flags OR-ed together.
|
|
|
|
So if at least a command has a given flag, it
|
|
|
|
will be set in this field. */
|
|
|
|
int cmd_inv_flags; /* Same as cmd_flags, OR-ing the ~flags. so that it
|
|
|
|
is possible to know if all the commands have a
|
|
|
|
certain flag. */
|
|
|
|
size_t argv_len_sums; /* mem used by all commands arguments */
|
|
|
|
int alloc_count; /* total number of multiCmd struct memory reserved. */
|
2010-06-22 00:07:48 +02:00
|
|
|
} multiState;
|
|
|
|
|
2013-12-03 17:43:53 +01:00
|
|
|
/* This structure holds the blocking operation state for a client.
|
2024-08-18 21:20:35 +08:00
|
|
|
* The fields used depend on client->bstate.btype. */
|
2010-11-08 19:38:01 -03:00
|
|
|
typedef struct blockingState {
|
2013-12-03 17:43:53 +01:00
|
|
|
/* Generic fields. */
|
2024-05-22 23:24:12 -07:00
|
|
|
blocking_type btype; /* Type of blocking op if CLIENT_BLOCKED. */
|
|
|
|
mstime_t timeout; /* Blocking operation timeout. If UNIX current time
|
|
|
|
* is > timeout then the operation timed out. */
|
|
|
|
int unblock_on_nokey; /* Whether to unblock the client when at least one of the keys
|
|
|
|
is deleted or does not exist anymore */
|
2024-08-20 13:54:53 +03:00
|
|
|
union {
|
|
|
|
listNode *client_waiting_acks_list_node; /* list node in server.clients_waiting_acks list. */
|
|
|
|
listNode *postponed_list_node; /* list node in server.postponed_clients */
|
|
|
|
listNode *generic_blocked_list_node; /* generic placeholder for blocked clients utility lists.
|
|
|
|
Since a client cannot be blocked multiple times, we can assume
|
|
|
|
it will be held in only one extra utility list, so it is ok to maintain
|
|
|
|
a union of these listNode references. */
|
|
|
|
};
|
2024-08-18 21:20:35 +08:00
|
|
|
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
/* BLOCKED_LIST, BLOCKED_ZSET and BLOCKED_STREAM or any other Keys related blocking */
|
2024-05-22 23:24:12 -07:00
|
|
|
dict *keys; /* The keys we are blocked on */
|
2017-09-07 09:30:50 +02:00
|
|
|
|
2023-03-16 00:16:16 +08:00
|
|
|
/* BLOCKED_WAIT and BLOCKED_WAITAOF */
|
2024-08-20 13:54:53 +03:00
|
|
|
int numreplicas; /* Number of replicas we are waiting for ACK. */
|
|
|
|
int numlocal; /* Indication if WAITAOF is waiting for local fsync. */
|
|
|
|
long long reploffset; /* Replication offset to reach. */
|
2016-10-07 11:55:35 +02:00
|
|
|
|
|
|
|
/* BLOCKED_MODULE */
|
2024-04-05 16:59:55 -07:00
|
|
|
void *module_blocked_handle; /* ValkeyModuleBlockedClient structure.
|
2016-10-07 11:55:35 +02:00
|
|
|
which is opaque for the Redis core, only
|
|
|
|
handled in module.c. */
|
Support for RM_Call on blocking commands (#11568)
Allow running blocking commands from within a module using `RM_Call`.
Today, when `RM_Call` is used, the fake client that is used to run command
is marked with `CLIENT_DENY_BLOCKING` flag. This flag tells the command
that it is not allowed to block the client and in case it needs to block, it must
fallback to some alternative (either return error or perform some default behavior).
For example, `BLPOP` fallback to simple `LPOP` if it is not allowed to block.
All the commands must respect the `CLIENT_DENY_BLOCKING` flag (including
module commands). When the command invocation finished, Redis asserts that
the client was not blocked.
This PR introduces the ability to call blocking command using `RM_Call` by
passing a callback that will be called when the client will get unblocked.
In order to do that, the user must explicitly say that he allow to perform blocking
command by passing a new format specifier argument, `K`, to the `RM_Call`
function. This new flag will tell Redis that it is allow to run blocking command
and block the client. In case the command got blocked, Redis will return a new
type of call reply (`REDISMODULE_REPLY_PROMISE`). This call reply indicates
that the command got blocked and the user can set the on_unblocked handler using
`RM_CallReplyPromiseSetUnblockHandler`.
When clients gets unblocked, it eventually reaches `processUnblockedClients` function.
This is where we check if the client is a fake module client and if it is, we call the unblock
callback instead of performing the usual unblock operations.
**Notice**: `RM_CallReplyPromiseSetUnblockHandler` must be called atomically
along side the command invocation (without releasing the Redis lock in between).
In addition, unlike other CallReply types, the promise call reply must be released
by the module when the Redis GIL is acquired.
The module can abort the execution on the blocking command (if it was not yet
executed) using `RM_CallReplyPromiseAbort`. the API will return `REDISMODULE_OK`
on success and `REDISMODULE_ERR` if the operation is already executed.
**Notice** that in case of misbehave module, Abort might finished successfully but the
operation will not really be aborted. This can only happened if the module do not respect
the disconnect callback of the blocked client.
For pure Redis commands this can not happened.
### Atomicity Guarantees
The API promise that the unblock handler will run atomically as an execution unit.
This means that all the operation performed on the unblock handler will be wrapped
with a multi exec transaction when replicated to the replica and AOF.
The API **do not** grantee any other atomicity properties such as when the unblock
handler will be called. This gives us the flexibility to strengthen the grantees (or not)
in the future if we will decide that we need a better guarantees.
That said, the implementation **does** provide a better guarantees when performing
pure Redis blocking command like `BLPOP`. In this case the unblock handler will run
atomically with the operation that got unblocked (for example, in case of `BLPOP`, the
unblock handler will run atomically with the `LPOP` operation that run when the command
got unblocked). This is an implementation detail that might be change in the future and the
module writer should not count on that.
### Calling blocking commands while running on script mode (`S`)
`RM_Call` script mode (`S`) was introduced on #0372. It is used for usecases where the
command that was invoked on `RM_Call` comes from a user input and we want to make
sure the user will not run dangerous commands like `shutdown`. Some command, such
as `BLPOP`, are marked with `NO_SCRIPT` flag, which means they will not be allowed on
script mode. Those commands are marked with `NO_SCRIPT` just because they are
blocking commands and not because they are dangerous. Now that we can run blocking
commands on RM_Call, there is no real reason not to allow such commands on script mode.
The underline problem is that the `NO_SCRIPT` flag is abused to also mark some of the
blocking commands (notice that those commands know not to block the client if it is not
allowed to do so, and have a fallback logic to such cases. So even if those commands
were not marked with `NO_SCRIPT` flag, it would not harm Redis, and today we can
already run those commands within multi exec).
In addition, not all blocking commands are marked with `NO_SCRIPT` flag, for example
`blmpop` are not marked and can run from within a script.
Those facts shows that there are some ambiguity about the meaning of the `NO_SCRIPT`
flag, and its not fully clear where it should be use.
The PR suggest that blocking commands should not be marked with `NO_SCRIPT` flag,
those commands should handle `CLIENT_DENY_BLOCKING` flag and only block when
it's safe (like they already does today). To achieve that, the PR removes the `NO_SCRIPT`
flag from the following commands:
* `blmove`
* `blpop`
* `brpop`
* `brpoplpush`
* `bzpopmax`
* `bzpopmin`
* `wait`
This might be considered a breaking change as now, on scripts, instead of getting
`command is not allowed from script` error, the user will get some fallback behavior
base on the command implementation. That said, the change matches the behavior
of scripts and multi exec with respect to those commands and allow running them on
`RM_Call` even when script mode is used.
### Additional RedisModule API and changes
* `RM_BlockClientSetPrivateData` - Set private data on the blocked client without the
need to unblock the client. This allows up to set the promise CallReply as the private
data of the blocked client and abort it if the client gets disconnected.
* `RM_BlockClientGetPrivateData` - Return the current private data set on a blocked client.
We need it so we will have access to this private data on the disconnect callback.
* On RM_Call, the returned reply will be added to the auto memory context only if auto
memory is enabled, this allows us to keep the call reply for longer time then the context
lifetime and does not force an unneeded borrow relationship between the CallReply and
the RedisModuleContext.
2023-03-16 14:04:31 +02:00
|
|
|
|
2024-04-05 16:59:55 -07:00
|
|
|
void *async_rm_call_handle; /* ValkeyModuleAsyncRMCallPromise structure.
|
Support for RM_Call on blocking commands (#11568)
Allow running blocking commands from within a module using `RM_Call`.
Today, when `RM_Call` is used, the fake client that is used to run command
is marked with `CLIENT_DENY_BLOCKING` flag. This flag tells the command
that it is not allowed to block the client and in case it needs to block, it must
fallback to some alternative (either return error or perform some default behavior).
For example, `BLPOP` fallback to simple `LPOP` if it is not allowed to block.
All the commands must respect the `CLIENT_DENY_BLOCKING` flag (including
module commands). When the command invocation finished, Redis asserts that
the client was not blocked.
This PR introduces the ability to call blocking command using `RM_Call` by
passing a callback that will be called when the client will get unblocked.
In order to do that, the user must explicitly say that he allow to perform blocking
command by passing a new format specifier argument, `K`, to the `RM_Call`
function. This new flag will tell Redis that it is allow to run blocking command
and block the client. In case the command got blocked, Redis will return a new
type of call reply (`REDISMODULE_REPLY_PROMISE`). This call reply indicates
that the command got blocked and the user can set the on_unblocked handler using
`RM_CallReplyPromiseSetUnblockHandler`.
When clients gets unblocked, it eventually reaches `processUnblockedClients` function.
This is where we check if the client is a fake module client and if it is, we call the unblock
callback instead of performing the usual unblock operations.
**Notice**: `RM_CallReplyPromiseSetUnblockHandler` must be called atomically
along side the command invocation (without releasing the Redis lock in between).
In addition, unlike other CallReply types, the promise call reply must be released
by the module when the Redis GIL is acquired.
The module can abort the execution on the blocking command (if it was not yet
executed) using `RM_CallReplyPromiseAbort`. the API will return `REDISMODULE_OK`
on success and `REDISMODULE_ERR` if the operation is already executed.
**Notice** that in case of misbehave module, Abort might finished successfully but the
operation will not really be aborted. This can only happened if the module do not respect
the disconnect callback of the blocked client.
For pure Redis commands this can not happened.
### Atomicity Guarantees
The API promise that the unblock handler will run atomically as an execution unit.
This means that all the operation performed on the unblock handler will be wrapped
with a multi exec transaction when replicated to the replica and AOF.
The API **do not** grantee any other atomicity properties such as when the unblock
handler will be called. This gives us the flexibility to strengthen the grantees (or not)
in the future if we will decide that we need a better guarantees.
That said, the implementation **does** provide a better guarantees when performing
pure Redis blocking command like `BLPOP`. In this case the unblock handler will run
atomically with the operation that got unblocked (for example, in case of `BLPOP`, the
unblock handler will run atomically with the `LPOP` operation that run when the command
got unblocked). This is an implementation detail that might be change in the future and the
module writer should not count on that.
### Calling blocking commands while running on script mode (`S`)
`RM_Call` script mode (`S`) was introduced on #0372. It is used for usecases where the
command that was invoked on `RM_Call` comes from a user input and we want to make
sure the user will not run dangerous commands like `shutdown`. Some command, such
as `BLPOP`, are marked with `NO_SCRIPT` flag, which means they will not be allowed on
script mode. Those commands are marked with `NO_SCRIPT` just because they are
blocking commands and not because they are dangerous. Now that we can run blocking
commands on RM_Call, there is no real reason not to allow such commands on script mode.
The underline problem is that the `NO_SCRIPT` flag is abused to also mark some of the
blocking commands (notice that those commands know not to block the client if it is not
allowed to do so, and have a fallback logic to such cases. So even if those commands
were not marked with `NO_SCRIPT` flag, it would not harm Redis, and today we can
already run those commands within multi exec).
In addition, not all blocking commands are marked with `NO_SCRIPT` flag, for example
`blmpop` are not marked and can run from within a script.
Those facts shows that there are some ambiguity about the meaning of the `NO_SCRIPT`
flag, and its not fully clear where it should be use.
The PR suggest that blocking commands should not be marked with `NO_SCRIPT` flag,
those commands should handle `CLIENT_DENY_BLOCKING` flag and only block when
it's safe (like they already does today). To achieve that, the PR removes the `NO_SCRIPT`
flag from the following commands:
* `blmove`
* `blpop`
* `brpop`
* `brpoplpush`
* `bzpopmax`
* `bzpopmin`
* `wait`
This might be considered a breaking change as now, on scripts, instead of getting
`command is not allowed from script` error, the user will get some fallback behavior
base on the command implementation. That said, the change matches the behavior
of scripts and multi exec with respect to those commands and allow running them on
`RM_Call` even when script mode is used.
### Additional RedisModule API and changes
* `RM_BlockClientSetPrivateData` - Set private data on the blocked client without the
need to unblock the client. This allows up to set the promise CallReply as the private
data of the blocked client and abort it if the client gets disconnected.
* `RM_BlockClientGetPrivateData` - Return the current private data set on a blocked client.
We need it so we will have access to this private data on the disconnect callback.
* On RM_Call, the returned reply will be added to the auto memory context only if auto
memory is enabled, this allows us to keep the call reply for longer time then the context
lifetime and does not force an unneeded borrow relationship between the CallReply and
the RedisModuleContext.
2023-03-16 14:04:31 +02:00
|
|
|
which is opaque for the Redis core, only
|
|
|
|
handled in module.c. */
|
2010-11-08 19:38:01 -03:00
|
|
|
} blockingState;
|
|
|
|
|
A reimplementation of blocking operation internals.
Redis provides support for blocking operations such as BLPOP or BRPOP.
This operations are identical to normal LPOP and RPOP operations as long
as there are elements in the target list, but if the list is empty they
block waiting for new data to arrive to the list.
All the clients blocked waiting for th same list are served in a FIFO
way, so the first that blocked is the first to be served when there is
more data pushed by another client into the list.
The previous implementation of blocking operations was conceived to
serve clients in the context of push operations. For for instance:
1) There is a client "A" blocked on list "foo".
2) The client "B" performs `LPUSH foo somevalue`.
3) The client "A" is served in the context of the "B" LPUSH,
synchronously.
Processing things in a synchronous way was useful as if "A" pushes a
value that is served by "B", from the point of view of the database is a
NOP (no operation) thing, that is, nothing is replicated, nothing is
written in the AOF file, and so forth.
However later we implemented two things:
1) Variadic LPUSH that could add multiple values to a list in the
context of a single call.
2) BRPOPLPUSH that was a version of BRPOP that also provided a "PUSH"
side effect when receiving data.
This forced us to make the synchronous implementation more complex. If
client "B" is waiting for data, and "A" pushes three elemnents in a
single call, we needed to propagate an LPUSH with a missing argument
in the AOF and replication link. We also needed to make sure to
replicate the LPUSH side of BRPOPLPUSH, but only if in turn did not
happened to serve another blocking client into another list ;)
This were complex but with a few of mutually recursive functions
everything worked as expected... until one day we introduced scripting
in Redis.
Scripting + synchronous blocking operations = Issue #614.
Basically you can't "rewrite" a script to have just a partial effect on
the replicas and AOF file if the script happened to serve a few blocked
clients.
The solution to all this problems, implemented by this commit, is to
change the way we serve blocked clients. Instead of serving the blocked
clients synchronously, in the context of the command performing the PUSH
operation, it is now an asynchronous and iterative process:
1) If a key that has clients blocked waiting for data is the subject of
a list push operation, We simply mark keys as "ready" and put it into a
queue.
2) Every command pushing stuff on lists, as a variadic LPUSH, a script,
or whatever it is, is replicated verbatim without any rewriting.
3) Every time a Redis command, a MULTI/EXEC block, or a script,
completed its execution, we run the list of keys ready to serve blocked
clients (as more data arrived), and process this list serving the
blocked clients.
4) As a result of "3" maybe more keys are ready again for other clients
(as a result of BRPOPLPUSH we may have push operations), so we iterate
back to step "3" if it's needed.
The new code has a much simpler semantics, and a simpler to understand
implementation, with the disadvantage of not being able to "optmize out"
a PUSH+BPOP as a No OP.
This commit will be tested with care before the final merge, more tests
will be added likely.
2012-09-04 10:37:49 +02:00
|
|
|
/* The following structure represents a node in the server.ready_keys list,
|
|
|
|
* where we accumulate all the keys that had clients blocked with a blocking
|
|
|
|
* operation such as B[LR]POP, but received new data in the context of the
|
|
|
|
* last executed command.
|
|
|
|
*
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
* After the execution of every command or script, we iterate over this list to check
|
A reimplementation of blocking operation internals.
Redis provides support for blocking operations such as BLPOP or BRPOP.
This operations are identical to normal LPOP and RPOP operations as long
as there are elements in the target list, but if the list is empty they
block waiting for new data to arrive to the list.
All the clients blocked waiting for th same list are served in a FIFO
way, so the first that blocked is the first to be served when there is
more data pushed by another client into the list.
The previous implementation of blocking operations was conceived to
serve clients in the context of push operations. For for instance:
1) There is a client "A" blocked on list "foo".
2) The client "B" performs `LPUSH foo somevalue`.
3) The client "A" is served in the context of the "B" LPUSH,
synchronously.
Processing things in a synchronous way was useful as if "A" pushes a
value that is served by "B", from the point of view of the database is a
NOP (no operation) thing, that is, nothing is replicated, nothing is
written in the AOF file, and so forth.
However later we implemented two things:
1) Variadic LPUSH that could add multiple values to a list in the
context of a single call.
2) BRPOPLPUSH that was a version of BRPOP that also provided a "PUSH"
side effect when receiving data.
This forced us to make the synchronous implementation more complex. If
client "B" is waiting for data, and "A" pushes three elemnents in a
single call, we needed to propagate an LPUSH with a missing argument
in the AOF and replication link. We also needed to make sure to
replicate the LPUSH side of BRPOPLPUSH, but only if in turn did not
happened to serve another blocking client into another list ;)
This were complex but with a few of mutually recursive functions
everything worked as expected... until one day we introduced scripting
in Redis.
Scripting + synchronous blocking operations = Issue #614.
Basically you can't "rewrite" a script to have just a partial effect on
the replicas and AOF file if the script happened to serve a few blocked
clients.
The solution to all this problems, implemented by this commit, is to
change the way we serve blocked clients. Instead of serving the blocked
clients synchronously, in the context of the command performing the PUSH
operation, it is now an asynchronous and iterative process:
1) If a key that has clients blocked waiting for data is the subject of
a list push operation, We simply mark keys as "ready" and put it into a
queue.
2) Every command pushing stuff on lists, as a variadic LPUSH, a script,
or whatever it is, is replicated verbatim without any rewriting.
3) Every time a Redis command, a MULTI/EXEC block, or a script,
completed its execution, we run the list of keys ready to serve blocked
clients (as more data arrived), and process this list serving the
blocked clients.
4) As a result of "3" maybe more keys are ready again for other clients
(as a result of BRPOPLPUSH we may have push operations), so we iterate
back to step "3" if it's needed.
The new code has a much simpler semantics, and a simpler to understand
implementation, with the disadvantage of not being able to "optmize out"
a PUSH+BPOP as a No OP.
This commit will be tested with care before the final merge, more tests
will be added likely.
2012-09-04 10:37:49 +02:00
|
|
|
* if as a result we should serve data to clients blocked, unblocking them.
|
|
|
|
* Note that server.ready_keys will not have duplicates as there dictionary
|
2024-04-09 01:24:03 -07:00
|
|
|
* also called ready_keys in every structure representing a database,
|
A reimplementation of blocking operation internals.
Redis provides support for blocking operations such as BLPOP or BRPOP.
This operations are identical to normal LPOP and RPOP operations as long
as there are elements in the target list, but if the list is empty they
block waiting for new data to arrive to the list.
All the clients blocked waiting for th same list are served in a FIFO
way, so the first that blocked is the first to be served when there is
more data pushed by another client into the list.
The previous implementation of blocking operations was conceived to
serve clients in the context of push operations. For for instance:
1) There is a client "A" blocked on list "foo".
2) The client "B" performs `LPUSH foo somevalue`.
3) The client "A" is served in the context of the "B" LPUSH,
synchronously.
Processing things in a synchronous way was useful as if "A" pushes a
value that is served by "B", from the point of view of the database is a
NOP (no operation) thing, that is, nothing is replicated, nothing is
written in the AOF file, and so forth.
However later we implemented two things:
1) Variadic LPUSH that could add multiple values to a list in the
context of a single call.
2) BRPOPLPUSH that was a version of BRPOP that also provided a "PUSH"
side effect when receiving data.
This forced us to make the synchronous implementation more complex. If
client "B" is waiting for data, and "A" pushes three elemnents in a
single call, we needed to propagate an LPUSH with a missing argument
in the AOF and replication link. We also needed to make sure to
replicate the LPUSH side of BRPOPLPUSH, but only if in turn did not
happened to serve another blocking client into another list ;)
This were complex but with a few of mutually recursive functions
everything worked as expected... until one day we introduced scripting
in Redis.
Scripting + synchronous blocking operations = Issue #614.
Basically you can't "rewrite" a script to have just a partial effect on
the replicas and AOF file if the script happened to serve a few blocked
clients.
The solution to all this problems, implemented by this commit, is to
change the way we serve blocked clients. Instead of serving the blocked
clients synchronously, in the context of the command performing the PUSH
operation, it is now an asynchronous and iterative process:
1) If a key that has clients blocked waiting for data is the subject of
a list push operation, We simply mark keys as "ready" and put it into a
queue.
2) Every command pushing stuff on lists, as a variadic LPUSH, a script,
or whatever it is, is replicated verbatim without any rewriting.
3) Every time a Redis command, a MULTI/EXEC block, or a script,
completed its execution, we run the list of keys ready to serve blocked
clients (as more data arrived), and process this list serving the
blocked clients.
4) As a result of "3" maybe more keys are ready again for other clients
(as a result of BRPOPLPUSH we may have push operations), so we iterate
back to step "3" if it's needed.
The new code has a much simpler semantics, and a simpler to understand
implementation, with the disadvantage of not being able to "optmize out"
a PUSH+BPOP as a No OP.
This commit will be tested with care before the final merge, more tests
will be added likely.
2012-09-04 10:37:49 +02:00
|
|
|
* where we make sure to remember if a given key was already added in the
|
|
|
|
* server.ready_keys list. */
|
|
|
|
typedef struct readyList {
|
2024-04-03 10:02:43 +07:00
|
|
|
serverDb *db;
|
A reimplementation of blocking operation internals.
Redis provides support for blocking operations such as BLPOP or BRPOP.
This operations are identical to normal LPOP and RPOP operations as long
as there are elements in the target list, but if the list is empty they
block waiting for new data to arrive to the list.
All the clients blocked waiting for th same list are served in a FIFO
way, so the first that blocked is the first to be served when there is
more data pushed by another client into the list.
The previous implementation of blocking operations was conceived to
serve clients in the context of push operations. For for instance:
1) There is a client "A" blocked on list "foo".
2) The client "B" performs `LPUSH foo somevalue`.
3) The client "A" is served in the context of the "B" LPUSH,
synchronously.
Processing things in a synchronous way was useful as if "A" pushes a
value that is served by "B", from the point of view of the database is a
NOP (no operation) thing, that is, nothing is replicated, nothing is
written in the AOF file, and so forth.
However later we implemented two things:
1) Variadic LPUSH that could add multiple values to a list in the
context of a single call.
2) BRPOPLPUSH that was a version of BRPOP that also provided a "PUSH"
side effect when receiving data.
This forced us to make the synchronous implementation more complex. If
client "B" is waiting for data, and "A" pushes three elemnents in a
single call, we needed to propagate an LPUSH with a missing argument
in the AOF and replication link. We also needed to make sure to
replicate the LPUSH side of BRPOPLPUSH, but only if in turn did not
happened to serve another blocking client into another list ;)
This were complex but with a few of mutually recursive functions
everything worked as expected... until one day we introduced scripting
in Redis.
Scripting + synchronous blocking operations = Issue #614.
Basically you can't "rewrite" a script to have just a partial effect on
the replicas and AOF file if the script happened to serve a few blocked
clients.
The solution to all this problems, implemented by this commit, is to
change the way we serve blocked clients. Instead of serving the blocked
clients synchronously, in the context of the command performing the PUSH
operation, it is now an asynchronous and iterative process:
1) If a key that has clients blocked waiting for data is the subject of
a list push operation, We simply mark keys as "ready" and put it into a
queue.
2) Every command pushing stuff on lists, as a variadic LPUSH, a script,
or whatever it is, is replicated verbatim without any rewriting.
3) Every time a Redis command, a MULTI/EXEC block, or a script,
completed its execution, we run the list of keys ready to serve blocked
clients (as more data arrived), and process this list serving the
blocked clients.
4) As a result of "3" maybe more keys are ready again for other clients
(as a result of BRPOPLPUSH we may have push operations), so we iterate
back to step "3" if it's needed.
The new code has a much simpler semantics, and a simpler to understand
implementation, with the disadvantage of not being able to "optmize out"
a PUSH+BPOP as a No OP.
This commit will be tested with care before the final merge, more tests
will be added likely.
2012-09-04 10:37:49 +02:00
|
|
|
robj *key;
|
|
|
|
} readyList;
|
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* This structure represents a user. This is useful for ACLs, the
|
2019-01-10 16:33:48 +01:00
|
|
|
* user is associated to the connection after the connection is authenticated.
|
|
|
|
* If there is no associated user, the connection uses the default user. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define USER_COMMAND_BITS_COUNT \
|
|
|
|
1024 /* The total number of command bits \
|
|
|
|
in the user structure. The last valid \
|
|
|
|
command ID we can set in the user \
|
|
|
|
is USER_COMMAND_BITS_COUNT-1. */
|
|
|
|
#define USER_FLAG_ENABLED (1 << 0) /* The user is active. */
|
|
|
|
#define USER_FLAG_DISABLED (1 << 1) /* The user is disabled. */
|
|
|
|
#define USER_FLAG_NOPASS \
|
|
|
|
(1 << 2) /* The user requires no password, any \
|
|
|
|
provided password will work. For the \
|
|
|
|
default user, this also means that \
|
|
|
|
no AUTH is needed, and every \
|
|
|
|
connection is immediately \
|
|
|
|
authenticated. */
|
|
|
|
#define USER_FLAG_SANITIZE_PAYLOAD \
|
|
|
|
(1 << 3) /* The user require a deep RESTORE \
|
|
|
|
* payload sanitization. */
|
|
|
|
#define USER_FLAG_SANITIZE_PAYLOAD_SKIP \
|
|
|
|
(1 << 4) /* The user should skip the \
|
|
|
|
* deep sanitization of RESTORE \
|
|
|
|
* payload. */
|
|
|
|
|
|
|
|
#define SELECTOR_FLAG_ROOT \
|
|
|
|
(1 << 0) /* This is the root user permission \
|
|
|
|
* selector. */
|
|
|
|
#define SELECTOR_FLAG_ALLKEYS (1 << 1) /* The user can mention any key. */
|
|
|
|
#define SELECTOR_FLAG_ALLCOMMANDS (1 << 2) /* The user can run all commands. */
|
|
|
|
#define SELECTOR_FLAG_ALLCHANNELS \
|
|
|
|
(1 << 3) /* The user can mention any Pub/Sub \
|
|
|
|
channel. */
|
2022-01-20 13:05:27 -08:00
|
|
|
|
2020-05-04 14:11:43 +02:00
|
|
|
typedef struct {
|
2024-05-22 23:24:12 -07:00
|
|
|
sds name; /* The username as an SDS string. */
|
|
|
|
uint32_t flags; /* See USER_FLAG_* */
|
|
|
|
list *passwords; /* A list of SDS valid passwords for this user. */
|
|
|
|
list *selectors; /* A list of selectors this user validates commands
|
|
|
|
against. This list will always contain at least
|
|
|
|
one selector for backwards compatibility. */
|
2022-09-22 16:29:00 +03:00
|
|
|
robj *acl_string; /* cached string represent of ACLs */
|
2019-01-10 16:33:48 +01:00
|
|
|
} user;
|
|
|
|
|
2013-01-17 01:00:20 +08:00
|
|
|
/* With multiplexing we need to take per-client state.
|
2013-11-20 14:14:27 +08:00
|
|
|
* Clients are taken in a linked list. */
|
2019-10-17 18:10:50 +02:00
|
|
|
|
2024-05-22 23:24:12 -07:00
|
|
|
#define CLIENT_ID_AOF \
|
|
|
|
(UINT64_MAX) /* Reserved ID for the AOF client. If you \
|
|
|
|
need more reserved IDs use UINT64_MAX-1, \
|
|
|
|
-2, ... and so forth. */
|
2019-10-17 18:10:50 +02:00
|
|
|
|
2023-03-16 14:50:32 +08:00
|
|
|
/* Replication backlog is not a separate memory, it just is one consumer of
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
* the global replication buffer. This structure records the reference of
|
|
|
|
* replication buffers. Since the replication buffer block list may be very long,
|
|
|
|
* it would cost much time to search replication offset on partial resync, so
|
|
|
|
* we use one rax tree to index some blocks every REPL_BACKLOG_INDEX_PER_BLOCKS
|
|
|
|
* to make searching offset from replication buffer blocks list faster. */
|
|
|
|
typedef struct replBacklog {
|
|
|
|
listNode *ref_repl_buf_node; /* Referenced node of replication buffer blocks,
|
|
|
|
* see the definition of replBufBlock. */
|
|
|
|
size_t unindexed_count; /* The count from last creating index block. */
|
2022-04-25 22:59:39 +08:00
|
|
|
rax *blocks_index; /* The index of recorded blocks of replication
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
* buffer for quickly searching replication
|
|
|
|
* offset on partial resynchronization. */
|
|
|
|
long long histlen; /* Backlog actual data length */
|
2024-06-07 14:21:33 -07:00
|
|
|
long long offset; /* Replication "primary offset" of first
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
* byte in the replication backlog buffer.*/
|
|
|
|
} replBacklog;
|
|
|
|
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
typedef struct replDataBuf {
|
|
|
|
list *blocks; /* List of replDataBufBlock */
|
|
|
|
size_t len; /* Number of bytes stored in all blocks */
|
|
|
|
size_t peak;
|
|
|
|
} replDataBuf;
|
|
|
|
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
typedef struct {
|
|
|
|
list *clients;
|
|
|
|
size_t mem_usage_sum;
|
|
|
|
} clientMemUsageBucket;
|
|
|
|
|
Add reply_schema to command json files (internal for now) (#10273)
Work in progress towards implementing a reply schema as part of COMMAND DOCS, see #9845
Since ironing the details of the reply schema of each and every command can take a long time, we
would like to merge this PR when the infrastructure is ready, and let this mature in the unstable branch.
Meanwhile the changes of this PR are internal, they are part of the repo, but do not affect the produced build.
### Background
In #9656 we add a lot of information about Redis commands, but we are missing information about the replies
### Motivation
1. Documentation. This is the primary goal.
2. It should be possible, based on the output of COMMAND, to be able to generate client code in typed
languages. In order to do that, we need Redis to tell us, in detail, what each reply looks like.
3. We would like to build a fuzzer that verifies the reply structure (for now we use the existing
testsuite, see the "Testing" section)
### Schema
The idea is to supply some sort of schema for the various replies of each command.
The schema will describe the conceptual structure of the reply (for generated clients), as defined in RESP3.
Note that the reply structure itself may change, depending on the arguments (e.g. `XINFO STREAM`, with
and without the `FULL` modifier)
We decided to use the standard json-schema (see https://json-schema.org/) as the reply-schema.
Example for `BZPOPMIN`:
```
"reply_schema": {
"oneOf": [
{
"description": "Timeout reached and no elements were popped.",
"type": "null"
},
{
"description": "The keyname, popped member, and its score.",
"type": "array",
"minItems": 3,
"maxItems": 3,
"items": [
{
"description": "Keyname",
"type": "string"
},
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
]
}
```
#### Notes
1. It is ok that some commands' reply structure depends on the arguments and it's the caller's responsibility
to know which is the relevant one. this comes after looking at other request-reply systems like OpenAPI,
where the reply schema can also be oneOf and the caller is responsible to know which schema is the relevant one.
2. The reply schemas will describe RESP3 replies only. even though RESP3 is structured, we want to use reply
schema for documentation (and possibly to create a fuzzer that validates the replies)
3. For documentation, the description field will include an explanation of the scenario in which the reply is sent,
including any relation to arguments. for example, for `ZRANGE`'s two schemas we will need to state that one
is with `WITHSCORES` and the other is without.
4. For documentation, there will be another optional field "notes" in which we will add a short description of
the representation in RESP2, in case it's not trivial (RESP3's `ZRANGE`'s nested array vs. RESP2's flat
array, for example)
Given the above:
1. We can generate the "return" section of all commands in [redis-doc](https://redis.io/commands/)
(given that "description" and "notes" are comprehensive enough)
2. We can generate a client in a strongly typed language (but the return type could be a conceptual
`union` and the caller needs to know which schema is relevant). see the section below for RESP2 support.
3. We can create a fuzzer for RESP3.
### Limitations (because we are using the standard json-schema)
The problem is that Redis' replies are more diverse than what the json format allows. This means that,
when we convert the reply to a json (in order to validate the schema against it), we lose information (see
the "Testing" section below).
The other option would have been to extend the standard json-schema (and json format) to include stuff
like sets, bulk-strings, error-string, etc. but that would mean also extending the schema-validator - and that
seemed like too much work, so we decided to compromise.
Examples:
1. We cannot tell the difference between an "array" and a "set"
2. We cannot tell the difference between simple-string and bulk-string
3. we cannot verify true uniqueness of items in commands like ZRANGE: json-schema doesn't cover the
case of two identical members with different scores (e.g. `[["m1",6],["m1",7]]`) because `uniqueItems`
compares (member,score) tuples and not just the member name.
### Testing
This commit includes some changes inside Redis in order to verify the schemas (existing and future ones)
are indeed correct (i.e. describe the actual response of Redis).
To do that, we added a debugging feature to Redis that causes it to produce a log of all the commands
it executed and their replies.
For that, Redis needs to be compiled with `-DLOG_REQ_RES` and run with
`--reg-res-logfile <file> --client-default-resp 3` (the testsuite already does that if you run it with
`--log-req-res --force-resp3`)
You should run the testsuite with the above args (and `--dont-clean`) in order to make Redis generate
`.reqres` files (same dir as the `stdout` files) which contain request-response pairs.
These files are later on processed by `./utils/req-res-log-validator.py` which does:
1. Goes over req-res files, generated by redis-servers, spawned by the testsuite (see logreqres.c)
2. For each request-response pair, it validates the response against the request's reply_schema
(obtained from the extended COMMAND DOCS)
5. In order to get good coverage of the Redis commands, and all their different replies, we chose to use
the existing redis test suite, rather than attempt to write a fuzzer.
#### Notes about RESP2
1. We will not be able to use the testing tool to verify RESP2 replies (we are ok with that, it's time to
accept RESP3 as the future RESP)
2. Since the majority of the test suite is using RESP2, and we want the server to reply with RESP3
so that we can validate it, we will need to know how to convert the actual reply to the one expected.
- number and boolean are always strings in RESP2 so the conversion is easy
- objects (maps) are always a flat array in RESP2
- others (nested array in RESP3's `ZRANGE` and others) will need some special per-command
handling (so the client will not be totally auto-generated)
Example for ZRANGE:
```
"reply_schema": {
"anyOf": [
{
"description": "A list of member elements",
"type": "array",
"uniqueItems": true,
"items": {
"type": "string"
}
},
{
"description": "Members and their scores. Returned in case `WITHSCORES` was used.",
"notes": "In RESP2 this is returned as a flat array",
"type": "array",
"uniqueItems": true,
"items": {
"type": "array",
"minItems": 2,
"maxItems": 2,
"items": [
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
}
]
}
```
### Other changes
1. Some tests that behave differently depending on the RESP are now being tested for both RESP,
regardless of the special log-req-res mode ("Pub/Sub PING" for example)
2. Update the history field of CLIENT LIST
3. Added basic tests for commands that were not covered at all by the testsuite
### TODO
- [x] (maybe a different PR) add a "condition" field to anyOf/oneOf schemas that refers to args. e.g.
when `SET` return NULL, the condition is `arguments.get||arguments.condition`, for `OK` the condition
is `!arguments.get`, and for `string` the condition is `arguments.get` - https://github.com/redis/redis/issues/11896
- [x] (maybe a different PR) also run `runtest-cluster` in the req-res logging mode
- [x] add the new tests to GH actions (i.e. compile with `-DLOG_REQ_RES`, run the tests, and run the validator)
- [x] (maybe a different PR) figure out a way to warn about (sub)schemas that are uncovered by the output
of the tests - https://github.com/redis/redis/issues/11897
- [x] (probably a separate PR) add all missing schemas
- [x] check why "SDOWN is triggered by misconfigured instance replying with errors" fails with --log-req-res
- [x] move the response transformers to their own file (run both regular, cluster, and sentinel tests - need to
fight with the tcl including mechanism a bit)
- [x] issue: module API - https://github.com/redis/redis/issues/11898
- [x] (probably a separate PR): improve schemas: add `required` to `object`s - https://github.com/redis/redis/issues/11899
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Hanna Fadida <hanna.fadida@redislabs.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Shaya Potter <shaya@redislabs.com>
2023-03-11 09:14:16 +01:00
|
|
|
#ifdef LOG_REQ_RES
|
|
|
|
/* Structure used to log client's requests and their
|
|
|
|
* responses (see logreqres.c) */
|
|
|
|
typedef struct {
|
|
|
|
/* General */
|
|
|
|
int argv_logged; /* 1 if the command was logged */
|
|
|
|
/* Vars for log buffer */
|
|
|
|
unsigned char *buf; /* Buffer holding the data (request and response) */
|
|
|
|
size_t used;
|
|
|
|
size_t capacity;
|
|
|
|
/* Vars for offsets within the client's reply */
|
|
|
|
struct {
|
|
|
|
/* General */
|
|
|
|
int saved; /* 1 if we already saved the offset (first time we call addReply*) */
|
|
|
|
/* Offset within the static reply buffer */
|
|
|
|
int bufpos;
|
|
|
|
/* Offset within the reply block list */
|
|
|
|
struct {
|
|
|
|
int index;
|
|
|
|
size_t used;
|
|
|
|
} last_node;
|
|
|
|
} offset;
|
|
|
|
} clientReqResInfo;
|
|
|
|
#endif
|
|
|
|
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
typedef enum {
|
|
|
|
CLIENT_IDLE = 0, /* Initial state: client is idle. */
|
|
|
|
CLIENT_PENDING_IO = 1, /* Main-thread sets this state when client is sent to IO-thread for read/write. */
|
|
|
|
CLIENT_COMPLETED_IO = 2 /* IO-thread sets this state after completing IO operation. */
|
|
|
|
} clientIOState;
|
|
|
|
|
2024-07-01 02:33:10 +08:00
|
|
|
typedef struct ClientFlags {
|
2024-07-01 13:45:14 -07:00
|
|
|
uint64_t primary : 1; /* This client is a primary */
|
|
|
|
uint64_t replica : 1; /* This client is a replica */
|
|
|
|
uint64_t monitor : 1; /* This client is a replica monitor, see MONITOR */
|
|
|
|
uint64_t multi : 1; /* This client is in a MULTI context */
|
|
|
|
uint64_t blocked : 1; /* The client is waiting in a blocking operation */
|
|
|
|
uint64_t dirty_cas : 1; /* Watched keys modified. EXEC will fail. */
|
|
|
|
uint64_t close_after_reply : 1; /* Close after writing entire reply. */
|
|
|
|
uint64_t unblocked : 1; /* This client was unblocked and is stored in server.unblocked_clients */
|
|
|
|
uint64_t script : 1; /* This is a non connected client used by Lua */
|
|
|
|
uint64_t asking : 1; /* Client issued the ASKING command */
|
|
|
|
uint64_t close_asap : 1; /* Close this client ASAP */
|
|
|
|
uint64_t unix_socket : 1; /* Client connected via Unix domain socket */
|
|
|
|
uint64_t dirty_exec : 1; /* EXEC will fail for errors while queueing */
|
|
|
|
uint64_t primary_force_reply : 1; /* Queue replies even if is primary */
|
|
|
|
uint64_t force_aof : 1; /* Force AOF propagation of current cmd. */
|
|
|
|
uint64_t force_repl : 1; /* Force replication of current cmd. */
|
|
|
|
uint64_t pre_psync : 1; /* Instance don't understand PSYNC. */
|
|
|
|
uint64_t readonly : 1; /* Cluster client is in read-only state. */
|
|
|
|
uint64_t pubsub : 1; /* Client is in Pub/Sub mode. */
|
|
|
|
uint64_t prevent_aof_prop : 1; /* Don't propagate to AOF. */
|
|
|
|
uint64_t prevent_repl_prop : 1; /* Don't propagate to replicas. */
|
|
|
|
uint64_t prevent_prop : 1; /* Don't propagate to AOF or replicas. */
|
|
|
|
uint64_t pending_write : 1; /* Client has output to send but a write handler is yet not installed. */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
uint64_t pending_read : 1; /* Client has output to send but a write handler is yet not installed. */
|
2024-07-01 13:45:14 -07:00
|
|
|
uint64_t reply_off : 1; /* Don't send replies to client. */
|
|
|
|
uint64_t reply_skip_next : 1; /* Set CLIENT_REPLY_SKIP for next cmd */
|
|
|
|
uint64_t reply_skip : 1; /* Don't send just this reply. */
|
|
|
|
uint64_t lua_debug : 1; /* Run EVAL in debug mode. */
|
|
|
|
uint64_t lua_debug_sync : 1; /* EVAL debugging without fork() */
|
|
|
|
uint64_t module : 1; /* Non connected client used by some module. */
|
|
|
|
uint64_t protected : 1; /* Client should not be freed for now. */
|
|
|
|
uint64_t executing_command : 1; /* Indicates that the client is currently in the process of handling a command. */
|
|
|
|
uint64_t pending_command : 1; /* Indicates the client has a fully parsed command ready for execution. */
|
|
|
|
uint64_t tracking : 1; /* Client enabled keys tracking in order to perform client side caching. */
|
|
|
|
uint64_t tracking_broken_redir : 1; /* Target client is invalid. */
|
|
|
|
uint64_t tracking_bcast : 1; /* Tracking in BCAST mode. */
|
|
|
|
uint64_t tracking_optin : 1; /* Tracking in opt-in mode. */
|
|
|
|
uint64_t tracking_optout : 1; /* Tracking in opt-out mode. */
|
|
|
|
uint64_t tracking_caching : 1; /* CACHING yes/no was given, depending on optin/optout mode. */
|
|
|
|
uint64_t tracking_noloop : 1; /* Don't send invalidation messages about writes performed by myself. */
|
|
|
|
uint64_t in_to_table : 1; /* This client is in the timeout table. */
|
|
|
|
uint64_t protocol_error : 1; /* Protocol error chatting with it. */
|
|
|
|
uint64_t close_after_command : 1; /* Close after executing commands and writing entire reply. */
|
|
|
|
uint64_t deny_blocking : 1; /* Indicate that the client should not be blocked. */
|
|
|
|
uint64_t repl_rdbonly : 1; /* This client is a replica that only wants RDB without replication buffer. */
|
|
|
|
uint64_t no_evict : 1; /* This client is protected against client memory eviction. */
|
|
|
|
uint64_t allow_oom : 1; /* Client used by RM_Call is allowed to fully execute scripts even when in OOM */
|
|
|
|
uint64_t no_touch : 1; /* This client will not touch LFU/LRU stats. */
|
|
|
|
uint64_t pushing : 1; /* This client is pushing notifications. */
|
|
|
|
uint64_t module_auth_has_result : 1; /* Indicates a client in the middle of module based auth had been authenticated
|
|
|
|
from the Module. */
|
|
|
|
uint64_t module_prevent_aof_prop : 1; /* Module client do not want to propagate to AOF */
|
|
|
|
uint64_t module_prevent_repl_prop : 1; /* Module client do not want to propagate to replica */
|
|
|
|
uint64_t reprocessing_command : 1; /* The client is re-processing the command. */
|
|
|
|
uint64_t replication_done : 1; /* Indicate that replication has been done on the client */
|
|
|
|
uint64_t authenticated : 1; /* Indicate a client has successfully authenticated */
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
uint64_t
|
|
|
|
protected_rdb_channel : 1; /* Dual channel replication sync: Protects the RDB client from premature \
|
|
|
|
* release during full sync. This flag is used to ensure that the RDB client, which \
|
|
|
|
* references the first replication data block required by the replica, is not \
|
|
|
|
* released prematurely. Protecting the client is crucial for prevention of \
|
|
|
|
* synchronization failures: \
|
|
|
|
* If the RDB client is released before the replica initiates PSYNC, the primary \
|
|
|
|
* will reduce the reference count (o->refcount) of the block needed by the replica.
|
|
|
|
* \
|
|
|
|
* This could potentially lead to the removal of the required data block, resulting \
|
|
|
|
* in synchronization failures. Such failures could occur even in scenarios where \
|
|
|
|
* the replica only needs an additional 4KB beyond the minimum size of the
|
|
|
|
* repl_backlog.
|
|
|
|
* By using this flag, we ensure that the RDB client remains intact until the replica
|
|
|
|
* \ has successfully initiated PSYNC. */
|
|
|
|
uint64_t repl_rdb_channel : 1; /* Dual channel replication sync: track a connection which is used for rdb snapshot */
|
|
|
|
uint64_t reserved : 7; /* Reserved for future use */
|
2024-07-01 02:33:10 +08:00
|
|
|
} ClientFlags;
|
|
|
|
|
2015-07-26 15:20:46 +02:00
|
|
|
typedef struct client {
|
2024-07-01 02:33:10 +08:00
|
|
|
uint64_t id; /* Client incremental unique ID. */
|
|
|
|
union {
|
|
|
|
uint64_t raw_flag;
|
|
|
|
struct ClientFlags flag;
|
|
|
|
};
|
2019-09-12 10:56:54 +03:00
|
|
|
connection *conn;
|
2024-05-22 23:24:12 -07:00
|
|
|
int resp; /* RESP protocol version. Can be 2 or 3. */
|
2024-06-27 19:00:45 +08:00
|
|
|
uint32_t capa; /* Client capabilities: CLIENT_CAPA* macros. */
|
2024-05-22 23:24:12 -07:00
|
|
|
serverDb *db; /* Pointer to currently SELECTed DB. */
|
|
|
|
robj *name; /* As set by CLIENT SETNAME. */
|
|
|
|
robj *lib_name; /* The client library name as set by CLIENT SETINFO. */
|
|
|
|
robj *lib_ver; /* The client library version as set by CLIENT SETINFO. */
|
|
|
|
sds querybuf; /* Buffer we use to accumulate client queries. */
|
|
|
|
size_t qb_pos; /* The position we have read in querybuf. */
|
|
|
|
size_t querybuf_peak; /* Recent (100ms or more) peak of querybuf size. */
|
|
|
|
int argc; /* Num of arguments of current command. */
|
|
|
|
robj **argv; /* Arguments of current command. */
|
|
|
|
int argv_len; /* Size of argv array (may be more than argc) */
|
|
|
|
int original_argc; /* Num of arguments of original command if arguments were rewritten. */
|
|
|
|
robj **original_argv; /* Arguments of original command if arguments were rewritten. */
|
|
|
|
size_t argv_len_sum; /* Sum of lengths of objects in argv list. */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
volatile uint8_t io_read_state; /* Indicate the IO read state of the client */
|
|
|
|
volatile uint8_t io_write_state; /* Indicate the IO write state of the client */
|
|
|
|
uint8_t cur_tid; /* ID of IO thread currently performing IO for this client */
|
|
|
|
int nread; /* Number of bytes of the last read. */
|
|
|
|
int nwritten; /* Number of bytes of the last write. */
|
|
|
|
int read_flags; /* Client Read flags - used to communicate the client read state. */
|
|
|
|
uint16_t write_flags; /* Client Write flags - used to communicate the client write state. */
|
2024-05-22 23:24:12 -07:00
|
|
|
struct serverCommand *cmd, *lastcmd; /* Last command executed. */
|
|
|
|
struct serverCommand *realcmd; /* The original command that was executed by the client,
|
|
|
|
Used to update error stats in case the c->cmd was modified
|
|
|
|
during the command invocation (like on GEOADD for example). */
|
Io thread work offload (#763)
### IO-Threads Work Offloading
This PR is the 2nd of 3 PRs intended to achieve the goal of 1M requests
per second.
(1st PR: https://github.com/valkey-io/valkey/pull/758)
This PR offloads additional work to the I/O threads, beyond the current
read-parse/write operations, to better utilize the I/O threads and
reduce the load on the main thread.
It contains the following 3 commits:
### Poll Offload
Currently, the main thread is responsible for executing the poll-wait
system call, while the IO threads wait for tasks from the main thread.
The poll-wait operation is expensive and can consume up to 30% of the
main thread's time. We could have let the IO threads do the poll-wait by
themselves, with each thread listening to some of the clients and
notifying the main thread when a client's command is ready to execute.
However, the current approach, where the main thread listens for events
from the network, has several benefits. The main thread remains in
charge, allowing it to know the state of each client
(idle/read/write/close) at any given time. Additionally, it makes the
threads flexible, enabling us to drain an IO thread's job queue and stop
a thread when the load is light without modifying the event loop and
moving its clients to a different IO thread. Furthermore, with this
approach, the IO threads don't need to wait for both messages from the
network and from the main thread instead, the threads wait only for
tasks from the main thread.
To enjoy the benefits of both the main thread remaining in charge and
the poll being offloaded, we propose offloading the poll-wait as a
single-time, non-blocking job to one of the IO threads. The IO thread
will perform a poll-wait non-blocking call while the main thread
processes the client commands. Later, in `aeProcessEvents`, instead of
sleeping on the poll, we check for the IO thread's poll-wait results.
The poll-wait will be offloaded in `beforeSleep` only when there are
ready events for the main thread to process. If no events are pending,
the main thread will revert to the current behavior and sleep on the
poll by itself.
**Implementation Details**
A new call back `custompoll` was added to the `aeEventLoop` when not set
to `NULL` the ae will call the `custompoll` callback instead of the
`aeApiPoll`.
When the poll is offloaded we will set the `custompoll` to
`getIOThreadPollResults` and send a poll-job to the thread. the thread
will take a mutex, call a non-blocking (with timeout 0) to `aePoll`
which will populate the fired events array. the IO thread will set the
`server.io_fired_events` to the number of the returning `numevents`,
later the main-thread in `custompoll` will return the
`server.io_fired_events` and will set the `customPoll` back to `NULL`.
To ensure thread safety when accessing server.el, all functions that
modify the eventloop events were wrapped with a mutex to ensure mutual
exclusion when modifying the events.
### Command Lookup Offload
As the IO thread parses the command from the client's Querybuf, it can
perform a command lookup in the commands dictionary, which can consume
up to ~5% of the main-thread runtime.
**Implementation details**
The IO thread will store the looked-up command in the client's new field
`io_parsed_cmd` field. We can't use `c->cmd` for that since we use
`c->cmd `to check if a command was reprocessed or not.
To ensure thread safety when accessing the command dictionary, we make
sure the main thread isn't changing the dictionary while IO threads are
accessing it. This is accomplished by introducing a new flag called
`no_incremental_rehash` for the `dictType` commands. When performing
`dictResize`, we will rehash the entire dictionary in place rather than
deferring the process.
### Free Offload
Since the command arguments are allocated by the I/O thread, it would be
beneficial if they were also freed by the same thread. If the main
thread frees objects allocated by the I/O thread, two issues arise:
1. During the freeing process, the main thread needs to access the SDS
pointed to by the object to get its length.
2. With Jemalloc, each thread manages thread local pool (`tcache`) of
buffers for quick reallocation without accessing the arena. If the main
thread constantly frees objects allocated by other threads, those
threads will have to frequently access the shared arena to obtain new
memory allocations
**Implementation Details**
When freeing the client's argv, we will send the argv array to the
thread that allocated it. The thread will be identified by the client
ID. When freeing an object during `dbOverwrite`, we will offload the
object free as well. We will extend this to offload the free during
`dbDelete` in a future PR, as its effects on defrag/memory evictions
need to be studied.
---------
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-19 05:21:45 +03:00
|
|
|
struct serverCommand *io_parsed_cmd; /* The command that was parsed by the IO thread. */
|
2024-05-22 23:24:12 -07:00
|
|
|
user *user; /* User associated with this connection. If the
|
|
|
|
user is set to NULL the connection can do
|
|
|
|
anything (admin). */
|
|
|
|
int reqtype; /* Request protocol type: PROTO_REQ_* */
|
|
|
|
int multibulklen; /* Number of multi bulk arguments left to read. */
|
|
|
|
long bulklen; /* Length of bulk argument in multi bulk request. */
|
|
|
|
list *reply; /* List of reply objects to send to the client. */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
listNode *io_last_reply_block; /* Last client reply block when sent to IO thread */
|
2024-05-22 23:24:12 -07:00
|
|
|
unsigned long long reply_bytes; /* Tot bytes of objects in reply list. */
|
|
|
|
list *deferred_reply_errors; /* Used for module thread safe contexts. */
|
|
|
|
size_t sentlen; /* Amount of bytes already sent in the current
|
|
|
|
buffer or object being sent. */
|
|
|
|
time_t ctime; /* Client creation time. */
|
2024-06-07 14:21:33 -07:00
|
|
|
long duration; /* Current command duration. Used for measuring latency of blocking/non-blocking cmds */
|
|
|
|
int slot; /* The slot the client is executing against. Set to -1 if no slot is being used */
|
|
|
|
dictEntry *cur_script; /* Cached pointer to the dictEntry of the script being executed. */
|
|
|
|
time_t last_interaction; /* Time of the last interaction, used for timeout */
|
2012-01-23 16:12:37 +01:00
|
|
|
time_t obuf_soft_limit_reached_time;
|
2024-06-07 14:21:33 -07:00
|
|
|
int repl_state; /* Replication state if this is a replica. */
|
|
|
|
int repl_start_cmd_stream_on_ack; /* Install replica write handler on first ACK. */
|
2024-05-22 23:24:12 -07:00
|
|
|
int repldbfd; /* Replication DB file descriptor. */
|
|
|
|
off_t repldboff; /* Replication DB file offset. */
|
|
|
|
off_t repldbsize; /* Replication DB file size. */
|
|
|
|
sds replpreamble; /* Replication DB preamble. */
|
2024-06-07 14:21:33 -07:00
|
|
|
long long read_reploff; /* Read replication offset if this is a primary. */
|
|
|
|
long long reploff; /* Applied replication offset if this is a primary. */
|
2024-05-22 23:24:12 -07:00
|
|
|
long long repl_applied; /* Applied replication data count in querybuf, if this is a replica. */
|
2024-06-07 14:21:33 -07:00
|
|
|
long long repl_ack_off; /* Replication ack offset, if this is a replica. */
|
|
|
|
long long repl_aof_off; /* Replication AOF fsync ack offset, if this is a replica. */
|
|
|
|
long long repl_ack_time; /* Replication ack time, if this is a replica. */
|
2024-05-22 23:24:12 -07:00
|
|
|
long long repl_last_partial_write; /* The last time the server did a partial write from the RDB child pipe to this
|
|
|
|
replica */
|
2024-06-07 14:21:33 -07:00
|
|
|
long long psync_initial_offset; /* FULLRESYNC reply offset other replicas
|
|
|
|
copying this replica output buffer
|
2024-05-22 23:24:12 -07:00
|
|
|
should use. */
|
2024-06-07 14:21:33 -07:00
|
|
|
char replid[CONFIG_RUN_ID_SIZE + 1]; /* primary replication ID (if primary). */
|
|
|
|
int replica_listening_port; /* As configured with: REPLCONF listening-port */
|
|
|
|
char *replica_addr; /* Optionally given by REPLCONF ip-address */
|
2024-05-27 23:03:34 +02:00
|
|
|
int replica_version; /* Version on the form 0xMMmmpp. */
|
2024-06-07 14:21:33 -07:00
|
|
|
short replica_capa; /* Replica capabilities: REPLICA_CAPA_* bitwise OR. */
|
|
|
|
short replica_req; /* Replica requirements: REPLICA_REQ_* */
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
uint64_t associated_rdb_client_id; /* The client id of this replica's rdb connection */
|
|
|
|
time_t rdb_client_disconnect_time; /* Time of the first freeClient call on this client. Used for delaying free. */
|
2024-05-22 23:24:12 -07:00
|
|
|
multiState mstate; /* MULTI/EXEC state */
|
|
|
|
blockingState bstate; /* blocking state */
|
|
|
|
long long woff; /* Last write global replication offset. */
|
|
|
|
list *watched_keys; /* Keys WATCHED for MULTI/EXEC CAS */
|
|
|
|
dict *pubsub_channels; /* channels a client is interested in (SUBSCRIBE) */
|
|
|
|
dict *pubsub_patterns; /* patterns a client is interested in (PSUBSCRIBE) */
|
|
|
|
dict *pubsubshard_channels; /* shard level channels a client is interested in (SSUBSCRIBE) */
|
|
|
|
sds peerid; /* Cached peer ID. */
|
|
|
|
sds sockname; /* Cached connection target address. */
|
|
|
|
listNode *client_list_node; /* list node in client list */
|
|
|
|
void *module_blocked_client; /* Pointer to the ValkeyModuleBlockedClient associated with this
|
|
|
|
* client. This is set in case of module authentication before the
|
|
|
|
* unblocked client is reprocessed to handle reply callbacks. */
|
|
|
|
void *module_auth_ctx; /* Ongoing / attempted module based auth callback's ctx.
|
|
|
|
* This is only tracked within the context of the command attempting
|
|
|
|
* authentication. If not NULL, it means module auth is in progress. */
|
2024-04-05 16:59:55 -07:00
|
|
|
ValkeyModuleUserChangedFunc auth_callback; /* Module callback to execute
|
2024-05-22 23:24:12 -07:00
|
|
|
* when the authenticated user
|
|
|
|
* changes. */
|
|
|
|
void *auth_callback_privdata; /* Private data that is passed when the auth
|
|
|
|
* changed callback is executed. Opaque for
|
|
|
|
* the Server Core. */
|
|
|
|
void *auth_module; /* The module that owns the callback, which is used
|
|
|
|
* to disconnect the client if the module is
|
|
|
|
* unloaded for cleanup. Opaque for the Server Core.*/
|
2010-08-30 14:44:34 +02:00
|
|
|
|
2019-06-29 20:08:41 -04:00
|
|
|
/* If this client is in tracking mode and this field is non zero,
|
2023-03-16 14:50:32 +08:00
|
|
|
* invalidation messages for keys fetched by this client will be sent to
|
2019-06-29 20:08:41 -04:00
|
|
|
* the specified client ID. */
|
|
|
|
uint64_t client_tracking_redirection;
|
2020-02-11 17:26:27 +01:00
|
|
|
rax *client_tracking_prefixes; /* A dictionary of prefixes we are already
|
|
|
|
subscribed to in BCAST mode, in the
|
|
|
|
context of client side caching. */
|
2022-12-06 22:26:56 -08:00
|
|
|
/* In updateClientMemoryUsage() we track the memory usage of
|
2020-04-07 12:07:09 +02:00
|
|
|
* each client and add it to the sum of all the clients of a given type,
|
|
|
|
* however we need to remember what was the old contribution of each
|
2021-06-10 20:39:33 +08:00
|
|
|
* client, and in which category the client was, in order to remove it
|
2020-04-07 12:07:09 +02:00
|
|
|
* before adding it the new value. */
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
size_t last_memory_usage;
|
|
|
|
int last_memory_type;
|
|
|
|
|
|
|
|
listNode *mem_usage_bucket_node;
|
|
|
|
clientMemUsageBucket *mem_usage_bucket;
|
|
|
|
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
listNode *ref_repl_buf_node; /* Referenced node of replication buffer blocks,
|
|
|
|
* see the definition of replBufBlock. */
|
|
|
|
size_t ref_block_pos; /* Access position of referenced buffer block,
|
|
|
|
* i.e. the next offset to send. */
|
|
|
|
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
/* list node in clients_pending_write or in clients_pending_io_write list */
|
2022-09-15 06:39:47 +03:00
|
|
|
listNode clients_pending_write_node;
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
listNode pending_read_list_node; /* list node in clients_pending_io_read list */
|
2010-08-30 14:44:34 +02:00
|
|
|
/* Response buffer */
|
2024-05-22 23:24:12 -07:00
|
|
|
size_t buf_peak; /* Peak used size of buffer in last 5 sec interval. */
|
2022-02-22 11:19:38 +02:00
|
|
|
mstime_t buf_peak_last_reset_time; /* keeps the last time the buffer peak value was reset */
|
2015-02-24 11:31:04 -05:00
|
|
|
int bufpos;
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
size_t io_last_bufpos; /* The client's bufpos at the time it was sent to the IO thread */
|
2021-06-08 18:40:12 +08:00
|
|
|
size_t buf_usable_size; /* Usable size of buffer. */
|
2022-02-22 11:19:38 +02:00
|
|
|
char *buf;
|
Add reply_schema to command json files (internal for now) (#10273)
Work in progress towards implementing a reply schema as part of COMMAND DOCS, see #9845
Since ironing the details of the reply schema of each and every command can take a long time, we
would like to merge this PR when the infrastructure is ready, and let this mature in the unstable branch.
Meanwhile the changes of this PR are internal, they are part of the repo, but do not affect the produced build.
### Background
In #9656 we add a lot of information about Redis commands, but we are missing information about the replies
### Motivation
1. Documentation. This is the primary goal.
2. It should be possible, based on the output of COMMAND, to be able to generate client code in typed
languages. In order to do that, we need Redis to tell us, in detail, what each reply looks like.
3. We would like to build a fuzzer that verifies the reply structure (for now we use the existing
testsuite, see the "Testing" section)
### Schema
The idea is to supply some sort of schema for the various replies of each command.
The schema will describe the conceptual structure of the reply (for generated clients), as defined in RESP3.
Note that the reply structure itself may change, depending on the arguments (e.g. `XINFO STREAM`, with
and without the `FULL` modifier)
We decided to use the standard json-schema (see https://json-schema.org/) as the reply-schema.
Example for `BZPOPMIN`:
```
"reply_schema": {
"oneOf": [
{
"description": "Timeout reached and no elements were popped.",
"type": "null"
},
{
"description": "The keyname, popped member, and its score.",
"type": "array",
"minItems": 3,
"maxItems": 3,
"items": [
{
"description": "Keyname",
"type": "string"
},
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
]
}
```
#### Notes
1. It is ok that some commands' reply structure depends on the arguments and it's the caller's responsibility
to know which is the relevant one. this comes after looking at other request-reply systems like OpenAPI,
where the reply schema can also be oneOf and the caller is responsible to know which schema is the relevant one.
2. The reply schemas will describe RESP3 replies only. even though RESP3 is structured, we want to use reply
schema for documentation (and possibly to create a fuzzer that validates the replies)
3. For documentation, the description field will include an explanation of the scenario in which the reply is sent,
including any relation to arguments. for example, for `ZRANGE`'s two schemas we will need to state that one
is with `WITHSCORES` and the other is without.
4. For documentation, there will be another optional field "notes" in which we will add a short description of
the representation in RESP2, in case it's not trivial (RESP3's `ZRANGE`'s nested array vs. RESP2's flat
array, for example)
Given the above:
1. We can generate the "return" section of all commands in [redis-doc](https://redis.io/commands/)
(given that "description" and "notes" are comprehensive enough)
2. We can generate a client in a strongly typed language (but the return type could be a conceptual
`union` and the caller needs to know which schema is relevant). see the section below for RESP2 support.
3. We can create a fuzzer for RESP3.
### Limitations (because we are using the standard json-schema)
The problem is that Redis' replies are more diverse than what the json format allows. This means that,
when we convert the reply to a json (in order to validate the schema against it), we lose information (see
the "Testing" section below).
The other option would have been to extend the standard json-schema (and json format) to include stuff
like sets, bulk-strings, error-string, etc. but that would mean also extending the schema-validator - and that
seemed like too much work, so we decided to compromise.
Examples:
1. We cannot tell the difference between an "array" and a "set"
2. We cannot tell the difference between simple-string and bulk-string
3. we cannot verify true uniqueness of items in commands like ZRANGE: json-schema doesn't cover the
case of two identical members with different scores (e.g. `[["m1",6],["m1",7]]`) because `uniqueItems`
compares (member,score) tuples and not just the member name.
### Testing
This commit includes some changes inside Redis in order to verify the schemas (existing and future ones)
are indeed correct (i.e. describe the actual response of Redis).
To do that, we added a debugging feature to Redis that causes it to produce a log of all the commands
it executed and their replies.
For that, Redis needs to be compiled with `-DLOG_REQ_RES` and run with
`--reg-res-logfile <file> --client-default-resp 3` (the testsuite already does that if you run it with
`--log-req-res --force-resp3`)
You should run the testsuite with the above args (and `--dont-clean`) in order to make Redis generate
`.reqres` files (same dir as the `stdout` files) which contain request-response pairs.
These files are later on processed by `./utils/req-res-log-validator.py` which does:
1. Goes over req-res files, generated by redis-servers, spawned by the testsuite (see logreqres.c)
2. For each request-response pair, it validates the response against the request's reply_schema
(obtained from the extended COMMAND DOCS)
5. In order to get good coverage of the Redis commands, and all their different replies, we chose to use
the existing redis test suite, rather than attempt to write a fuzzer.
#### Notes about RESP2
1. We will not be able to use the testing tool to verify RESP2 replies (we are ok with that, it's time to
accept RESP3 as the future RESP)
2. Since the majority of the test suite is using RESP2, and we want the server to reply with RESP3
so that we can validate it, we will need to know how to convert the actual reply to the one expected.
- number and boolean are always strings in RESP2 so the conversion is easy
- objects (maps) are always a flat array in RESP2
- others (nested array in RESP3's `ZRANGE` and others) will need some special per-command
handling (so the client will not be totally auto-generated)
Example for ZRANGE:
```
"reply_schema": {
"anyOf": [
{
"description": "A list of member elements",
"type": "array",
"uniqueItems": true,
"items": {
"type": "string"
}
},
{
"description": "Members and their scores. Returned in case `WITHSCORES` was used.",
"notes": "In RESP2 this is returned as a flat array",
"type": "array",
"uniqueItems": true,
"items": {
"type": "array",
"minItems": 2,
"maxItems": 2,
"items": [
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
}
]
}
```
### Other changes
1. Some tests that behave differently depending on the RESP are now being tested for both RESP,
regardless of the special log-req-res mode ("Pub/Sub PING" for example)
2. Update the history field of CLIENT LIST
3. Added basic tests for commands that were not covered at all by the testsuite
### TODO
- [x] (maybe a different PR) add a "condition" field to anyOf/oneOf schemas that refers to args. e.g.
when `SET` return NULL, the condition is `arguments.get||arguments.condition`, for `OK` the condition
is `!arguments.get`, and for `string` the condition is `arguments.get` - https://github.com/redis/redis/issues/11896
- [x] (maybe a different PR) also run `runtest-cluster` in the req-res logging mode
- [x] add the new tests to GH actions (i.e. compile with `-DLOG_REQ_RES`, run the tests, and run the validator)
- [x] (maybe a different PR) figure out a way to warn about (sub)schemas that are uncovered by the output
of the tests - https://github.com/redis/redis/issues/11897
- [x] (probably a separate PR) add all missing schemas
- [x] check why "SDOWN is triggered by misconfigured instance replying with errors" fails with --log-req-res
- [x] move the response transformers to their own file (run both regular, cluster, and sentinel tests - need to
fight with the tcl including mechanism a bit)
- [x] issue: module API - https://github.com/redis/redis/issues/11898
- [x] (probably a separate PR): improve schemas: add `required` to `object`s - https://github.com/redis/redis/issues/11899
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Hanna Fadida <hanna.fadida@redislabs.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Shaya Potter <shaya@redislabs.com>
2023-03-11 09:14:16 +01:00
|
|
|
#ifdef LOG_REQ_RES
|
|
|
|
clientReqResInfo reqres;
|
|
|
|
#endif
|
2024-07-26 19:06:16 -04:00
|
|
|
unsigned long long net_input_bytes; /* Total network input bytes read from this client. */
|
|
|
|
unsigned long long net_input_bytes_curr_cmd; /* Total network input bytes read for the
|
|
|
|
* execution of this client's current command. */
|
|
|
|
unsigned long long net_output_bytes; /* Total network output bytes sent to this client. */
|
|
|
|
unsigned long long commands_processed; /* Total count of commands this client executed. */
|
|
|
|
unsigned long long
|
|
|
|
net_output_bytes_curr_cmd; /* Total network output bytes sent to this client, by the current command. */
|
2015-07-26 15:20:46 +02:00
|
|
|
} client;
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2022-10-07 20:19:34 +03:00
|
|
|
/* ACL information */
|
|
|
|
typedef struct aclInfo {
|
2024-05-22 23:24:12 -07:00
|
|
|
long long user_auth_failures; /* Auth failure counts on user level */
|
|
|
|
long long invalid_cmd_accesses; /* Invalid command accesses that user doesn't have permission to */
|
|
|
|
long long invalid_key_accesses; /* Invalid key accesses that user doesn't have permission to */
|
2022-10-07 20:19:34 +03:00
|
|
|
long long invalid_channel_accesses; /* Invalid channel accesses that user doesn't have permission to */
|
|
|
|
} aclInfo;
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
struct saveparam {
|
|
|
|
time_t seconds;
|
|
|
|
int changes;
|
|
|
|
};
|
|
|
|
|
2016-06-13 09:39:44 +02:00
|
|
|
struct moduleLoadQueueEntry {
|
2016-06-05 10:03:34 +03:00
|
|
|
sds path;
|
|
|
|
int argc;
|
2016-06-13 09:51:06 +02:00
|
|
|
robj **argv;
|
2016-06-05 10:03:34 +03:00
|
|
|
};
|
|
|
|
|
Sentinel: Fix Config Dependency and Rewrite Sequence (#8271)
This commit fixes a well known and an annoying issue in Sentinel mode.
Cause of this issue:
Currently, Redis rewrite process works well in server mode, however in sentinel mode,
the sentinel config has variant semantics for different configurations, in example configuration
https://github.com/redis/redis/blob/unstable/sentinel.conf, we put comments on these.
However the rewrite process only treat the sentinel config as a single option. During rewrite
process, it will mess up with the lines and comments.
Approaches:
In order to solve this issue, we need to differentiate different subconfig options in sentinel separately,
for example, sentinel monitor <master-name> <ip> <redis-port> <quorum>
we can treat it as sentinel monitor option, instead of the sentinel option.
This commit also fixes the dependency issue when putting configurations in sentinel.conf.
For example before this commit,we must put
`sentinel monitor <master-name> <ip> <redis-port> <quorum>` before
`sentinel auth-pass <master-name> <password>` for a single master,
otherwise the server cannot start and will return error. This commit fixes this issue, as long as
the monitoring master was configured, no matter the sequence is, the sentinel can start and run properly.
2021-01-26 02:31:54 -05:00
|
|
|
struct sentinelLoadQueueEntry {
|
|
|
|
int argc;
|
|
|
|
sds *argv;
|
|
|
|
int linenum;
|
|
|
|
sds line;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct sentinelConfig {
|
|
|
|
list *pre_monitor_cfg;
|
|
|
|
list *monitor_cfg;
|
|
|
|
list *post_monitor_cfg;
|
|
|
|
};
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
struct sharedObjectsStruct {
|
2024-05-22 23:24:12 -07:00
|
|
|
robj *ok, *err, *emptybulk, *czero, *cone, *pong, *space, *queued, *null[4], *nullarray[4], *emptymap[4],
|
|
|
|
*emptyset[4], *emptyarray, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr, *outofrangeerr, *noscripterr,
|
2024-06-07 14:21:33 -07:00
|
|
|
*loadingerr, *slowevalerr, *slowscripterr, *slowmoduleerr, *bgsaveerr, *primarydownerr, *roreplicaerr,
|
2024-05-22 23:24:12 -07:00
|
|
|
*execaborterr, *noautherr, *noreplicaserr, *busykeyerr, *oomerr, *plus, *messagebulk, *pmessagebulk,
|
|
|
|
*subscribebulk, *unsubscribebulk, *psubscribebulk, *punsubscribebulk, *del, *unlink, *rpop, *lpop, *lpush,
|
|
|
|
*rpoplpush, *lmove, *blmove, *zpopmin, *zpopmax, *emptyscan, *multi, *exec, *left, *right, *hset, *srem,
|
|
|
|
*xgroup, *xclaim, *script, *replconf, *eval, *persist, *set, *pexpireat, *pexpire, *time, *pxat, *absttl,
|
|
|
|
*retrycount, *force, *justid, *entriesread, *lastid, *ping, *setid, *keepttl, *load, *createconsumer, *getack,
|
|
|
|
*special_asterick, *special_equals, *default_username, *redacted, *ssubscribebulk, *sunsubscribebulk,
|
2024-06-07 20:32:27 -07:00
|
|
|
*smessagebulk, *select[PROTO_SHARED_SELECT_CMDS], *integers[OBJ_SHARED_INTEGERS],
|
|
|
|
*mbulkhdr[OBJ_SHARED_BULKHDR_LEN], /* "*<value>\r\n" */
|
|
|
|
*bulkhdr[OBJ_SHARED_BULKHDR_LEN], /* "$<value>\r\n" */
|
|
|
|
*maphdr[OBJ_SHARED_BULKHDR_LEN], /* "%<value>\r\n" */
|
|
|
|
*sethdr[OBJ_SHARED_BULKHDR_LEN]; /* "~<value>\r\n" */
|
2015-08-04 09:20:55 +02:00
|
|
|
sds minstring, maxstring;
|
2010-06-22 00:07:48 +02:00
|
|
|
};
|
|
|
|
|
2011-04-28 19:00:33 +02:00
|
|
|
/* ZSETs use a specialized version of Skiplists */
|
|
|
|
typedef struct zskiplistNode {
|
2015-08-04 09:20:55 +02:00
|
|
|
sds ele;
|
2011-04-28 19:00:33 +02:00
|
|
|
double score;
|
|
|
|
struct zskiplistNode *backward;
|
|
|
|
struct zskiplistLevel {
|
|
|
|
struct zskiplistNode *forward;
|
2017-12-08 16:09:27 +08:00
|
|
|
unsigned long span;
|
2011-04-28 19:00:33 +02:00
|
|
|
} level[];
|
|
|
|
} zskiplistNode;
|
|
|
|
|
|
|
|
typedef struct zskiplist {
|
|
|
|
struct zskiplistNode *header, *tail;
|
|
|
|
unsigned long length;
|
|
|
|
int level;
|
|
|
|
} zskiplist;
|
|
|
|
|
|
|
|
typedef struct zset {
|
|
|
|
dict *dict;
|
|
|
|
zskiplist *zsl;
|
|
|
|
} zset;
|
|
|
|
|
2012-01-23 16:12:37 +01:00
|
|
|
typedef struct clientBufferLimitsConfig {
|
2012-01-24 10:43:30 +01:00
|
|
|
unsigned long long hard_limit_bytes;
|
|
|
|
unsigned long long soft_limit_bytes;
|
2012-01-23 16:12:37 +01:00
|
|
|
time_t soft_limit_seconds;
|
|
|
|
} clientBufferLimitsConfig;
|
|
|
|
|
2015-07-28 16:58:04 +02:00
|
|
|
extern clientBufferLimitsConfig clientBufferLimitsDefaults[CLIENT_TYPE_OBUF_COUNT];
|
2013-05-13 18:34:18 +02:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* The serverOp structure defines an Operation, that is an instance of
|
2012-02-29 00:46:50 +01:00
|
|
|
* a command with an argument vector, database ID, propagation target
|
2015-07-27 09:41:48 +02:00
|
|
|
* (PROPAGATE_*), and command pointer.
|
2012-02-29 00:46:50 +01:00
|
|
|
*
|
|
|
|
* Currently only used to additionally propagate more commands to AOF/Replication
|
|
|
|
* after the propagation of the executed command. */
|
2024-04-04 01:30:30 +07:00
|
|
|
typedef struct serverOp {
|
2012-02-28 18:03:08 +01:00
|
|
|
robj **argv;
|
|
|
|
int argc, dbid, target;
|
2024-04-04 01:30:30 +07:00
|
|
|
} serverOp;
|
2012-02-29 00:46:50 +01:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* Defines an array of Operations. There is an API to add to this
|
Squash merging 125 typo/grammar/comment/doc PRs (#7773)
List of squashed commits or PRs
===============================
commit 66801ea
Author: hwware <wen.hui.ware@gmail.com>
Date: Mon Jan 13 00:54:31 2020 -0500
typo fix in acl.c
commit 46f55db
Author: Itamar Haber <itamar@redislabs.com>
Date: Sun Sep 6 18:24:11 2020 +0300
Updates a couple of comments
Specifically:
* RM_AutoMemory completed instead of pointing to docs
* Updated link to custom type doc
commit 61a2aa0
Author: xindoo <xindoo@qq.com>
Date: Tue Sep 1 19:24:59 2020 +0800
Correct errors in code comments
commit a5871d1
Author: yz1509 <pro-756@qq.com>
Date: Tue Sep 1 18:36:06 2020 +0800
fix typos in module.c
commit 41eede7
Author: bookug <bookug@qq.com>
Date: Sat Aug 15 01:11:33 2020 +0800
docs: fix typos in comments
commit c303c84
Author: lazy-snail <ws.niu@outlook.com>
Date: Fri Aug 7 11:15:44 2020 +0800
fix spelling in redis.conf
commit 1eb76bf
Author: zhujian <zhujianxyz@gmail.com>
Date: Thu Aug 6 15:22:10 2020 +0800
add a missing 'n' in comment
commit 1530ec2
Author: Daniel Dai <764122422@qq.com>
Date: Mon Jul 27 00:46:35 2020 -0400
fix spelling in tracking.c
commit e517b31
Author: Hunter-Chen <huntcool001@gmail.com>
Date: Fri Jul 17 22:33:32 2020 +0800
Update redis.conf
Co-authored-by: Itamar Haber <itamar@redislabs.com>
commit c300eff
Author: Hunter-Chen <huntcool001@gmail.com>
Date: Fri Jul 17 22:33:23 2020 +0800
Update redis.conf
Co-authored-by: Itamar Haber <itamar@redislabs.com>
commit 4c058a8
Author: 陈浩鹏 <chenhaopeng@heytea.com>
Date: Thu Jun 25 19:00:56 2020 +0800
Grammar fix and clarification
commit 5fcaa81
Author: bodong.ybd <bodong.ybd@alibaba-inc.com>
Date: Fri Jun 19 10:09:00 2020 +0800
Fix typos
commit 4caca9a
Author: Pruthvi P <pruthvi@ixigo.com>
Date: Fri May 22 00:33:22 2020 +0530
Fix typo eviciton => eviction
commit b2a25f6
Author: Brad Dunbar <dunbarb2@gmail.com>
Date: Sun May 17 12:39:59 2020 -0400
Fix a typo.
commit 12842ae
Author: hwware <wen.hui.ware@gmail.com>
Date: Sun May 3 17:16:59 2020 -0400
fix spelling in redis conf
commit ddba07c
Author: Chris Lamb <chris@chris-lamb.co.uk>
Date: Sat May 2 23:25:34 2020 +0100
Correct a "conflicts" spelling error.
commit 8fc7bf2
Author: Nao YONASHIRO <yonashiro@r.recruit.co.jp>
Date: Thu Apr 30 10:25:27 2020 +0900
docs: fix EXPIRE_FAST_CYCLE_DURATION to ACTIVE_EXPIRE_CYCLE_FAST_DURATION
commit 9b2b67a
Author: Brad Dunbar <dunbarb2@gmail.com>
Date: Fri Apr 24 11:46:22 2020 -0400
Fix a typo.
commit 0746f10
Author: devilinrust <63737265+devilinrust@users.noreply.github.com>
Date: Thu Apr 16 00:17:53 2020 +0200
Fix typos in server.c
commit 92b588d
Author: benjessop12 <56115861+benjessop12@users.noreply.github.com>
Date: Mon Apr 13 13:43:55 2020 +0100
Fix spelling mistake in lazyfree.c
commit 1da37aa
Merge: 2d4ba28 af347a8
Author: hwware <wen.hui.ware@gmail.com>
Date: Thu Mar 5 22:41:31 2020 -0500
Merge remote-tracking branch 'upstream/unstable' into expiretypofix
commit 2d4ba28
Author: hwware <wen.hui.ware@gmail.com>
Date: Mon Mar 2 00:09:40 2020 -0500
fix typo in expire.c
commit 1a746f7
Author: SennoYuki <minakami1yuki@gmail.com>
Date: Thu Feb 27 16:54:32 2020 +0800
fix typo
commit 8599b1a
Author: dongheejeong <donghee950403@gmail.com>
Date: Sun Feb 16 20:31:43 2020 +0000
Fix typo in server.c
commit f38d4e8
Author: hwware <wen.hui.ware@gmail.com>
Date: Sun Feb 2 22:58:38 2020 -0500
fix typo in evict.c
commit fe143fc
Author: Leo Murillo <leonardo.murillo@gmail.com>
Date: Sun Feb 2 01:57:22 2020 -0600
Fix a few typos in redis.conf
commit 1ab4d21
Author: viraja1 <anchan.viraj@gmail.com>
Date: Fri Dec 27 17:15:58 2019 +0530
Fix typo in Latency API docstring
commit ca1f70e
Author: gosth <danxuedexing@qq.com>
Date: Wed Dec 18 15:18:02 2019 +0800
fix typo in sort.c
commit a57c06b
Author: ZYunH <zyunhjob@163.com>
Date: Mon Dec 16 22:28:46 2019 +0800
fix-zset-typo
commit b8c92b5
Author: git-hulk <hulk.website@gmail.com>
Date: Mon Dec 16 15:51:42 2019 +0800
FIX: typo in cluster.c, onformation->information
commit 9dd981c
Author: wujm2007 <jim.wujm@gmail.com>
Date: Mon Dec 16 09:37:52 2019 +0800
Fix typo
commit e132d7a
Author: Sebastien Williams-Wynn <s.williamswynn.mail@gmail.com>
Date: Fri Nov 15 00:14:07 2019 +0000
Minor typo change
commit 47f44d5
Author: happynote3966 <01ssrmikururudevice01@gmail.com>
Date: Mon Nov 11 22:08:48 2019 +0900
fix comment typo in redis-cli.c
commit b8bdb0d
Author: fulei <fulei@kuaishou.com>
Date: Wed Oct 16 18:00:17 2019 +0800
Fix a spelling mistake of comments in defragDictBucketCallback
commit 0def46a
Author: fulei <fulei@kuaishou.com>
Date: Wed Oct 16 13:09:27 2019 +0800
fix some spelling mistakes of comments in defrag.c
commit f3596fd
Author: Phil Rajchgot <tophil@outlook.com>
Date: Sun Oct 13 02:02:32 2019 -0400
Typo and grammar fixes
Redis and its documentation are great -- just wanted to submit a few corrections in the spirit of Hacktoberfest. Thanks for all your work on this project. I use it all the time and it works beautifully.
commit 2b928cd
Author: KangZhiDong <worldkzd@gmail.com>
Date: Sun Sep 1 07:03:11 2019 +0800
fix typos
commit 33aea14
Author: Axlgrep <axlgrep@gmail.com>
Date: Tue Aug 27 11:02:18 2019 +0800
Fixed eviction spelling issues
commit e282a80
Author: Simen Flatby <simen@oms.no>
Date: Tue Aug 20 15:25:51 2019 +0200
Update comments to reflect prop name
In the comments the prop is referenced as replica-validity-factor,
but it is really named cluster-replica-validity-factor.
commit 74d1f9a
Author: Jim Green <jimgreen2013@qq.com>
Date: Tue Aug 20 20:00:31 2019 +0800
fix comment error, the code is ok
commit eea1407
Author: Liao Tonglang <liaotonglang@gmail.com>
Date: Fri May 31 10:16:18 2019 +0800
typo fix
fix cna't to can't
commit 0da553c
Author: KAWACHI Takashi <tkawachi@gmail.com>
Date: Wed Jul 17 00:38:16 2019 +0900
Fix typo
commit 7fc8fb6
Author: Michael Prokop <mika@grml.org>
Date: Tue May 28 17:58:42 2019 +0200
Typo fixes
s/familar/familiar/
s/compatiblity/compatibility/
s/ ot / to /
s/itsef/itself/
commit 5f46c9d
Author: zhumoing <34539422+zhumoing@users.noreply.github.com>
Date: Tue May 21 21:16:50 2019 +0800
typo-fixes
typo-fixes
commit 321dfe1
Author: wxisme <850885154@qq.com>
Date: Sat Mar 16 15:10:55 2019 +0800
typo fix
commit b4fb131
Merge: 267e0e6 3df1eb8
Author: Nikitas Bastas <nikitasbst@gmail.com>
Date: Fri Feb 8 22:55:45 2019 +0200
Merge branch 'unstable' of antirez/redis into unstable
commit 267e0e6
Author: Nikitas Bastas <nikitasbst@gmail.com>
Date: Wed Jan 30 21:26:04 2019 +0200
Minor typo fix
commit 30544e7
Author: inshal96 <39904558+inshal96@users.noreply.github.com>
Date: Fri Jan 4 16:54:50 2019 +0500
remove an extra 'a' in the comments
commit 337969d
Author: BrotherGao <yangdongheng11@gmail.com>
Date: Sat Dec 29 12:37:29 2018 +0800
fix typo in redis.conf
commit 9f4b121
Merge: 423a030 e504583
Author: BrotherGao <yangdongheng@xiaomi.com>
Date: Sat Dec 29 11:41:12 2018 +0800
Merge branch 'unstable' of antirez/redis into unstable
commit 423a030
Merge: 42b02b7 46a51cd
Author: 杨东衡 <yangdongheng@xiaomi.com>
Date: Tue Dec 4 23:56:11 2018 +0800
Merge branch 'unstable' of antirez/redis into unstable
commit 42b02b7
Merge: 68c0e6e b8febe6
Author: Dongheng Yang <yangdongheng11@gmail.com>
Date: Sun Oct 28 15:54:23 2018 +0800
Merge pull request #1 from antirez/unstable
update local data
commit 714b589
Author: Christian <crifei93@gmail.com>
Date: Fri Dec 28 01:17:26 2018 +0100
fix typo "resulution"
commit e23259d
Author: garenchan <1412950785@qq.com>
Date: Wed Dec 26 09:58:35 2018 +0800
fix typo: segfauls -> segfault
commit a9359f8
Author: xjp <jianping_xie@aliyun.com>
Date: Tue Dec 18 17:31:44 2018 +0800
Fixed REDISMODULE_H spell bug
commit a12c3e4
Author: jdiaz <jrd.palacios@gmail.com>
Date: Sat Dec 15 23:39:52 2018 -0600
Fixes hyperloglog hash function comment block description
commit 770eb11
Author: 林上耀 <1210tom@163.com>
Date: Sun Nov 25 17:16:10 2018 +0800
fix typo
commit fd97fbb
Author: Chris Lamb <chris@chris-lamb.co.uk>
Date: Fri Nov 23 17:14:01 2018 +0100
Correct "unsupported" typo.
commit a85522d
Author: Jungnam Lee <jungnam.lee@oracle.com>
Date: Thu Nov 8 23:01:29 2018 +0900
fix typo in test comments
commit ade8007
Author: Arun Kumar <palerdot@users.noreply.github.com>
Date: Tue Oct 23 16:56:35 2018 +0530
Fixed grammatical typo
Fixed typo for word 'dictionary'
commit 869ee39
Author: Hamid Alaei <hamid.a85@gmail.com>
Date: Sun Aug 12 16:40:02 2018 +0430
fix documentations: (ThreadSafeContextStart/Stop -> ThreadSafeContextLock/Unlock), minor typo
commit f89d158
Author: Mayank Jain <mayankjain255@gmail.com>
Date: Tue Jul 31 23:01:21 2018 +0530
Updated README.md with some spelling corrections.
Made correction in spelling of some misspelled words.
commit 892198e
Author: dsomeshwar <someshwar.dhayalan@gmail.com>
Date: Sat Jul 21 23:23:04 2018 +0530
typo fix
commit 8a4d780
Author: Itamar Haber <itamar@redislabs.com>
Date: Mon Apr 30 02:06:52 2018 +0300
Fixes some typos
commit e3acef6
Author: Noah Rosamilia <ivoahivoah@gmail.com>
Date: Sat Mar 3 23:41:21 2018 -0500
Fix typo in /deps/README.md
commit 04442fb
Author: WuYunlong <xzsyeb@126.com>
Date: Sat Mar 3 10:32:42 2018 +0800
Fix typo in readSyncBulkPayload() comment.
commit 9f36880
Author: WuYunlong <xzsyeb@126.com>
Date: Sat Mar 3 10:20:37 2018 +0800
replication.c comment: run_id -> replid.
commit f866b4a
Author: Francesco 'makevoid' Canessa <makevoid@gmail.com>
Date: Thu Feb 22 22:01:56 2018 +0000
fix comment typo in server.c
commit 0ebc69b
Author: 줍 <jubee0124@gmail.com>
Date: Mon Feb 12 16:38:48 2018 +0900
Fix typo in redis.conf
Fix `five behaviors` to `eight behaviors` in [this sentence ](antirez/redis@unstable/redis.conf#L564)
commit b50a620
Author: martinbroadhurst <martinbroadhurst@users.noreply.github.com>
Date: Thu Dec 28 12:07:30 2017 +0000
Fix typo in valgrind.sup
commit 7d8f349
Author: Peter Boughton <peter@sorcerersisle.com>
Date: Mon Nov 27 19:52:19 2017 +0000
Update CONTRIBUTING; refer doc updates to redis-doc repo.
commit 02dec7e
Author: Klauswk <klauswk1@hotmail.com>
Date: Tue Oct 24 16:18:38 2017 -0200
Fix typo in comment
commit e1efbc8
Author: chenshi <baiwfg2@gmail.com>
Date: Tue Oct 3 18:26:30 2017 +0800
Correct two spelling errors of comments
commit 93327d8
Author: spacewander <spacewanderlzx@gmail.com>
Date: Wed Sep 13 16:47:24 2017 +0800
Update the comment for OBJ_ENCODING_EMBSTR_SIZE_LIMIT's value
The value of OBJ_ENCODING_EMBSTR_SIZE_LIMIT is 44 now instead of 39.
commit 63d361f
Author: spacewander <spacewanderlzx@gmail.com>
Date: Tue Sep 12 15:06:42 2017 +0800
Fix <prevlen> related doc in ziplist.c
According to the definition of ZIP_BIG_PREVLEN and other related code,
the guard of single byte <prevlen> should be 254 instead of 255.
commit ebe228d
Author: hanael80 <hanael80@gmail.com>
Date: Tue Aug 15 09:09:40 2017 +0900
Fix typo
commit 6b696e6
Author: Matt Robenolt <matt@ydekproductions.com>
Date: Mon Aug 14 14:50:47 2017 -0700
Fix typo in LATENCY DOCTOR output
commit a2ec6ae
Author: caosiyang <caosiyang@qiyi.com>
Date: Tue Aug 15 14:15:16 2017 +0800
Fix a typo: form => from
commit 3ab7699
Author: caosiyang <caosiyang@qiyi.com>
Date: Thu Aug 10 18:40:33 2017 +0800
Fix a typo: replicationFeedSlavesFromMaster() => replicationFeedSlavesFromMasterStream()
commit 72d43ef
Author: caosiyang <caosiyang@qiyi.com>
Date: Tue Aug 8 15:57:25 2017 +0800
fix a typo: servewr => server
commit 707c958
Author: Bo Cai <charpty@gmail.com>
Date: Wed Jul 26 21:49:42 2017 +0800
redis-cli.c typo: conut -> count.
Signed-off-by: Bo Cai <charpty@gmail.com>
commit b9385b2
Author: JackDrogon <jack.xsuperman@gmail.com>
Date: Fri Jun 30 14:22:31 2017 +0800
Fix some spell problems
commit 20d9230
Author: akosel <aaronjkosel@gmail.com>
Date: Sun Jun 4 19:35:13 2017 -0500
Fix typo
commit b167bfc
Author: Krzysiek Witkowicz <krzysiekwitkowicz@gmail.com>
Date: Mon May 22 21:32:27 2017 +0100
Fix #4008 small typo in comment
commit 2b78ac8
Author: Jake Clarkson <jacobwclarkson@gmail.com>
Date: Wed Apr 26 15:49:50 2017 +0100
Correct typo in tests/unit/hyperloglog.tcl
commit b0f1cdb
Author: Qi Luo <qiluo-msft@users.noreply.github.com>
Date: Wed Apr 19 14:25:18 2017 -0700
Fix typo
commit a90b0f9
Author: charsyam <charsyam@naver.com>
Date: Thu Mar 16 18:19:53 2017 +0900
fix typos
fix typos
fix typos
commit 8430a79
Author: Richard Hart <richardhart92@gmail.com>
Date: Mon Mar 13 22:17:41 2017 -0400
Fixed log message typo in listenToPort.
commit 481a1c2
Author: Vinod Kumar <kumar003vinod@gmail.com>
Date: Sun Jan 15 23:04:51 2017 +0530
src/db.c: Correct "save" -> "safe" typo
commit 586b4d3
Author: wangshaonan <wshn13@gmail.com>
Date: Wed Dec 21 20:28:27 2016 +0800
Fix typo they->the in helloworld.c
commit c1c4b5e
Author: Jenner <hypxm@qq.com>
Date: Mon Dec 19 16:39:46 2016 +0800
typo error
commit 1ee1a3f
Author: tielei <43289893@qq.com>
Date: Mon Jul 18 13:52:25 2016 +0800
fix some comments
commit 11a41fb
Author: Otto Kekäläinen <otto@seravo.fi>
Date: Sun Jul 3 10:23:55 2016 +0100
Fix spelling in documentation and comments
commit 5fb5d82
Author: francischan <f1ancis621@gmail.com>
Date: Tue Jun 28 00:19:33 2016 +0800
Fix outdated comments about redis.c file.
It should now refer to server.c file.
commit 6b254bc
Author: lmatt-bit <lmatt123n@gmail.com>
Date: Thu Apr 21 21:45:58 2016 +0800
Refine the comment of dictRehashMilliseconds func
SLAVECONF->REPLCONF in comment - by andyli029
commit ee9869f
Author: clark.kang <charsyam@naver.com>
Date: Tue Mar 22 11:09:51 2016 +0900
fix typos
commit f7b3b11
Author: Harisankar H <harisankarh@gmail.com>
Date: Wed Mar 9 11:49:42 2016 +0530
Typo correction: "faield" --> "failed"
Typo correction: "faield" --> "failed"
commit 3fd40fc
Author: Itamar Haber <itamar@redislabs.com>
Date: Thu Feb 25 10:31:51 2016 +0200
Fixes a typo in comments
commit 621c160
Author: Prayag Verma <prayag.verma@gmail.com>
Date: Mon Feb 1 12:36:20 2016 +0530
Fix typo in Readme.md
Spelling mistakes -
`eviciton` > `eviction`
`familar` > `familiar`
commit d7d07d6
Author: WonCheol Lee <toctoc21c@gmail.com>
Date: Wed Dec 30 15:11:34 2015 +0900
Typo fixed
commit a4dade7
Author: Felix Bünemann <buenemann@louis.info>
Date: Mon Dec 28 11:02:55 2015 +0100
[ci skip] Improve supervised upstart config docs
This mentions that "expect stop" is required for supervised upstart
to work correctly. See http://upstart.ubuntu.com/cookbook/#expect-stop
for an explanation.
commit d9caba9
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:30:03 2015 +1100
README: Remove trailing whitespace
commit 72d42e5
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:29:32 2015 +1100
README: Fix typo. th => the
commit dd6e957
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:29:20 2015 +1100
README: Fix typo. familar => familiar
commit 3a12b23
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:28:54 2015 +1100
README: Fix typo. eviciton => eviction
commit 2d1d03b
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:21:45 2015 +1100
README: Fix typo. sever => server
commit 3973b06
Author: Itamar Haber <itamar@garantiadata.com>
Date: Sat Dec 19 17:01:20 2015 +0200
Typo fix
commit 4f2e460
Author: Steve Gao <fu@2token.com>
Date: Fri Dec 4 10:22:05 2015 +0800
Update README - fix typos
commit b21667c
Author: binyan <binbin.yan@nokia.com>
Date: Wed Dec 2 22:48:37 2015 +0800
delete redundancy color judge in sdscatcolor
commit 88894c7
Author: binyan <binbin.yan@nokia.com>
Date: Wed Dec 2 22:14:42 2015 +0800
the example output shoule be HelloWorld
commit 2763470
Author: binyan <binbin.yan@nokia.com>
Date: Wed Dec 2 17:41:39 2015 +0800
modify error word keyevente
Signed-off-by: binyan <binbin.yan@nokia.com>
commit 0847b3d
Author: Bruno Martins <bscmartins@gmail.com>
Date: Wed Nov 4 11:37:01 2015 +0000
typo
commit bbb9e9e
Author: dawedawe <dawedawe@gmx.de>
Date: Fri Mar 27 00:46:41 2015 +0100
typo: zimap -> zipmap
commit 5ed297e
Author: Axel Advento <badwolf.bloodseeker.rev@gmail.com>
Date: Tue Mar 3 15:58:29 2015 +0800
Fix 'salve' typos to 'slave'
commit edec9d6
Author: LudwikJaniuk <ludvig.janiuk@gmail.com>
Date: Wed Jun 12 14:12:47 2019 +0200
Update README.md
Co-Authored-By: Qix <Qix-@users.noreply.github.com>
commit 692a7af
Author: LudwikJaniuk <ludvig.janiuk@gmail.com>
Date: Tue May 28 14:32:04 2019 +0200
grammar
commit d962b0a
Author: Nick Frost <nickfrostatx@gmail.com>
Date: Wed Jul 20 15:17:12 2016 -0700
Minor grammar fix
commit 24fff01aaccaf5956973ada8c50ceb1462e211c6 (typos)
Author: Chad Miller <chadm@squareup.com>
Date: Tue Sep 8 13:46:11 2020 -0400
Fix faulty comment about operation of unlink()
commit 3cd5c1f3326c52aa552ada7ec797c6bb16452355
Author: Kevin <kevin.xgr@gmail.com>
Date: Wed Nov 20 00:13:50 2019 +0800
Fix typo in server.c.
From a83af59 Mon Sep 17 00:00:00 2001
From: wuwo <wuwo@wacai.com>
Date: Fri, 17 Mar 2017 20:37:45 +0800
Subject: [PATCH] falure to failure
From c961896 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=B7=A6=E6=87=B6?= <veficos@gmail.com>
Date: Sat, 27 May 2017 15:33:04 +0800
Subject: [PATCH] fix typo
From e600ef2 Mon Sep 17 00:00:00 2001
From: "rui.zou" <rui.zou@yunify.com>
Date: Sat, 30 Sep 2017 12:38:15 +0800
Subject: [PATCH] fix a typo
From c7d07fa Mon Sep 17 00:00:00 2001
From: Alexandre Perrin <alex@kaworu.ch>
Date: Thu, 16 Aug 2018 10:35:31 +0200
Subject: [PATCH] deps README.md typo
From b25cb67 Mon Sep 17 00:00:00 2001
From: Guy Korland <gkorland@gmail.com>
Date: Wed, 26 Sep 2018 10:55:37 +0300
Subject: [PATCH 1/2] fix typos in header
From ad28ca6 Mon Sep 17 00:00:00 2001
From: Guy Korland <gkorland@gmail.com>
Date: Wed, 26 Sep 2018 11:02:36 +0300
Subject: [PATCH 2/2] fix typos
commit 34924cdedd8552466fc22c1168d49236cb7ee915
Author: Adrian Lynch <adi_ady_ade@hotmail.com>
Date: Sat Apr 4 21:59:15 2015 +0100
Typos fixed
commit fd2a1e7
Author: Jan <jsteemann@users.noreply.github.com>
Date: Sat Oct 27 19:13:01 2018 +0200
Fix typos
Fix typos
commit e14e47c1a234b53b0e103c5f6a1c61481cbcbb02
Author: Andy Lester <andy@petdance.com>
Date: Fri Aug 2 22:30:07 2019 -0500
Fix multiple misspellings of "following"
commit 79b948ce2dac6b453fe80995abbcaac04c213d5a
Author: Andy Lester <andy@petdance.com>
Date: Fri Aug 2 22:24:28 2019 -0500
Fix misspelling of create-cluster
commit 1fffde52666dc99ab35efbd31071a4c008cb5a71
Author: Andy Lester <andy@petdance.com>
Date: Wed Jul 31 17:57:56 2019 -0500
Fix typos
commit 204c9ba9651e9e05fd73936b452b9a30be456cfe
Author: Xiaobo Zhu <xiaobo.zhu@shopee.com>
Date: Tue Aug 13 22:19:25 2019 +0800
fix typos
Squashed commit of the following:
commit 1d9aaf8
Author: danmedani <danmedani@gmail.com>
Date: Sun Aug 2 11:40:26 2015 -0700
README typo fix.
Squashed commit of the following:
commit 32bfa7c
Author: Erik Dubbelboer <erik@dubbelboer.com>
Date: Mon Jul 6 21:15:08 2015 +0200
Fixed grammer
Squashed commit of the following:
commit b24f69c
Author: Sisir Koppaka <sisir.koppaka@gmail.com>
Date: Mon Mar 2 22:38:45 2015 -0500
utils/hashtable/rehashing.c: Fix typos
Squashed commit of the following:
commit 4e04082
Author: Erik Dubbelboer <erik@dubbelboer.com>
Date: Mon Mar 23 08:22:21 2015 +0000
Small config file documentation improvements
Squashed commit of the following:
commit acb8773
Author: ctd1500 <ctd1500@gmail.com>
Date: Fri May 8 01:52:48 2015 -0700
Typo and grammar fixes in readme
commit 2eb75b6
Author: ctd1500 <ctd1500@gmail.com>
Date: Fri May 8 01:36:18 2015 -0700
fixed redis.conf comment
Squashed commit of the following:
commit a8249a2
Author: Masahiko Sawada <sawada.mshk@gmail.com>
Date: Fri Dec 11 11:39:52 2015 +0530
Revise correction of typos.
Squashed commit of the following:
commit 3c02028
Author: zhaojun11 <zhaojun11@jd.com>
Date: Wed Jan 17 19:05:28 2018 +0800
Fix typos include two code typos in cluster.c and latency.c
Squashed commit of the following:
commit 9dba47c
Author: q191201771 <191201771@qq.com>
Date: Sat Jan 4 11:31:04 2020 +0800
fix function listCreate comment in adlist.c
Update src/server.c
commit 2c7c2cb536e78dd211b1ac6f7bda00f0f54faaeb
Author: charpty <charpty@gmail.com>
Date: Tue May 1 23:16:59 2018 +0800
server.c typo: modules system dictionary type comment
Signed-off-by: charpty <charpty@gmail.com>
commit a8395323fb63cb59cb3591cb0f0c8edb7c29a680
Author: Itamar Haber <itamar@redislabs.com>
Date: Sun May 6 00:25:18 2018 +0300
Updates test_helper.tcl's help with undocumented options
Specifically:
* Host
* Port
* Client
commit bde6f9ced15755cd6407b4af7d601b030f36d60b
Author: wxisme <850885154@qq.com>
Date: Wed Aug 8 15:19:19 2018 +0800
fix comments in deps files
commit 3172474ba991532ab799ee1873439f3402412331
Author: wxisme <850885154@qq.com>
Date: Wed Aug 8 14:33:49 2018 +0800
fix some comments
commit 01b6f2b6858b5cf2ce4ad5092d2c746e755f53f0
Author: Thor Juhasz <thor@juhasz.pro>
Date: Sun Nov 18 14:37:41 2018 +0100
Minor fixes to comments
Found some parts a little unclear on a first read, which prompted me to have a better look at the file and fix some minor things I noticed.
Fixing minor typos and grammar. There are no changes to configuration options.
These changes are only meant to help the user better understand the explanations to the various configuration options
2020-09-10 13:43:38 +03:00
|
|
|
* structure in an easy way.
|
2012-02-29 00:46:50 +01:00
|
|
|
*
|
2024-04-03 11:26:20 +07:00
|
|
|
* int serverOpArrayAppend(serverOpArray *oa, int dbid, robj **argv, int argc, int target);
|
|
|
|
* void serverOpArrayFree(serverOpArray *oa);
|
2012-02-29 00:46:50 +01:00
|
|
|
*/
|
2024-04-03 11:26:20 +07:00
|
|
|
typedef struct serverOpArray {
|
2024-04-04 01:30:30 +07:00
|
|
|
serverOp *ops;
|
2022-08-24 12:51:36 +03:00
|
|
|
int numops;
|
|
|
|
int capacity;
|
2024-04-03 11:26:20 +07:00
|
|
|
} serverOpArray;
|
2012-02-28 18:03:08 +01:00
|
|
|
|
2016-09-15 09:42:51 +02:00
|
|
|
/* This structure is returned by the getMemoryOverheadData() function in
|
|
|
|
* order to return memory overhead information. */
|
2024-04-03 09:29:42 +07:00
|
|
|
struct serverMemOverhead {
|
2016-09-16 10:43:19 +02:00
|
|
|
size_t peak_allocated;
|
2016-09-15 09:42:51 +02:00
|
|
|
size_t total_allocated;
|
|
|
|
size_t startup_allocated;
|
|
|
|
size_t repl_backlog;
|
2024-06-07 14:21:33 -07:00
|
|
|
size_t clients_replicas;
|
2016-09-15 09:42:51 +02:00
|
|
|
size_t clients_normal;
|
2021-12-16 21:56:59 -08:00
|
|
|
size_t cluster_links;
|
2016-09-15 09:42:51 +02:00
|
|
|
size_t aof_buffer;
|
2018-07-22 21:16:00 +03:00
|
|
|
size_t lua_caches;
|
2021-10-07 14:41:26 +03:00
|
|
|
size_t functions_caches;
|
2016-09-15 09:42:51 +02:00
|
|
|
size_t overhead_total;
|
|
|
|
size_t dataset;
|
2016-09-16 16:36:53 +02:00
|
|
|
size_t total_keys;
|
|
|
|
size_t bytes_per_key;
|
2016-09-15 17:33:11 +02:00
|
|
|
float dataset_perc;
|
2016-09-16 10:43:19 +02:00
|
|
|
float peak_perc;
|
2018-02-18 17:36:21 +02:00
|
|
|
float total_frag;
|
2018-12-02 15:29:20 +02:00
|
|
|
ssize_t total_frag_bytes;
|
2018-02-18 17:36:21 +02:00
|
|
|
float allocator_frag;
|
2018-12-02 15:29:20 +02:00
|
|
|
ssize_t allocator_frag_bytes;
|
2018-02-18 17:36:21 +02:00
|
|
|
float allocator_rss;
|
2018-12-02 15:29:20 +02:00
|
|
|
ssize_t allocator_rss_bytes;
|
2018-02-18 17:36:21 +02:00
|
|
|
float rss_extra;
|
|
|
|
size_t rss_extra_bytes;
|
2016-09-15 09:42:51 +02:00
|
|
|
size_t num_dbs;
|
2024-03-01 13:41:24 +08:00
|
|
|
size_t overhead_db_hashtable_lut;
|
|
|
|
size_t overhead_db_hashtable_rehashing;
|
|
|
|
unsigned long db_dict_rehashing_count;
|
2016-09-15 09:42:51 +02:00
|
|
|
struct {
|
|
|
|
size_t dbid;
|
|
|
|
size_t overhead_ht_main;
|
|
|
|
size_t overhead_ht_expires;
|
|
|
|
} *db;
|
|
|
|
};
|
|
|
|
|
Set replicas to panic on disk errors, and optionally panic on replication errors (#10504)
* Till now, replicas that were unable to persist, would still execute the commands
they got from the master, now they'll panic by default, and we add a new
`replica-ignore-disk-errors` config to change that.
* Till now, when a command failed on a replica or AOF-loading, it only logged a
warning and a stat, we add a new `propagation-error-behavior` config to allow
panicking in that state (may become the default one day)
Note that commands that fail on the replica can either indicate a bug that could
cause data inconsistency between the replica and the master, or they could be
in some cases (specifically in previous versions), a result of a command (e.g. EVAL)
that failed on the master, but still had to be propagated to fail on the replica as well.
2022-04-26 03:25:33 -07:00
|
|
|
/* Replication error behavior determines the replica behavior
|
|
|
|
* when it receives an error over the replication stream. In
|
|
|
|
* either case the error is logged. */
|
2022-04-26 19:29:05 +08:00
|
|
|
typedef enum {
|
Set replicas to panic on disk errors, and optionally panic on replication errors (#10504)
* Till now, replicas that were unable to persist, would still execute the commands
they got from the master, now they'll panic by default, and we add a new
`replica-ignore-disk-errors` config to change that.
* Till now, when a command failed on a replica or AOF-loading, it only logged a
warning and a stat, we add a new `propagation-error-behavior` config to allow
panicking in that state (may become the default one day)
Note that commands that fail on the replica can either indicate a bug that could
cause data inconsistency between the replica and the master, or they could be
in some cases (specifically in previous versions), a result of a command (e.g. EVAL)
that failed on the master, but still had to be propagated to fail on the replica as well.
2022-04-26 03:25:33 -07:00
|
|
|
PROPAGATION_ERR_BEHAVIOR_IGNORE = 0,
|
|
|
|
PROPAGATION_ERR_BEHAVIOR_PANIC,
|
|
|
|
PROPAGATION_ERR_BEHAVIOR_PANIC_ON_REPLICAS
|
|
|
|
} replicationErrorBehavior;
|
|
|
|
|
PSYNC2: different improvements to Redis replication.
The gist of the changes is that now, partial resynchronizations between
slaves and masters (without the need of a full resync with RDB transfer
and so forth), work in a number of cases when it was impossible
in the past. For instance:
1. When a slave is promoted to mastrer, the slaves of the old master can
partially resynchronize with the new master.
2. Chained slalves (slaves of slaves) can be moved to replicate to other
slaves or the master itsef, without requiring a full resync.
3. The master itself, after being turned into a slave, is able to
partially resynchronize with the new master, when it joins replication
again.
In order to obtain this, the following main changes were operated:
* Slaves also take a replication backlog, not just masters.
* Same stream replication for all the slaves and sub slaves. The
replication stream is identical from the top level master to its slaves
and is also the same from the slaves to their sub-slaves and so forth.
This means that if a slave is later promoted to master, it has the
same replication backlong, and can partially resynchronize with its
slaves (that were previously slaves of the old master).
* A given replication history is no longer identified by the `runid` of
a Redis node. There is instead a `replication ID` which changes every
time the instance has a new history no longer coherent with the past
one. So, for example, slaves publish the same replication history of
their master, however when they are turned into masters, they publish
a new replication ID, but still remember the old ID, so that they are
able to partially resynchronize with slaves of the old master (up to a
given offset).
* The replication protocol was slightly modified so that a new extended
+CONTINUE reply from the master is able to inform the slave of a
replication ID change.
* REPLCONF CAPA is used in order to notify masters that a slave is able
to understand the new +CONTINUE reply.
* The RDB file was extended with an auxiliary field that is able to
select a given DB after loading in the slave, so that the slave can
continue receiving the replication stream from the point it was
disconnected without requiring the master to insert "SELECT" statements.
This is useful in order to guarantee the "same stream" property, because
the slave must be able to accumulate an identical backlog.
* Slave pings to sub-slaves are now sent in a special form, when the
top-level master is disconnected, in order to don't interfer with the
replication stream. We just use out of band "\n" bytes as in other parts
of the Redis protocol.
An old design document is available here:
https://gist.github.com/antirez/ae068f95c0d084891305
However the implementation is not identical to the description because
during the work to implement it, different changes were needed in order
to make things working well.
2016-11-09 11:31:06 +01:00
|
|
|
/* This structure can be optionally passed to RDB save/load functions in
|
|
|
|
* order to implement additional functionalities, by storing and loading
|
|
|
|
* metadata to the RDB file.
|
|
|
|
*
|
2021-11-30 20:10:49 +08:00
|
|
|
* For example, to use select a DB at load time, useful in
|
2024-06-07 14:21:33 -07:00
|
|
|
* replication in order to make sure that chained replicas (replicas of replicas)
|
PSYNC2: different improvements to Redis replication.
The gist of the changes is that now, partial resynchronizations between
slaves and masters (without the need of a full resync with RDB transfer
and so forth), work in a number of cases when it was impossible
in the past. For instance:
1. When a slave is promoted to mastrer, the slaves of the old master can
partially resynchronize with the new master.
2. Chained slalves (slaves of slaves) can be moved to replicate to other
slaves or the master itsef, without requiring a full resync.
3. The master itself, after being turned into a slave, is able to
partially resynchronize with the new master, when it joins replication
again.
In order to obtain this, the following main changes were operated:
* Slaves also take a replication backlog, not just masters.
* Same stream replication for all the slaves and sub slaves. The
replication stream is identical from the top level master to its slaves
and is also the same from the slaves to their sub-slaves and so forth.
This means that if a slave is later promoted to master, it has the
same replication backlong, and can partially resynchronize with its
slaves (that were previously slaves of the old master).
* A given replication history is no longer identified by the `runid` of
a Redis node. There is instead a `replication ID` which changes every
time the instance has a new history no longer coherent with the past
one. So, for example, slaves publish the same replication history of
their master, however when they are turned into masters, they publish
a new replication ID, but still remember the old ID, so that they are
able to partially resynchronize with slaves of the old master (up to a
given offset).
* The replication protocol was slightly modified so that a new extended
+CONTINUE reply from the master is able to inform the slave of a
replication ID change.
* REPLCONF CAPA is used in order to notify masters that a slave is able
to understand the new +CONTINUE reply.
* The RDB file was extended with an auxiliary field that is able to
select a given DB after loading in the slave, so that the slave can
continue receiving the replication stream from the point it was
disconnected without requiring the master to insert "SELECT" statements.
This is useful in order to guarantee the "same stream" property, because
the slave must be able to accumulate an identical backlog.
* Slave pings to sub-slaves are now sent in a special form, when the
top-level master is disconnected, in order to don't interfer with the
replication stream. We just use out of band "\n" bytes as in other parts
of the Redis protocol.
An old design document is available here:
https://gist.github.com/antirez/ae068f95c0d084891305
However the implementation is not identical to the description because
during the work to implement it, different changes were needed in order
to make things working well.
2016-11-09 11:31:06 +01:00
|
|
|
* select the correct DB and are able to accept the stream coming from the
|
2024-06-07 14:21:33 -07:00
|
|
|
* top-level primary. */
|
PSYNC2: different improvements to Redis replication.
The gist of the changes is that now, partial resynchronizations between
slaves and masters (without the need of a full resync with RDB transfer
and so forth), work in a number of cases when it was impossible
in the past. For instance:
1. When a slave is promoted to mastrer, the slaves of the old master can
partially resynchronize with the new master.
2. Chained slalves (slaves of slaves) can be moved to replicate to other
slaves or the master itsef, without requiring a full resync.
3. The master itself, after being turned into a slave, is able to
partially resynchronize with the new master, when it joins replication
again.
In order to obtain this, the following main changes were operated:
* Slaves also take a replication backlog, not just masters.
* Same stream replication for all the slaves and sub slaves. The
replication stream is identical from the top level master to its slaves
and is also the same from the slaves to their sub-slaves and so forth.
This means that if a slave is later promoted to master, it has the
same replication backlong, and can partially resynchronize with its
slaves (that were previously slaves of the old master).
* A given replication history is no longer identified by the `runid` of
a Redis node. There is instead a `replication ID` which changes every
time the instance has a new history no longer coherent with the past
one. So, for example, slaves publish the same replication history of
their master, however when they are turned into masters, they publish
a new replication ID, but still remember the old ID, so that they are
able to partially resynchronize with slaves of the old master (up to a
given offset).
* The replication protocol was slightly modified so that a new extended
+CONTINUE reply from the master is able to inform the slave of a
replication ID change.
* REPLCONF CAPA is used in order to notify masters that a slave is able
to understand the new +CONTINUE reply.
* The RDB file was extended with an auxiliary field that is able to
select a given DB after loading in the slave, so that the slave can
continue receiving the replication stream from the point it was
disconnected without requiring the master to insert "SELECT" statements.
This is useful in order to guarantee the "same stream" property, because
the slave must be able to accumulate an identical backlog.
* Slave pings to sub-slaves are now sent in a special form, when the
top-level master is disconnected, in order to don't interfer with the
replication stream. We just use out of band "\n" bytes as in other parts
of the Redis protocol.
An old design document is available here:
https://gist.github.com/antirez/ae068f95c0d084891305
However the implementation is not identical to the description because
during the work to implement it, different changes were needed in order
to make things working well.
2016-11-09 11:31:06 +01:00
|
|
|
typedef struct rdbSaveInfo {
|
2016-11-10 12:35:29 +01:00
|
|
|
/* Used saving and loading. */
|
2024-06-07 14:21:33 -07:00
|
|
|
int repl_stream_db; /* DB to select in server.primary client. */
|
2016-11-10 12:35:29 +01:00
|
|
|
|
|
|
|
/* Used only loading. */
|
2024-05-22 23:24:12 -07:00
|
|
|
int repl_id_is_set; /* True if repl_id field is set. */
|
|
|
|
char repl_id[CONFIG_RUN_ID_SIZE + 1]; /* Replication ID. */
|
|
|
|
long long repl_offset; /* Replication offset. */
|
PSYNC2: different improvements to Redis replication.
The gist of the changes is that now, partial resynchronizations between
slaves and masters (without the need of a full resync with RDB transfer
and so forth), work in a number of cases when it was impossible
in the past. For instance:
1. When a slave is promoted to mastrer, the slaves of the old master can
partially resynchronize with the new master.
2. Chained slalves (slaves of slaves) can be moved to replicate to other
slaves or the master itsef, without requiring a full resync.
3. The master itself, after being turned into a slave, is able to
partially resynchronize with the new master, when it joins replication
again.
In order to obtain this, the following main changes were operated:
* Slaves also take a replication backlog, not just masters.
* Same stream replication for all the slaves and sub slaves. The
replication stream is identical from the top level master to its slaves
and is also the same from the slaves to their sub-slaves and so forth.
This means that if a slave is later promoted to master, it has the
same replication backlong, and can partially resynchronize with its
slaves (that were previously slaves of the old master).
* A given replication history is no longer identified by the `runid` of
a Redis node. There is instead a `replication ID` which changes every
time the instance has a new history no longer coherent with the past
one. So, for example, slaves publish the same replication history of
their master, however when they are turned into masters, they publish
a new replication ID, but still remember the old ID, so that they are
able to partially resynchronize with slaves of the old master (up to a
given offset).
* The replication protocol was slightly modified so that a new extended
+CONTINUE reply from the master is able to inform the slave of a
replication ID change.
* REPLCONF CAPA is used in order to notify masters that a slave is able
to understand the new +CONTINUE reply.
* The RDB file was extended with an auxiliary field that is able to
select a given DB after loading in the slave, so that the slave can
continue receiving the replication stream from the point it was
disconnected without requiring the master to insert "SELECT" statements.
This is useful in order to guarantee the "same stream" property, because
the slave must be able to accumulate an identical backlog.
* Slave pings to sub-slaves are now sent in a special form, when the
top-level master is disconnected, in order to don't interfer with the
replication stream. We just use out of band "\n" bytes as in other parts
of the Redis protocol.
An old design document is available here:
https://gist.github.com/antirez/ae068f95c0d084891305
However the implementation is not identical to the description because
during the work to implement it, different changes were needed in order
to make things working well.
2016-11-09 11:31:06 +01:00
|
|
|
} rdbSaveInfo;
|
|
|
|
|
2024-05-22 23:24:12 -07:00
|
|
|
#define RDB_SAVE_INFO_INIT {-1, 0, "0000000000000000000000000000000000000000", -1}
|
PSYNC2: different improvements to Redis replication.
The gist of the changes is that now, partial resynchronizations between
slaves and masters (without the need of a full resync with RDB transfer
and so forth), work in a number of cases when it was impossible
in the past. For instance:
1. When a slave is promoted to mastrer, the slaves of the old master can
partially resynchronize with the new master.
2. Chained slalves (slaves of slaves) can be moved to replicate to other
slaves or the master itsef, without requiring a full resync.
3. The master itself, after being turned into a slave, is able to
partially resynchronize with the new master, when it joins replication
again.
In order to obtain this, the following main changes were operated:
* Slaves also take a replication backlog, not just masters.
* Same stream replication for all the slaves and sub slaves. The
replication stream is identical from the top level master to its slaves
and is also the same from the slaves to their sub-slaves and so forth.
This means that if a slave is later promoted to master, it has the
same replication backlong, and can partially resynchronize with its
slaves (that were previously slaves of the old master).
* A given replication history is no longer identified by the `runid` of
a Redis node. There is instead a `replication ID` which changes every
time the instance has a new history no longer coherent with the past
one. So, for example, slaves publish the same replication history of
their master, however when they are turned into masters, they publish
a new replication ID, but still remember the old ID, so that they are
able to partially resynchronize with slaves of the old master (up to a
given offset).
* The replication protocol was slightly modified so that a new extended
+CONTINUE reply from the master is able to inform the slave of a
replication ID change.
* REPLCONF CAPA is used in order to notify masters that a slave is able
to understand the new +CONTINUE reply.
* The RDB file was extended with an auxiliary field that is able to
select a given DB after loading in the slave, so that the slave can
continue receiving the replication stream from the point it was
disconnected without requiring the master to insert "SELECT" statements.
This is useful in order to guarantee the "same stream" property, because
the slave must be able to accumulate an identical backlog.
* Slave pings to sub-slaves are now sent in a special form, when the
top-level master is disconnected, in order to don't interfer with the
replication stream. We just use out of band "\n" bytes as in other parts
of the Redis protocol.
An old design document is available here:
https://gist.github.com/antirez/ae068f95c0d084891305
However the implementation is not identical to the description because
during the work to implement it, different changes were needed in order
to make things working well.
2016-11-09 11:31:06 +01:00
|
|
|
|
2018-06-19 16:59:45 +03:00
|
|
|
struct malloc_stats {
|
2018-02-18 17:36:21 +02:00
|
|
|
size_t zmalloc_used;
|
|
|
|
size_t process_rss;
|
|
|
|
size_t allocator_allocated;
|
|
|
|
size_t allocator_active;
|
|
|
|
size_t allocator_resident;
|
2024-02-21 00:11:09 +08:00
|
|
|
size_t allocator_muzzy;
|
|
|
|
size_t allocator_frag_smallbins_bytes;
|
2018-06-19 16:59:45 +03:00
|
|
|
};
|
2018-02-18 17:36:21 +02:00
|
|
|
|
2024-07-10 13:53:52 +02:00
|
|
|
/*-----------------------------------------------------------------------------
|
|
|
|
* Cached state per client connection type flags (bitwise or)
|
|
|
|
*-----------------------------------------------------------------------------*/
|
|
|
|
|
|
|
|
#define CACHE_CONN_TYPE_TLS (1 << 0)
|
|
|
|
#define CACHE_CONN_TYPE_IPv6 (1 << 1)
|
|
|
|
#define CACHE_CONN_TYPE_RESP3 (1 << 2)
|
|
|
|
#define CACHE_CONN_TYPE_MAX (1 << 3)
|
|
|
|
|
2019-09-12 11:10:22 +03:00
|
|
|
/*-----------------------------------------------------------------------------
|
|
|
|
* TLS Context Configuration
|
|
|
|
*----------------------------------------------------------------------------*/
|
|
|
|
|
2024-04-04 00:06:01 +07:00
|
|
|
typedef struct serverTLSContextConfig {
|
2024-05-22 23:24:12 -07:00
|
|
|
char *cert_file; /* Server side and optionally client side cert file name */
|
|
|
|
char *key_file; /* Private key filename for cert_file */
|
|
|
|
char *key_file_pass; /* Optional password for key_file */
|
|
|
|
char *client_cert_file; /* Certificate to use as a client; if none, use cert_file */
|
|
|
|
char *client_key_file; /* Private key filename for client_cert_file */
|
|
|
|
char *client_key_file_pass; /* Optional password for client_key_file */
|
2019-09-12 11:10:22 +03:00
|
|
|
char *dh_params_file;
|
|
|
|
char *ca_cert_file;
|
2019-10-08 17:57:05 +03:00
|
|
|
char *ca_cert_dir;
|
2019-09-12 11:10:22 +03:00
|
|
|
char *protocols;
|
|
|
|
char *ciphers;
|
|
|
|
char *ciphersuites;
|
|
|
|
int prefer_server_ciphers;
|
2020-07-10 11:33:47 +03:00
|
|
|
int session_caching;
|
|
|
|
int session_cache_size;
|
|
|
|
int session_cache_timeout;
|
2024-04-04 00:06:01 +07:00
|
|
|
} serverTLSContextConfig;
|
2019-09-12 11:10:22 +03:00
|
|
|
|
Implement Multi Part AOF mechanism to avoid AOFRW overheads. (#9788)
Implement Multi-Part AOF mechanism to avoid overheads during AOFRW.
Introducing a folder with multiple AOF files tracked by a manifest file.
The main issues with the the original AOFRW mechanism are:
* buffering of commands that are processed during rewrite (consuming a lot of RAM)
* freezes of the main process when the AOFRW completes to drain the remaining part of the buffer and fsync it.
* double disk IO for the data that arrives during AOFRW (had to be written to both the old and new AOF files)
The main modifications of this PR:
1. Remove the AOF rewrite buffer and related code.
2. Divide the AOF into multiple files, they are classified as two types, one is the the `BASE` type,
it represents the full amount of data (Maybe AOF or RDB format) after each AOFRW, there is only
one `BASE` file at most. The second is `INCR` type, may have more than one. They represent the
incremental commands since the last AOFRW.
3. Use a AOF manifest file to record and manage these AOF files mentioned above.
4. The original configuration of `appendfilename` will be the base part of the new file name, for example:
`appendonly.aof.1.base.rdb` and `appendonly.aof.2.incr.aof`
5. Add manifest-related TCL tests, and modified some existing tests that depend on the `appendfilename`
6. Remove the `aof_rewrite_buffer_length` field in info.
7. Add `aof-disable-auto-gc` configuration. By default we're automatically deleting HISTORY type AOFs.
It also gives users the opportunity to preserve the history AOFs. just for testing use now.
8. Add AOFRW limiting measure. When the AOFRW failures reaches the threshold (3 times now),
we will delay the execution of the next AOFRW by 1 minute. If the next AOFRW also fails, it will be
delayed by 2 minutes. The next is 4, 8, 16, the maximum delay is 60 minutes (1 hour). During the limit
period, we can still use the 'bgrewriteaof' command to execute AOFRW immediately.
9. Support upgrade (load) data from old version redis.
10. Add `appenddirname` configuration, as the directory name of the append only files. All AOF files and
manifest file will be placed in this directory.
11. Only the last AOF file (BASE or INCR) can be truncated. Otherwise redis will exit even if
`aof-load-truncated` is enabled.
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-04 01:14:13 +08:00
|
|
|
/*-----------------------------------------------------------------------------
|
|
|
|
* AOF manifest definition
|
|
|
|
*----------------------------------------------------------------------------*/
|
|
|
|
typedef enum {
|
2024-05-22 23:24:12 -07:00
|
|
|
AOF_FILE_TYPE_BASE = 'b', /* BASE file */
|
|
|
|
AOF_FILE_TYPE_HIST = 'h', /* HISTORY file */
|
|
|
|
AOF_FILE_TYPE_INCR = 'i', /* INCR file */
|
Implement Multi Part AOF mechanism to avoid AOFRW overheads. (#9788)
Implement Multi-Part AOF mechanism to avoid overheads during AOFRW.
Introducing a folder with multiple AOF files tracked by a manifest file.
The main issues with the the original AOFRW mechanism are:
* buffering of commands that are processed during rewrite (consuming a lot of RAM)
* freezes of the main process when the AOFRW completes to drain the remaining part of the buffer and fsync it.
* double disk IO for the data that arrives during AOFRW (had to be written to both the old and new AOF files)
The main modifications of this PR:
1. Remove the AOF rewrite buffer and related code.
2. Divide the AOF into multiple files, they are classified as two types, one is the the `BASE` type,
it represents the full amount of data (Maybe AOF or RDB format) after each AOFRW, there is only
one `BASE` file at most. The second is `INCR` type, may have more than one. They represent the
incremental commands since the last AOFRW.
3. Use a AOF manifest file to record and manage these AOF files mentioned above.
4. The original configuration of `appendfilename` will be the base part of the new file name, for example:
`appendonly.aof.1.base.rdb` and `appendonly.aof.2.incr.aof`
5. Add manifest-related TCL tests, and modified some existing tests that depend on the `appendfilename`
6. Remove the `aof_rewrite_buffer_length` field in info.
7. Add `aof-disable-auto-gc` configuration. By default we're automatically deleting HISTORY type AOFs.
It also gives users the opportunity to preserve the history AOFs. just for testing use now.
8. Add AOFRW limiting measure. When the AOFRW failures reaches the threshold (3 times now),
we will delay the execution of the next AOFRW by 1 minute. If the next AOFRW also fails, it will be
delayed by 2 minutes. The next is 4, 8, 16, the maximum delay is 60 minutes (1 hour). During the limit
period, we can still use the 'bgrewriteaof' command to execute AOFRW immediately.
9. Support upgrade (load) data from old version redis.
10. Add `appenddirname` configuration, as the directory name of the append only files. All AOF files and
manifest file will be placed in this directory.
11. Only the last AOF file (BASE or INCR) can be truncated. Otherwise redis will exit even if
`aof-load-truncated` is enabled.
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-04 01:14:13 +08:00
|
|
|
} aof_file_type;
|
|
|
|
|
|
|
|
typedef struct {
|
2024-05-22 23:24:12 -07:00
|
|
|
sds file_name; /* file name */
|
|
|
|
long long file_seq; /* file sequence */
|
|
|
|
aof_file_type file_type; /* file type */
|
Implement Multi Part AOF mechanism to avoid AOFRW overheads. (#9788)
Implement Multi-Part AOF mechanism to avoid overheads during AOFRW.
Introducing a folder with multiple AOF files tracked by a manifest file.
The main issues with the the original AOFRW mechanism are:
* buffering of commands that are processed during rewrite (consuming a lot of RAM)
* freezes of the main process when the AOFRW completes to drain the remaining part of the buffer and fsync it.
* double disk IO for the data that arrives during AOFRW (had to be written to both the old and new AOF files)
The main modifications of this PR:
1. Remove the AOF rewrite buffer and related code.
2. Divide the AOF into multiple files, they are classified as two types, one is the the `BASE` type,
it represents the full amount of data (Maybe AOF or RDB format) after each AOFRW, there is only
one `BASE` file at most. The second is `INCR` type, may have more than one. They represent the
incremental commands since the last AOFRW.
3. Use a AOF manifest file to record and manage these AOF files mentioned above.
4. The original configuration of `appendfilename` will be the base part of the new file name, for example:
`appendonly.aof.1.base.rdb` and `appendonly.aof.2.incr.aof`
5. Add manifest-related TCL tests, and modified some existing tests that depend on the `appendfilename`
6. Remove the `aof_rewrite_buffer_length` field in info.
7. Add `aof-disable-auto-gc` configuration. By default we're automatically deleting HISTORY type AOFs.
It also gives users the opportunity to preserve the history AOFs. just for testing use now.
8. Add AOFRW limiting measure. When the AOFRW failures reaches the threshold (3 times now),
we will delay the execution of the next AOFRW by 1 minute. If the next AOFRW also fails, it will be
delayed by 2 minutes. The next is 4, 8, 16, the maximum delay is 60 minutes (1 hour). During the limit
period, we can still use the 'bgrewriteaof' command to execute AOFRW immediately.
9. Support upgrade (load) data from old version redis.
10. Add `appenddirname` configuration, as the directory name of the append only files. All AOF files and
manifest file will be placed in this directory.
11. Only the last AOF file (BASE or INCR) can be truncated. Otherwise redis will exit even if
`aof-load-truncated` is enabled.
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-04 01:14:13 +08:00
|
|
|
} aofInfo;
|
|
|
|
|
|
|
|
typedef struct {
|
2024-05-22 23:24:12 -07:00
|
|
|
aofInfo *base_aof_info; /* BASE file information. NULL if there is no BASE file. */
|
|
|
|
list *incr_aof_list; /* INCR AOFs list. We may have multiple INCR AOF when rewrite fails. */
|
|
|
|
list *history_aof_list; /* HISTORY AOF list. When the AOFRW success, The aofInfo contained in
|
|
|
|
`base_aof_info` and `incr_aof_list` will be moved to this list. We
|
|
|
|
will delete these AOF files when AOFRW finish. */
|
|
|
|
long long curr_base_file_seq; /* The sequence number used by the current BASE file. */
|
|
|
|
long long curr_incr_file_seq; /* The sequence number used by the current INCR file. */
|
|
|
|
int dirty; /* 1 Indicates that the aofManifest in the memory is inconsistent with
|
|
|
|
disk, we need to persist it immediately. */
|
Implement Multi Part AOF mechanism to avoid AOFRW overheads. (#9788)
Implement Multi-Part AOF mechanism to avoid overheads during AOFRW.
Introducing a folder with multiple AOF files tracked by a manifest file.
The main issues with the the original AOFRW mechanism are:
* buffering of commands that are processed during rewrite (consuming a lot of RAM)
* freezes of the main process when the AOFRW completes to drain the remaining part of the buffer and fsync it.
* double disk IO for the data that arrives during AOFRW (had to be written to both the old and new AOF files)
The main modifications of this PR:
1. Remove the AOF rewrite buffer and related code.
2. Divide the AOF into multiple files, they are classified as two types, one is the the `BASE` type,
it represents the full amount of data (Maybe AOF or RDB format) after each AOFRW, there is only
one `BASE` file at most. The second is `INCR` type, may have more than one. They represent the
incremental commands since the last AOFRW.
3. Use a AOF manifest file to record and manage these AOF files mentioned above.
4. The original configuration of `appendfilename` will be the base part of the new file name, for example:
`appendonly.aof.1.base.rdb` and `appendonly.aof.2.incr.aof`
5. Add manifest-related TCL tests, and modified some existing tests that depend on the `appendfilename`
6. Remove the `aof_rewrite_buffer_length` field in info.
7. Add `aof-disable-auto-gc` configuration. By default we're automatically deleting HISTORY type AOFs.
It also gives users the opportunity to preserve the history AOFs. just for testing use now.
8. Add AOFRW limiting measure. When the AOFRW failures reaches the threshold (3 times now),
we will delay the execution of the next AOFRW by 1 minute. If the next AOFRW also fails, it will be
delayed by 2 minutes. The next is 4, 8, 16, the maximum delay is 60 minutes (1 hour). During the limit
period, we can still use the 'bgrewriteaof' command to execute AOFRW immediately.
9. Support upgrade (load) data from old version redis.
10. Add `appenddirname` configuration, as the directory name of the append only files. All AOF files and
manifest file will be placed in this directory.
11. Only the last AOF file (BASE or INCR) can be truncated. Otherwise redis will exit even if
`aof-load-truncated` is enabled.
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-04 01:14:13 +08:00
|
|
|
} aofManifest;
|
|
|
|
|
2011-03-29 17:51:15 +02:00
|
|
|
/*-----------------------------------------------------------------------------
|
|
|
|
* Global server state
|
|
|
|
*----------------------------------------------------------------------------*/
|
|
|
|
|
2014-08-07 12:14:21 +02:00
|
|
|
/* AIX defines hz to __hz, we don't use this define and in order to allow
|
2024-04-09 01:24:03 -07:00
|
|
|
* the server build on AIX we need to undef it. */
|
2014-07-29 16:39:37 -05:00
|
|
|
#ifdef _AIX
|
|
|
|
#undef hz
|
|
|
|
#endif
|
|
|
|
|
2020-09-20 13:43:28 +03:00
|
|
|
#define CHILD_TYPE_NONE 0
|
|
|
|
#define CHILD_TYPE_RDB 1
|
|
|
|
#define CHILD_TYPE_AOF 2
|
|
|
|
#define CHILD_TYPE_LDB 3
|
|
|
|
#define CHILD_TYPE_MODULE 4
|
2016-09-19 13:45:20 +02:00
|
|
|
|
2021-02-16 16:06:51 +02:00
|
|
|
typedef enum childInfoType {
|
|
|
|
CHILD_INFO_TYPE_CURRENT_INFO,
|
|
|
|
CHILD_INFO_TYPE_AOF_COW_SIZE,
|
|
|
|
CHILD_INFO_TYPE_RDB_COW_SIZE,
|
|
|
|
CHILD_INFO_TYPE_MODULE_COW_SIZE
|
|
|
|
} childInfoType;
|
|
|
|
|
2024-04-04 01:34:18 +07:00
|
|
|
struct valkeyServer {
|
2011-01-14 00:19:19 +01:00
|
|
|
/* General */
|
2024-05-22 23:24:12 -07:00
|
|
|
pid_t pid; /* Main process pid. */
|
|
|
|
pthread_t main_thread_id; /* Main thread id */
|
|
|
|
char *configfile; /* Absolute config file path, or NULL */
|
|
|
|
char *executable; /* Absolute executable file path. */
|
|
|
|
char **exec_argv; /* Executable argv vector (copy). */
|
|
|
|
int dynamic_hz; /* Change hz value depending on # of clients. */
|
|
|
|
int config_hz; /* Configured HZ value. May be different than
|
|
|
|
the actual 'hz' field value if dynamic-hz
|
|
|
|
is enabled. */
|
|
|
|
mode_t umask; /* The umask value of the process on startup */
|
|
|
|
int hz; /* serverCron() calls frequency in hertz */
|
|
|
|
int in_fork_child; /* indication that this is a fork child */
|
2024-04-03 10:02:43 +07:00
|
|
|
serverDb *db;
|
2024-05-22 23:24:12 -07:00
|
|
|
dict *commands; /* Command table */
|
|
|
|
dict *orig_commands; /* Command table before command renaming. */
|
2011-01-14 00:19:19 +01:00
|
|
|
aeEventLoop *el;
|
Io thread work offload (#763)
### IO-Threads Work Offloading
This PR is the 2nd of 3 PRs intended to achieve the goal of 1M requests
per second.
(1st PR: https://github.com/valkey-io/valkey/pull/758)
This PR offloads additional work to the I/O threads, beyond the current
read-parse/write operations, to better utilize the I/O threads and
reduce the load on the main thread.
It contains the following 3 commits:
### Poll Offload
Currently, the main thread is responsible for executing the poll-wait
system call, while the IO threads wait for tasks from the main thread.
The poll-wait operation is expensive and can consume up to 30% of the
main thread's time. We could have let the IO threads do the poll-wait by
themselves, with each thread listening to some of the clients and
notifying the main thread when a client's command is ready to execute.
However, the current approach, where the main thread listens for events
from the network, has several benefits. The main thread remains in
charge, allowing it to know the state of each client
(idle/read/write/close) at any given time. Additionally, it makes the
threads flexible, enabling us to drain an IO thread's job queue and stop
a thread when the load is light without modifying the event loop and
moving its clients to a different IO thread. Furthermore, with this
approach, the IO threads don't need to wait for both messages from the
network and from the main thread instead, the threads wait only for
tasks from the main thread.
To enjoy the benefits of both the main thread remaining in charge and
the poll being offloaded, we propose offloading the poll-wait as a
single-time, non-blocking job to one of the IO threads. The IO thread
will perform a poll-wait non-blocking call while the main thread
processes the client commands. Later, in `aeProcessEvents`, instead of
sleeping on the poll, we check for the IO thread's poll-wait results.
The poll-wait will be offloaded in `beforeSleep` only when there are
ready events for the main thread to process. If no events are pending,
the main thread will revert to the current behavior and sleep on the
poll by itself.
**Implementation Details**
A new call back `custompoll` was added to the `aeEventLoop` when not set
to `NULL` the ae will call the `custompoll` callback instead of the
`aeApiPoll`.
When the poll is offloaded we will set the `custompoll` to
`getIOThreadPollResults` and send a poll-job to the thread. the thread
will take a mutex, call a non-blocking (with timeout 0) to `aePoll`
which will populate the fired events array. the IO thread will set the
`server.io_fired_events` to the number of the returning `numevents`,
later the main-thread in `custompoll` will return the
`server.io_fired_events` and will set the `customPoll` back to `NULL`.
To ensure thread safety when accessing server.el, all functions that
modify the eventloop events were wrapped with a mutex to ensure mutual
exclusion when modifying the events.
### Command Lookup Offload
As the IO thread parses the command from the client's Querybuf, it can
perform a command lookup in the commands dictionary, which can consume
up to ~5% of the main-thread runtime.
**Implementation details**
The IO thread will store the looked-up command in the client's new field
`io_parsed_cmd` field. We can't use `c->cmd` for that since we use
`c->cmd `to check if a command was reprocessed or not.
To ensure thread safety when accessing the command dictionary, we make
sure the main thread isn't changing the dictionary while IO threads are
accessing it. This is accomplished by introducing a new flag called
`no_incremental_rehash` for the `dictType` commands. When performing
`dictResize`, we will rehash the entire dictionary in place rather than
deferring the process.
### Free Offload
Since the command arguments are allocated by the I/O thread, it would be
beneficial if they were also freed by the same thread. If the main
thread frees objects allocated by the I/O thread, two issues arise:
1. During the freeing process, the main thread needs to access the SDS
pointed to by the object to get its length.
2. With Jemalloc, each thread manages thread local pool (`tcache`) of
buffers for quick reallocation without accessing the arena. If the main
thread constantly frees objects allocated by other threads, those
threads will have to frequently access the shared arena to obtain new
memory allocations
**Implementation Details**
When freeing the client's argv, we will send the argv array to the
thread that allocated it. The thread will be identified by the client
ID. When freeing an object during `dbOverwrite`, we will offload the
object free as well. We will extend this to offload the free during
`dbDelete` in a future PR, as its effects on defrag/memory evictions
need to be studied.
---------
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-19 05:21:45 +03:00
|
|
|
_Atomic AeIoState io_poll_state; /* Indicates the state of the IO polling. */
|
|
|
|
int io_ae_fired_events; /* Number of poll events received by the IO thread. */
|
2024-05-22 23:24:12 -07:00
|
|
|
rax *errors; /* Errors table */
|
|
|
|
unsigned int lruclock; /* Clock for LRU eviction */
|
2022-01-02 08:50:15 +01:00
|
|
|
volatile sig_atomic_t shutdown_asap; /* Shutdown ordered by signal handler. */
|
2024-05-22 23:24:12 -07:00
|
|
|
mstime_t shutdown_mstime; /* Timestamp to limit graceful shutdown. */
|
|
|
|
int last_sig_received; /* Indicates the last SIGNAL received, if any (e.g., SIGINT or SIGTERM). */
|
|
|
|
int shutdown_flags; /* Flags passed to prepareForShutdown(). */
|
|
|
|
int activerehashing; /* Incremental rehash in serverCron() */
|
|
|
|
int active_defrag_running; /* Active defragmentation running (holds current scan aggressiveness) */
|
|
|
|
char *pidfile; /* PID file path */
|
|
|
|
int arch_bits; /* 32 or 64 depending on sizeof(long) */
|
|
|
|
int cronloops; /* Number of times the cron function run */
|
|
|
|
char runid[CONFIG_RUN_ID_SIZE + 1]; /* ID always different at every exec. */
|
|
|
|
int sentinel_mode; /* True if this instance is a Sentinel. */
|
|
|
|
size_t initial_memory_usage; /* Bytes used after initialization. */
|
|
|
|
int always_show_logo; /* Show logo even for non-stdout logging. */
|
|
|
|
int in_exec; /* Are we inside EXEC? */
|
|
|
|
int busy_module_yield_flags; /* Are we inside a busy module? (triggered by RM_Yield). see BUSY_MODULE_YIELD_ flags. */
|
2022-01-20 09:05:53 +02:00
|
|
|
const char *busy_module_yield_reply; /* When non-null, we are inside RM_Yield. */
|
2024-05-22 23:24:12 -07:00
|
|
|
char *ignore_warnings; /* Config: warnings that should be ignored. */
|
|
|
|
int client_pause_in_transaction; /* Was a client pause executed during this Exec? */
|
|
|
|
int thp_enabled; /* If true, THP is enabled. */
|
|
|
|
size_t page_size; /* The page size of OS. */
|
2016-03-06 13:44:24 +01:00
|
|
|
/* Modules */
|
2018-12-20 12:06:24 +01:00
|
|
|
dict *moduleapi; /* Exported core APIs dictionary for modules. */
|
|
|
|
dict *sharedapi; /* Like moduleapi but containing the APIs that
|
|
|
|
modules share with each other. */
|
2024-05-22 23:24:12 -07:00
|
|
|
dict *module_configs_queue; /* Dict that stores module configurations from .conf file until after modules are loaded
|
|
|
|
during startup or arguments to loadex. */
|
2016-03-06 13:44:24 +01:00
|
|
|
list *loadmodule_queue; /* List of modules to load at startup. */
|
Add event loop support to the module API (#10001)
Modules can now register sockets/pipe to the Redis main thread event loop and do network operations asynchronously. Previously, modules had to maintain an event loop and another thread for asynchronous network operations.
Also, if a module is calling API functions after doing some network operations, it had to synchronize its event loop thread's access with Redis main thread by locking the GIL, causing contention on the lock. After this commit, no synchronization is needed as module can operate in Redis main thread context. So, this commit may improve the performance for some use cases.
Added three functions to the module API:
* RedisModule_EventLoopAdd(int fd, int mask, RedisModuleEventLoopFunc func, void *user_data)
* RedisModule_EventLoopDel(int fd, int mask)
* RedisModule_EventLoopAddOneShot(RedisModuleEventLoopOneShotFunc func, void *user_data) - This function can be called from other threads to trigger callback on Redis main thread. Callback will be triggered only once. If Redis main thread is sleeping, this call will wake up the Redis main thread.
Event loop callbacks are called by Redis main thread after locking the GIL. Inside callbacks, modules can operate as if they are holding the GIL.
Added REDISMODULE_EVENT_EVENTLOOP event with two subevents:
* REDISMODULE_SUBEVENT_EVENTLOOP_BEFORE_SLEEP
* REDISMODULE_SUBEVENT_EVENTLOOP_AFTER_SLEEP
These events are for modules that want to participate in the before and after sleep action. e.g It might be useful to implement batching : Read data from the network, write all to a file in one go on BEFORE_SLEEP event.
2022-01-18 14:10:07 +03:00
|
|
|
int module_pipe[2]; /* Pipe used to awake the event loop by module threads. */
|
Refactory fork child related infra, Unify child pid
This is a refactory commit, isn't suppose to have any actual impact.
it does the following:
- keep just one server struct fork child pid variable instead of 3
- have one server struct variable indicating the purpose of the current fork
child.
- redisFork is now responsible of updating the server struct with the pid,
which means it can be the one that calls updateDictResizePolicy
- move child info pipe handling into redisFork instead of having them
repeated outside
- there are two classes of fork purposes, mutually exclusive group (AOF, RDB,
Module), and one that can create several forks to coexist in parallel (LDB,
but maybe Modules some day too, Module API allows for that).
- minor fix to killRDBChild:
unlike killAppendOnlyChild and TerminateModuleForkChild, the killRDBChild
doesn't clear the pid variable or call wait4, so checkChildrenDone does
the cleanup for it.
This commit removes the explicit calls to rdbRemoveTempFile, closeChildInfoPipe,
updateDictResizePolicy, which didn't do any harm, but where unnecessary.
2020-12-16 15:14:04 +02:00
|
|
|
pid_t child_pid; /* PID of current child */
|
|
|
|
int child_type; /* Type of current child */
|
2024-05-26 17:41:11 +01:00
|
|
|
_Atomic int module_gil_acquiring; /* Indicates whether the GIL is being acquiring by the main thread. */
|
2011-01-14 00:19:19 +01:00
|
|
|
/* Networking */
|
2024-05-22 23:24:12 -07:00
|
|
|
int port; /* TCP listening port */
|
|
|
|
int tls_port; /* TLS listening port */
|
|
|
|
int tcp_backlog; /* TCP listen() backlog */
|
|
|
|
char *bindaddr[CONFIG_BINDADDR_MAX]; /* Addresses we should bind to */
|
|
|
|
int bindaddr_count; /* Number of addresses in server.bindaddr[] */
|
|
|
|
char *bind_source_addr; /* Source address to bind on for outgoing connections */
|
|
|
|
char *unixsocket; /* UNIX socket path */
|
|
|
|
unsigned int unixsocketperm; /* UNIX socket permission (see mode_t) */
|
Introduce .listen into connection type
Introduce listen method into connection type, this allows no hard code
of listen logic. Originally, we initialize server during startup like
this:
if (server.port)
listenToPort(server.port,&server.ipfd);
if (server.tls_port)
listenToPort(server.port,&server.tlsfd);
if (server.unixsocket)
anetUnixServer(...server.unixsocket...);
...
if (createSocketAcceptHandler(&server.ipfd, acceptTcpHandler) != C_OK)
if (createSocketAcceptHandler(&server.tlsfd, acceptTcpHandler) != C_OK)
if (createSocketAcceptHandler(&server.sofd, acceptTcpHandler) != C_OK)
...
If a new connection type gets supported, we have to add more hard code
to setup listener.
Introduce .listen and refactor listener, and Unix socket supports this.
this allows to setup listener arguments and create listener in a loop.
What's more, '.listen' is defined in connection.h, so we should include
server.h to import 'struct socketFds', but server.h has already include
'connection.h'. To avoid including loop(also to make code reasonable),
define 'struct connListener' in connection.h instead of 'struct socketFds'
in server.h. This leads this commit to get more changes.
There are more fields in 'struct connListener', hence it's possible to
simplify changeBindAddr & applyTLSPort() & updatePort() into a single
logic: update the listener config from the server.xxx, and re-create
the listener.
Because of the new field 'priv' in struct connListener, we expect to pass
this to the accept handler(even it's not used currently), this may be used
in the future.
Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
2022-07-27 12:18:28 +08:00
|
|
|
connListener listeners[CONN_TYPE_MAX]; /* TCP/Unix/TLS even more types */
|
2024-05-22 23:24:12 -07:00
|
|
|
uint32_t socket_mark_id; /* ID for listen socket marking */
|
|
|
|
connListener clistener; /* Cluster bus listener */
|
|
|
|
list *clients; /* List of active clients */
|
|
|
|
list *clients_to_close; /* Clients to close asynchronously */
|
|
|
|
list *clients_pending_write; /* There is to write or install handler. */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
list *clients_pending_io_read; /* List of clients with pending read to be process by I/O threads. */
|
|
|
|
list *clients_pending_io_write; /* List of clients with pending write to be process by I/O threads. */
|
2024-06-07 14:21:33 -07:00
|
|
|
list *replicas, *monitors; /* List of replicas and MONITORs */
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
rax *replicas_waiting_psync; /* Radix tree for tracking replicas awaiting partial synchronization.
|
|
|
|
* Key: RDB client ID
|
|
|
|
* Value: RDB client object
|
|
|
|
* This structure holds dual-channel sync replicas from the start of their
|
|
|
|
* RDB transfer until their main channel establishes partial synchronization. */
|
2024-05-22 23:24:12 -07:00
|
|
|
client *current_client; /* The client that triggered the command execution (External or AOF). */
|
|
|
|
client *executing_client; /* The client executing the current command (possibly script or module). */
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
|
Add reply_schema to command json files (internal for now) (#10273)
Work in progress towards implementing a reply schema as part of COMMAND DOCS, see #9845
Since ironing the details of the reply schema of each and every command can take a long time, we
would like to merge this PR when the infrastructure is ready, and let this mature in the unstable branch.
Meanwhile the changes of this PR are internal, they are part of the repo, but do not affect the produced build.
### Background
In #9656 we add a lot of information about Redis commands, but we are missing information about the replies
### Motivation
1. Documentation. This is the primary goal.
2. It should be possible, based on the output of COMMAND, to be able to generate client code in typed
languages. In order to do that, we need Redis to tell us, in detail, what each reply looks like.
3. We would like to build a fuzzer that verifies the reply structure (for now we use the existing
testsuite, see the "Testing" section)
### Schema
The idea is to supply some sort of schema for the various replies of each command.
The schema will describe the conceptual structure of the reply (for generated clients), as defined in RESP3.
Note that the reply structure itself may change, depending on the arguments (e.g. `XINFO STREAM`, with
and without the `FULL` modifier)
We decided to use the standard json-schema (see https://json-schema.org/) as the reply-schema.
Example for `BZPOPMIN`:
```
"reply_schema": {
"oneOf": [
{
"description": "Timeout reached and no elements were popped.",
"type": "null"
},
{
"description": "The keyname, popped member, and its score.",
"type": "array",
"minItems": 3,
"maxItems": 3,
"items": [
{
"description": "Keyname",
"type": "string"
},
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
]
}
```
#### Notes
1. It is ok that some commands' reply structure depends on the arguments and it's the caller's responsibility
to know which is the relevant one. this comes after looking at other request-reply systems like OpenAPI,
where the reply schema can also be oneOf and the caller is responsible to know which schema is the relevant one.
2. The reply schemas will describe RESP3 replies only. even though RESP3 is structured, we want to use reply
schema for documentation (and possibly to create a fuzzer that validates the replies)
3. For documentation, the description field will include an explanation of the scenario in which the reply is sent,
including any relation to arguments. for example, for `ZRANGE`'s two schemas we will need to state that one
is with `WITHSCORES` and the other is without.
4. For documentation, there will be another optional field "notes" in which we will add a short description of
the representation in RESP2, in case it's not trivial (RESP3's `ZRANGE`'s nested array vs. RESP2's flat
array, for example)
Given the above:
1. We can generate the "return" section of all commands in [redis-doc](https://redis.io/commands/)
(given that "description" and "notes" are comprehensive enough)
2. We can generate a client in a strongly typed language (but the return type could be a conceptual
`union` and the caller needs to know which schema is relevant). see the section below for RESP2 support.
3. We can create a fuzzer for RESP3.
### Limitations (because we are using the standard json-schema)
The problem is that Redis' replies are more diverse than what the json format allows. This means that,
when we convert the reply to a json (in order to validate the schema against it), we lose information (see
the "Testing" section below).
The other option would have been to extend the standard json-schema (and json format) to include stuff
like sets, bulk-strings, error-string, etc. but that would mean also extending the schema-validator - and that
seemed like too much work, so we decided to compromise.
Examples:
1. We cannot tell the difference between an "array" and a "set"
2. We cannot tell the difference between simple-string and bulk-string
3. we cannot verify true uniqueness of items in commands like ZRANGE: json-schema doesn't cover the
case of two identical members with different scores (e.g. `[["m1",6],["m1",7]]`) because `uniqueItems`
compares (member,score) tuples and not just the member name.
### Testing
This commit includes some changes inside Redis in order to verify the schemas (existing and future ones)
are indeed correct (i.e. describe the actual response of Redis).
To do that, we added a debugging feature to Redis that causes it to produce a log of all the commands
it executed and their replies.
For that, Redis needs to be compiled with `-DLOG_REQ_RES` and run with
`--reg-res-logfile <file> --client-default-resp 3` (the testsuite already does that if you run it with
`--log-req-res --force-resp3`)
You should run the testsuite with the above args (and `--dont-clean`) in order to make Redis generate
`.reqres` files (same dir as the `stdout` files) which contain request-response pairs.
These files are later on processed by `./utils/req-res-log-validator.py` which does:
1. Goes over req-res files, generated by redis-servers, spawned by the testsuite (see logreqres.c)
2. For each request-response pair, it validates the response against the request's reply_schema
(obtained from the extended COMMAND DOCS)
5. In order to get good coverage of the Redis commands, and all their different replies, we chose to use
the existing redis test suite, rather than attempt to write a fuzzer.
#### Notes about RESP2
1. We will not be able to use the testing tool to verify RESP2 replies (we are ok with that, it's time to
accept RESP3 as the future RESP)
2. Since the majority of the test suite is using RESP2, and we want the server to reply with RESP3
so that we can validate it, we will need to know how to convert the actual reply to the one expected.
- number and boolean are always strings in RESP2 so the conversion is easy
- objects (maps) are always a flat array in RESP2
- others (nested array in RESP3's `ZRANGE` and others) will need some special per-command
handling (so the client will not be totally auto-generated)
Example for ZRANGE:
```
"reply_schema": {
"anyOf": [
{
"description": "A list of member elements",
"type": "array",
"uniqueItems": true,
"items": {
"type": "string"
}
},
{
"description": "Members and their scores. Returned in case `WITHSCORES` was used.",
"notes": "In RESP2 this is returned as a flat array",
"type": "array",
"uniqueItems": true,
"items": {
"type": "array",
"minItems": 2,
"maxItems": 2,
"items": [
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
}
]
}
```
### Other changes
1. Some tests that behave differently depending on the RESP are now being tested for both RESP,
regardless of the special log-req-res mode ("Pub/Sub PING" for example)
2. Update the history field of CLIENT LIST
3. Added basic tests for commands that were not covered at all by the testsuite
### TODO
- [x] (maybe a different PR) add a "condition" field to anyOf/oneOf schemas that refers to args. e.g.
when `SET` return NULL, the condition is `arguments.get||arguments.condition`, for `OK` the condition
is `!arguments.get`, and for `string` the condition is `arguments.get` - https://github.com/redis/redis/issues/11896
- [x] (maybe a different PR) also run `runtest-cluster` in the req-res logging mode
- [x] add the new tests to GH actions (i.e. compile with `-DLOG_REQ_RES`, run the tests, and run the validator)
- [x] (maybe a different PR) figure out a way to warn about (sub)schemas that are uncovered by the output
of the tests - https://github.com/redis/redis/issues/11897
- [x] (probably a separate PR) add all missing schemas
- [x] check why "SDOWN is triggered by misconfigured instance replying with errors" fails with --log-req-res
- [x] move the response transformers to their own file (run both regular, cluster, and sentinel tests - need to
fight with the tcl including mechanism a bit)
- [x] issue: module API - https://github.com/redis/redis/issues/11898
- [x] (probably a separate PR): improve schemas: add `required` to `object`s - https://github.com/redis/redis/issues/11899
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Hanna Fadida <hanna.fadida@redislabs.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Shaya Potter <shaya@redislabs.com>
2023-03-11 09:14:16 +01:00
|
|
|
#ifdef LOG_REQ_RES
|
2024-05-22 23:24:12 -07:00
|
|
|
char *req_res_logfile; /* Path of log file for logging all requests and their replies. If NULL, no logging will be
|
|
|
|
performed */
|
Add reply_schema to command json files (internal for now) (#10273)
Work in progress towards implementing a reply schema as part of COMMAND DOCS, see #9845
Since ironing the details of the reply schema of each and every command can take a long time, we
would like to merge this PR when the infrastructure is ready, and let this mature in the unstable branch.
Meanwhile the changes of this PR are internal, they are part of the repo, but do not affect the produced build.
### Background
In #9656 we add a lot of information about Redis commands, but we are missing information about the replies
### Motivation
1. Documentation. This is the primary goal.
2. It should be possible, based on the output of COMMAND, to be able to generate client code in typed
languages. In order to do that, we need Redis to tell us, in detail, what each reply looks like.
3. We would like to build a fuzzer that verifies the reply structure (for now we use the existing
testsuite, see the "Testing" section)
### Schema
The idea is to supply some sort of schema for the various replies of each command.
The schema will describe the conceptual structure of the reply (for generated clients), as defined in RESP3.
Note that the reply structure itself may change, depending on the arguments (e.g. `XINFO STREAM`, with
and without the `FULL` modifier)
We decided to use the standard json-schema (see https://json-schema.org/) as the reply-schema.
Example for `BZPOPMIN`:
```
"reply_schema": {
"oneOf": [
{
"description": "Timeout reached and no elements were popped.",
"type": "null"
},
{
"description": "The keyname, popped member, and its score.",
"type": "array",
"minItems": 3,
"maxItems": 3,
"items": [
{
"description": "Keyname",
"type": "string"
},
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
]
}
```
#### Notes
1. It is ok that some commands' reply structure depends on the arguments and it's the caller's responsibility
to know which is the relevant one. this comes after looking at other request-reply systems like OpenAPI,
where the reply schema can also be oneOf and the caller is responsible to know which schema is the relevant one.
2. The reply schemas will describe RESP3 replies only. even though RESP3 is structured, we want to use reply
schema for documentation (and possibly to create a fuzzer that validates the replies)
3. For documentation, the description field will include an explanation of the scenario in which the reply is sent,
including any relation to arguments. for example, for `ZRANGE`'s two schemas we will need to state that one
is with `WITHSCORES` and the other is without.
4. For documentation, there will be another optional field "notes" in which we will add a short description of
the representation in RESP2, in case it's not trivial (RESP3's `ZRANGE`'s nested array vs. RESP2's flat
array, for example)
Given the above:
1. We can generate the "return" section of all commands in [redis-doc](https://redis.io/commands/)
(given that "description" and "notes" are comprehensive enough)
2. We can generate a client in a strongly typed language (but the return type could be a conceptual
`union` and the caller needs to know which schema is relevant). see the section below for RESP2 support.
3. We can create a fuzzer for RESP3.
### Limitations (because we are using the standard json-schema)
The problem is that Redis' replies are more diverse than what the json format allows. This means that,
when we convert the reply to a json (in order to validate the schema against it), we lose information (see
the "Testing" section below).
The other option would have been to extend the standard json-schema (and json format) to include stuff
like sets, bulk-strings, error-string, etc. but that would mean also extending the schema-validator - and that
seemed like too much work, so we decided to compromise.
Examples:
1. We cannot tell the difference between an "array" and a "set"
2. We cannot tell the difference between simple-string and bulk-string
3. we cannot verify true uniqueness of items in commands like ZRANGE: json-schema doesn't cover the
case of two identical members with different scores (e.g. `[["m1",6],["m1",7]]`) because `uniqueItems`
compares (member,score) tuples and not just the member name.
### Testing
This commit includes some changes inside Redis in order to verify the schemas (existing and future ones)
are indeed correct (i.e. describe the actual response of Redis).
To do that, we added a debugging feature to Redis that causes it to produce a log of all the commands
it executed and their replies.
For that, Redis needs to be compiled with `-DLOG_REQ_RES` and run with
`--reg-res-logfile <file> --client-default-resp 3` (the testsuite already does that if you run it with
`--log-req-res --force-resp3`)
You should run the testsuite with the above args (and `--dont-clean`) in order to make Redis generate
`.reqres` files (same dir as the `stdout` files) which contain request-response pairs.
These files are later on processed by `./utils/req-res-log-validator.py` which does:
1. Goes over req-res files, generated by redis-servers, spawned by the testsuite (see logreqres.c)
2. For each request-response pair, it validates the response against the request's reply_schema
(obtained from the extended COMMAND DOCS)
5. In order to get good coverage of the Redis commands, and all their different replies, we chose to use
the existing redis test suite, rather than attempt to write a fuzzer.
#### Notes about RESP2
1. We will not be able to use the testing tool to verify RESP2 replies (we are ok with that, it's time to
accept RESP3 as the future RESP)
2. Since the majority of the test suite is using RESP2, and we want the server to reply with RESP3
so that we can validate it, we will need to know how to convert the actual reply to the one expected.
- number and boolean are always strings in RESP2 so the conversion is easy
- objects (maps) are always a flat array in RESP2
- others (nested array in RESP3's `ZRANGE` and others) will need some special per-command
handling (so the client will not be totally auto-generated)
Example for ZRANGE:
```
"reply_schema": {
"anyOf": [
{
"description": "A list of member elements",
"type": "array",
"uniqueItems": true,
"items": {
"type": "string"
}
},
{
"description": "Members and their scores. Returned in case `WITHSCORES` was used.",
"notes": "In RESP2 this is returned as a flat array",
"type": "array",
"uniqueItems": true,
"items": {
"type": "array",
"minItems": 2,
"maxItems": 2,
"items": [
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
}
]
}
```
### Other changes
1. Some tests that behave differently depending on the RESP are now being tested for both RESP,
regardless of the special log-req-res mode ("Pub/Sub PING" for example)
2. Update the history field of CLIENT LIST
3. Added basic tests for commands that were not covered at all by the testsuite
### TODO
- [x] (maybe a different PR) add a "condition" field to anyOf/oneOf schemas that refers to args. e.g.
when `SET` return NULL, the condition is `arguments.get||arguments.condition`, for `OK` the condition
is `!arguments.get`, and for `string` the condition is `arguments.get` - https://github.com/redis/redis/issues/11896
- [x] (maybe a different PR) also run `runtest-cluster` in the req-res logging mode
- [x] add the new tests to GH actions (i.e. compile with `-DLOG_REQ_RES`, run the tests, and run the validator)
- [x] (maybe a different PR) figure out a way to warn about (sub)schemas that are uncovered by the output
of the tests - https://github.com/redis/redis/issues/11897
- [x] (probably a separate PR) add all missing schemas
- [x] check why "SDOWN is triggered by misconfigured instance replying with errors" fails with --log-req-res
- [x] move the response transformers to their own file (run both regular, cluster, and sentinel tests - need to
fight with the tcl including mechanism a bit)
- [x] issue: module API - https://github.com/redis/redis/issues/11898
- [x] (probably a separate PR): improve schemas: add `required` to `object`s - https://github.com/redis/redis/issues/11899
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Hanna Fadida <hanna.fadida@redislabs.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Shaya Potter <shaya@redislabs.com>
2023-03-11 09:14:16 +01:00
|
|
|
unsigned int client_default_resp;
|
|
|
|
#endif
|
|
|
|
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
/* Stuff for client mem eviction */
|
2024-05-22 23:24:12 -07:00
|
|
|
clientMemUsageBucket *client_mem_usage_buckets;
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
|
2020-03-26 16:05:20 +01:00
|
|
|
rax *clients_timeout_table; /* Radix tree for blocked clients timeouts. */
|
2022-12-20 13:21:50 +05:30
|
|
|
int execution_nesting; /* Execution nesting level.
|
|
|
|
* e.g. call(), async module stuff (timers, events, etc.),
|
|
|
|
* cron stuff (active expire, eviction) */
|
2018-06-27 13:26:01 +02:00
|
|
|
rax *clients_index; /* Active clients dictionary by client ID. */
|
2024-05-22 23:24:12 -07:00
|
|
|
uint32_t paused_actions; /* Bitmask of actions that are currently paused */
|
|
|
|
list *postponed_clients; /* List of postponed clients */
|
2022-10-27 11:57:04 +03:00
|
|
|
pause_event client_pause_per_purpose[NUM_PAUSE_PURPOSES];
|
2024-05-28 09:27:51 -07:00
|
|
|
char neterr[ANET_ERR_LEN]; /* Error buffer for anet.c */
|
|
|
|
dict *migrate_cached_sockets; /* MIGRATE cached sockets */
|
|
|
|
_Atomic uint64_t next_client_id; /* Next client unique ID. Incremental. */
|
|
|
|
int protected_mode; /* Don't accept external connections. */
|
|
|
|
int io_threads_num; /* Number of IO threads to use. */
|
|
|
|
int io_threads_do_reads; /* Read and parse from IO threads? */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
int active_io_threads_num; /* Current number of active IO threads, includes main thread. */
|
|
|
|
int events_per_io_thread; /* Number of events on the event loop to trigger IO threads activation. */
|
2020-05-14 10:02:57 +02:00
|
|
|
long long events_processed_while_blocked; /* processEventsWhileBlocked() */
|
2024-05-22 23:24:12 -07:00
|
|
|
int enable_protected_configs; /* Enable the modification of protected configs, see PROTECTED_ACTION_ALLOWED_* */
|
|
|
|
int enable_debug_cmd; /* Enable DEBUG commands, see PROTECTED_ACTION_ALLOWED_* */
|
|
|
|
int enable_module_cmd; /* Enable MODULE commands, see PROTECTED_ACTION_ALLOWED_* */
|
2024-05-31 22:50:08 -07:00
|
|
|
int enable_debug_assert; /* Enable debug asserts */
|
2017-10-24 08:35:05 +02:00
|
|
|
|
2010-11-08 11:52:03 +01:00
|
|
|
/* RDB / AOF loading information */
|
2024-05-22 23:24:12 -07:00
|
|
|
volatile sig_atomic_t loading; /* We are loading data from disk if true */
|
Replica keep serving data during repl-diskless-load=swapdb for better availability (#9323)
For diskless replication in swapdb mode, considering we already spend replica memory
having a backup of current db to restore in case of failure, we can have the following benefits
by instead swapping database only in case we succeeded in transferring db from master:
- Avoid `LOADING` response during failed and successful synchronization for cases where the
replica is already up and running with data.
- Faster total time of diskless replication, because now we're moving from Transfer + Flush + Load
time to Transfer + Load only. Flushing the tempDb is done asynchronously after swapping.
- This could be implemented also for disk replication with similar benefits if consumers are willing
to spend the extra memory usage.
General notes:
- The concept of `backupDb` becomes `tempDb` for clarity.
- Async loading mode will only kick in if the replica is syncing from a master that has the same
repl-id the one it had before. i.e. the data it's getting belongs to a different time of the same timeline.
- New property in INFO: `async_loading` to differentiate from the blocking loading
- Slot to Key mapping is now a field of `redisDb` as it's more natural to access it from both server.db
and the tempDb that is passed around.
- Because this is affecting replicas only, we assume that if they are not readonly and write commands
during replication, they are lost after SYNC same way as before, but we're still denying CONFIG SET
here anyways to avoid complications.
Considerations for review:
- We have many cases where server.loading flag is used and even though I tried my best, there may
be cases where async_loading should be checked as well and cases where it shouldn't (would require
very good understanding of whole code)
- Several places that had different behavior depending on the loading flag where actually meant to just
handle commands coming from the AOF client differently than ones coming from real clients, changed
to check CLIENT_ID_AOF instead.
**Additional for Release Notes**
- Bugfix - server.dirty was not incremented for any kind of diskless replication, as effect it wouldn't
contribute on triggering next database SAVE
- New flag for RM_GetContextFlags module API: REDISMODULE_CTX_FLAGS_ASYNC_LOADING
- Deprecated RedisModuleEvent_ReplBackup. Starting from Redis 7.0, we don't fire this event.
Instead, we have the new RedisModuleEvent_ReplAsyncLoad holding 3 sub-events: STARTED,
ABORTED and COMPLETED.
- New module flag REDISMODULE_OPTIONS_HANDLE_REPL_ASYNC_LOAD for RedisModule_SetModuleOptions
to allow modules to declare they support the diskless replication with async loading (when absent, we fall
back to disk-based loading).
Co-authored-by: Eduardo Semprebon <edus@saxobank.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-11-04 09:46:50 +01:00
|
|
|
volatile sig_atomic_t async_loading; /* We are loading data without blocking the db being served */
|
2010-11-08 11:52:03 +01:00
|
|
|
off_t loading_total_bytes;
|
2020-11-05 11:46:16 +02:00
|
|
|
off_t loading_rdb_used_mem;
|
2010-11-08 11:52:03 +01:00
|
|
|
off_t loading_loaded_bytes;
|
|
|
|
time_t loading_start_time;
|
2012-12-12 15:59:22 +02:00
|
|
|
off_t loading_process_events_interval_bytes;
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
time_t loading_process_events_interval_ms;
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Fields used only for stats */
|
2024-05-22 23:24:12 -07:00
|
|
|
time_t stat_starttime; /* Server start time */
|
|
|
|
long long stat_numcommands; /* Number of processed commands */
|
|
|
|
long long stat_numconnections; /* Number of connections received */
|
|
|
|
long long stat_expiredkeys; /* Number of expired keys */
|
|
|
|
double stat_expired_stale_perc; /* Percentage of keys probably expired */
|
2021-06-10 20:39:33 +08:00
|
|
|
long long stat_expired_time_cap_reached_count; /* Early expire cycle stops.*/
|
2024-05-28 09:27:51 -07:00
|
|
|
long long stat_expire_cycle_time_used; /* Cumulative microseconds used. */
|
|
|
|
long long stat_evictedkeys; /* Number of evicted keys (maxmemory) */
|
|
|
|
long long stat_evictedclients; /* Number of evicted clients */
|
|
|
|
long long stat_evictedscripts; /* Number of evicted lua scripts. */
|
|
|
|
long long stat_total_eviction_exceeded_time; /* Total time over the memory limit, unit us */
|
|
|
|
monotime stat_last_eviction_exceeded_time; /* Timestamp of current eviction start, unit us */
|
|
|
|
long long stat_keyspace_hits; /* Number of successful lookups of keys */
|
|
|
|
long long stat_keyspace_misses; /* Number of failed lookups of keys */
|
|
|
|
long long stat_active_defrag_hits; /* number of allocations moved */
|
|
|
|
long long stat_active_defrag_misses; /* number of allocations scanned but not moved */
|
|
|
|
long long stat_active_defrag_key_hits; /* number of keys with moved allocations */
|
|
|
|
long long stat_active_defrag_key_misses; /* number of keys scanned and not moved */
|
|
|
|
long long stat_active_defrag_scanned; /* number of dictEntries scanned */
|
|
|
|
long long stat_total_active_defrag_time; /* Total time memory fragmentation over the limit, unit us */
|
|
|
|
monotime stat_last_active_defrag_time; /* Timestamp of current active defrag start */
|
|
|
|
size_t stat_peak_memory; /* Max used memory record */
|
|
|
|
long long stat_aof_rewrites; /* number of aof file rewrites performed */
|
|
|
|
long long stat_aofrw_consecutive_failures; /* The number of consecutive failures of aofrw */
|
|
|
|
long long stat_rdb_saves; /* number of rdb saves performed */
|
|
|
|
long long stat_fork_time; /* Time needed to perform latest fork() */
|
|
|
|
double stat_fork_rate; /* Fork rate in GB/sec. */
|
|
|
|
long long stat_total_forks; /* Total count of fork. */
|
|
|
|
long long stat_rejected_conn; /* Clients rejected because of maxclients */
|
2024-06-07 14:21:33 -07:00
|
|
|
long long stat_sync_full; /* Number of full resyncs with replicas. */
|
2024-05-28 09:27:51 -07:00
|
|
|
long long stat_sync_partial_ok; /* Number of accepted PSYNC requests. */
|
|
|
|
long long stat_sync_partial_err; /* Number of unaccepted PSYNC requests. */
|
|
|
|
list *slowlog; /* SLOWLOG list of commands */
|
|
|
|
long long slowlog_entry_id; /* SLOWLOG current entry ID */
|
|
|
|
long long slowlog_log_slower_than; /* SLOWLOG time limit (to get logged) */
|
|
|
|
unsigned long slowlog_max_len; /* SLOWLOG max number of items logged */
|
|
|
|
struct malloc_stats cron_malloc_stats; /* sampled in serverCron(). */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
long long stat_net_input_bytes; /* Bytes read from network. */
|
|
|
|
long long stat_net_output_bytes; /* Bytes written to network. */
|
|
|
|
long long stat_net_repl_input_bytes; /* Bytes read during replication, added to stat_net_input_bytes in 'info'. */
|
|
|
|
/* Bytes written during replication, added to stat_net_output_bytes in 'info'. */
|
|
|
|
long long stat_net_repl_output_bytes;
|
|
|
|
size_t stat_current_cow_peak; /* Peak size of copy on write bytes. */
|
|
|
|
size_t stat_current_cow_bytes; /* Copy on write bytes while child is active. */
|
|
|
|
monotime stat_current_cow_updated; /* Last update time of stat_current_cow_bytes */
|
2024-05-28 09:27:51 -07:00
|
|
|
size_t stat_current_save_keys_processed; /* Processed keys while child is active. */
|
|
|
|
size_t stat_current_save_keys_total; /* Number of keys when child started. */
|
|
|
|
size_t stat_rdb_cow_bytes; /* Copy on write bytes during RDB saving. */
|
|
|
|
size_t stat_aof_cow_bytes; /* Copy on write bytes during AOF rewrite. */
|
|
|
|
size_t stat_module_cow_bytes; /* Copy on write bytes during module fork. */
|
|
|
|
double stat_module_progress; /* Module save progress. */
|
|
|
|
size_t stat_clients_type_memory[CLIENT_TYPE_COUNT]; /* Mem usage by type */
|
|
|
|
size_t stat_cluster_links_memory; /* Mem usage by cluster links */
|
|
|
|
long long
|
2024-06-07 14:21:33 -07:00
|
|
|
stat_unexpected_error_replies; /* Number of unexpected (aof-loading, replica to primary, etc.) error replies */
|
2020-12-31 14:53:43 +00:00
|
|
|
long long stat_total_error_replies; /* Total number of issued error replies ( command + rejected errors ) */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
long long stat_dump_payload_sanitizations; /* Number deep dump payloads integrity validations. */
|
|
|
|
long long stat_io_reads_processed; /* Number of read events processed by IO threads */
|
|
|
|
long long stat_io_writes_processed; /* Number of write events processed by IO threads */
|
Io thread work offload (#763)
### IO-Threads Work Offloading
This PR is the 2nd of 3 PRs intended to achieve the goal of 1M requests
per second.
(1st PR: https://github.com/valkey-io/valkey/pull/758)
This PR offloads additional work to the I/O threads, beyond the current
read-parse/write operations, to better utilize the I/O threads and
reduce the load on the main thread.
It contains the following 3 commits:
### Poll Offload
Currently, the main thread is responsible for executing the poll-wait
system call, while the IO threads wait for tasks from the main thread.
The poll-wait operation is expensive and can consume up to 30% of the
main thread's time. We could have let the IO threads do the poll-wait by
themselves, with each thread listening to some of the clients and
notifying the main thread when a client's command is ready to execute.
However, the current approach, where the main thread listens for events
from the network, has several benefits. The main thread remains in
charge, allowing it to know the state of each client
(idle/read/write/close) at any given time. Additionally, it makes the
threads flexible, enabling us to drain an IO thread's job queue and stop
a thread when the load is light without modifying the event loop and
moving its clients to a different IO thread. Furthermore, with this
approach, the IO threads don't need to wait for both messages from the
network and from the main thread instead, the threads wait only for
tasks from the main thread.
To enjoy the benefits of both the main thread remaining in charge and
the poll being offloaded, we propose offloading the poll-wait as a
single-time, non-blocking job to one of the IO threads. The IO thread
will perform a poll-wait non-blocking call while the main thread
processes the client commands. Later, in `aeProcessEvents`, instead of
sleeping on the poll, we check for the IO thread's poll-wait results.
The poll-wait will be offloaded in `beforeSleep` only when there are
ready events for the main thread to process. If no events are pending,
the main thread will revert to the current behavior and sleep on the
poll by itself.
**Implementation Details**
A new call back `custompoll` was added to the `aeEventLoop` when not set
to `NULL` the ae will call the `custompoll` callback instead of the
`aeApiPoll`.
When the poll is offloaded we will set the `custompoll` to
`getIOThreadPollResults` and send a poll-job to the thread. the thread
will take a mutex, call a non-blocking (with timeout 0) to `aePoll`
which will populate the fired events array. the IO thread will set the
`server.io_fired_events` to the number of the returning `numevents`,
later the main-thread in `custompoll` will return the
`server.io_fired_events` and will set the `customPoll` back to `NULL`.
To ensure thread safety when accessing server.el, all functions that
modify the eventloop events were wrapped with a mutex to ensure mutual
exclusion when modifying the events.
### Command Lookup Offload
As the IO thread parses the command from the client's Querybuf, it can
perform a command lookup in the commands dictionary, which can consume
up to ~5% of the main-thread runtime.
**Implementation details**
The IO thread will store the looked-up command in the client's new field
`io_parsed_cmd` field. We can't use `c->cmd` for that since we use
`c->cmd `to check if a command was reprocessed or not.
To ensure thread safety when accessing the command dictionary, we make
sure the main thread isn't changing the dictionary while IO threads are
accessing it. This is accomplished by introducing a new flag called
`no_incremental_rehash` for the `dictType` commands. When performing
`dictResize`, we will rehash the entire dictionary in place rather than
deferring the process.
### Free Offload
Since the command arguments are allocated by the I/O thread, it would be
beneficial if they were also freed by the same thread. If the main
thread frees objects allocated by the I/O thread, two issues arise:
1. During the freeing process, the main thread needs to access the SDS
pointed to by the object to get its length.
2. With Jemalloc, each thread manages thread local pool (`tcache`) of
buffers for quick reallocation without accessing the arena. If the main
thread constantly frees objects allocated by other threads, those
threads will have to frequently access the shared arena to obtain new
memory allocations
**Implementation Details**
When freeing the client's argv, we will send the argv array to the
thread that allocated it. The thread will be identified by the client
ID. When freeing an object during `dbOverwrite`, we will offload the
object free as well. We will extend this to offload the free during
`dbDelete` in a future PR, as its effects on defrag/memory evictions
need to be studied.
---------
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-19 05:21:45 +03:00
|
|
|
long long stat_io_freed_objects; /* Number of objects freed by IO threads */
|
|
|
|
long long stat_poll_processed_by_io_threads; /* Total number of poll jobs processed by IO */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
long long stat_total_reads_processed; /* Total number of read events processed */
|
|
|
|
long long stat_total_writes_processed; /* Total number of write events processed */
|
|
|
|
long long stat_client_qbuf_limit_disconnections; /* Total number of clients reached query buf length limit */
|
2024-05-28 09:27:51 -07:00
|
|
|
long long stat_client_outbuf_limit_disconnections; /* Total number of clients reached output buf length limit */
|
2014-12-03 12:06:54 +01:00
|
|
|
/* The following two are used to track instantaneous metrics, like
|
|
|
|
* number of operations per second, network traffic. */
|
|
|
|
struct {
|
2023-05-13 01:13:15 +08:00
|
|
|
long long last_sample_base; /* The divisor of last sample window */
|
|
|
|
long long last_sample_value; /* The dividend of last sample window */
|
2015-07-27 09:41:48 +02:00
|
|
|
long long samples[STATS_METRIC_SAMPLES];
|
2014-12-03 12:06:54 +01:00
|
|
|
int idx;
|
2015-07-27 09:41:48 +02:00
|
|
|
} inst_metric[STATS_METRIC_COUNT];
|
2022-02-22 11:19:38 +02:00
|
|
|
long long stat_reply_buffer_shrinks; /* Total number of output buffer shrinks */
|
|
|
|
long long stat_reply_buffer_expands; /* Total number of output buffer expands */
|
2023-05-13 01:13:15 +08:00
|
|
|
monotime el_start;
|
|
|
|
/* The following two are used to record the max number of commands executed in one eventloop.
|
|
|
|
* Note that commands in transactions are also counted. */
|
|
|
|
long long el_cmd_cnt_start;
|
|
|
|
long long el_cmd_cnt_max;
|
|
|
|
/* The sum of active-expire, active-defrag and all other tasks done by cron and beforeSleep,
|
|
|
|
but excluding read, write and AOF, which are counted by other sets of metrics. */
|
2024-05-22 23:24:12 -07:00
|
|
|
monotime el_cron_duration;
|
2023-05-13 01:13:15 +08:00
|
|
|
durationStats duration_stats[EL_DURATION_TYPE_NUM];
|
2022-02-22 11:19:38 +02:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Configuration */
|
2024-05-22 23:24:12 -07:00
|
|
|
int verbosity; /* Loglevel verbosity */
|
|
|
|
int maxidletime; /* Client timeout in seconds */
|
|
|
|
int tcpkeepalive; /* Set SO_KEEPALIVE if non-zero. */
|
|
|
|
int active_expire_enabled; /* Can be disabled for testing purposes. */
|
|
|
|
int active_expire_effort; /* From 1 (default) to 10, active effort. */
|
|
|
|
int lazy_expire_disabled; /* If > 0, don't trigger lazy expire */
|
2016-12-30 03:37:52 +02:00
|
|
|
int active_defrag_enabled;
|
2024-05-22 23:24:12 -07:00
|
|
|
int sanitize_dump_payload; /* Enables deep sanitization for ziplist and listpack in RDB and RESTORE. */
|
|
|
|
int skip_checksum_validation; /* Disable checksum validation for RDB and RESTORE payload. */
|
|
|
|
int jemalloc_bg_thread; /* Enable jemalloc background thread */
|
2024-02-06 19:39:07 +08:00
|
|
|
int active_defrag_configuration_changed; /* defrag configuration has been changed and need to reconsider
|
|
|
|
* active_defrag_running in computeDefragCycles. */
|
2024-05-22 23:24:12 -07:00
|
|
|
size_t active_defrag_ignore_bytes; /* minimum amount of fragmentation waste to start active defrag */
|
|
|
|
int active_defrag_threshold_lower; /* minimum percentage of fragmentation to start active defrag */
|
|
|
|
int active_defrag_threshold_upper; /* maximum percentage of fragmentation at which we use maximum effort */
|
|
|
|
int active_defrag_cycle_min; /* minimal effort for defrag in CPU percentage */
|
|
|
|
int active_defrag_cycle_max; /* maximal effort for defrag in CPU percentage */
|
|
|
|
unsigned long active_defrag_max_scan_fields; /* maximum number of fields of set/hash/zset/list to process from
|
|
|
|
within the main dict scan */
|
|
|
|
size_t client_max_querybuf_len; /* Limit for client query buffer length */
|
|
|
|
int dbnum; /* Total number of configured DBs */
|
|
|
|
int supervised; /* 1 if supervised, 0 otherwise. */
|
|
|
|
int supervised_mode; /* See SUPERVISED_* */
|
|
|
|
int daemonize; /* True if running as a daemon */
|
|
|
|
int set_proc_title; /* True if change proc title */
|
|
|
|
char *proc_title_template; /* Process title template format */
|
2015-07-28 16:58:04 +02:00
|
|
|
clientBufferLimitsConfig client_obuf_limits[CLIENT_TYPE_OBUF_COUNT];
|
2024-05-22 23:24:12 -07:00
|
|
|
int extended_redis_compat; /* True if extended Redis OSS compatibility is enabled */
|
|
|
|
int pause_cron; /* Don't run cron tasks (debug) */
|
|
|
|
int dict_resizing; /* Whether to allow main dict and expired dict to be resized (debug) */
|
|
|
|
int latency_tracking_enabled; /* 1 if extended latency tracking is enabled, 0 otherwise. */
|
Added INFO LATENCYSTATS section: latency by percentile distribution/latency by cumulative distribution of latencies (#9462)
# Short description
The Redis extended latency stats track per command latencies and enables:
- exporting the per-command percentile distribution via the `INFO LATENCYSTATS` command.
**( percentile distribution is not mergeable between cluster nodes ).**
- exporting the per-command cumulative latency distributions via the `LATENCY HISTOGRAM` command.
Using the cumulative distribution of latencies we can merge several stats from different cluster nodes
to calculate aggregate metrics .
By default, the extended latency monitoring is enabled since the overhead of keeping track of the
command latency is very small.
If you don't want to track extended latency metrics, you can easily disable it at runtime using the command:
- `CONFIG SET latency-tracking no`
By default, the exported latency percentiles are the p50, p99, and p999.
You can alter them at runtime using the command:
- `CONFIG SET latency-tracking-info-percentiles "0.0 50.0 100.0"`
## Some details:
- The total size per histogram should sit around 40 KiB. We only allocate those 40KiB when a command
was called for the first time.
- With regards to the WRITE overhead As seen below, there is no measurable overhead on the achievable
ops/sec or full latency spectrum on the client. Including also the measured redis-benchmark for unstable
vs this branch.
- We track from 1 nanosecond to 1 second ( everything above 1 second is considered +Inf )
## `INFO LATENCYSTATS` exposition format
- Format: `latency_percentiles_usec_<CMDNAME>:p0=XX,p50....`
## `LATENCY HISTOGRAM [command ...]` exposition format
Return a cumulative distribution of latencies in the format of a histogram for the specified command names.
The histogram is composed of a map of time buckets:
- Each representing a latency range, between 1 nanosecond and roughly 1 second.
- Each bucket covers twice the previous bucket's range.
- Empty buckets are not printed.
- Everything above 1 sec is considered +Inf.
- At max there will be log2(1000000000)=30 buckets
We reply a map for each command in the format:
`<command name> : { `calls`: <total command calls> , `histogram` : { <bucket 1> : latency , < bucket 2> : latency, ... } }`
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-05 12:01:05 +00:00
|
|
|
double *latency_tracking_info_percentiles; /* Extended latency tracking info output percentile list configuration. */
|
|
|
|
int latency_tracking_info_percentiles_len;
|
2024-05-22 23:24:12 -07:00
|
|
|
unsigned int max_new_tls_conns_per_cycle; /* The maximum number of tls connections that will be accepted during each
|
|
|
|
invocation of the event loop. */
|
|
|
|
unsigned int max_new_conns_per_cycle; /* The maximum number of tcp connections that will be accepted during each
|
|
|
|
invocation of the event loop. */
|
2011-12-21 10:05:32 +01:00
|
|
|
/* AOF persistence */
|
2024-05-28 09:27:51 -07:00
|
|
|
int aof_enabled; /* AOF configuration */
|
|
|
|
int aof_state; /* AOF_(ON|OFF|WAIT_REWRITE) */
|
|
|
|
int aof_fsync; /* Kind of fsync() policy */
|
|
|
|
char *aof_filename; /* Basename of the AOF file and manifest file */
|
|
|
|
char *aof_dirname; /* Name of the AOF directory */
|
|
|
|
int aof_no_fsync_on_rewrite; /* Don't fsync if a rewrite is in prog. */
|
|
|
|
int aof_rewrite_perc; /* Rewrite AOF if % growth is > M and... */
|
|
|
|
off_t aof_rewrite_min_size; /* the AOF file is at least N bytes. */
|
|
|
|
off_t aof_rewrite_base_size; /* AOF size on latest startup or rewrite. */
|
|
|
|
off_t aof_current_size; /* AOF current size (Including BASE + INCRs). */
|
|
|
|
off_t aof_last_incr_size; /* The size of the latest incr AOF. */
|
|
|
|
off_t aof_last_incr_fsync_offset; /* AOF offset which is already requested to be synced to disk.
|
|
|
|
* Compare with the aof_last_incr_size. */
|
|
|
|
int aof_flush_sleep; /* Micros to sleep before flush. (used by tests) */
|
|
|
|
int aof_rewrite_scheduled; /* Rewrite once BGSAVE terminates. */
|
|
|
|
sds aof_buf; /* AOF buffer, written before entering the event loop */
|
|
|
|
int aof_fd; /* File descriptor of currently selected AOF file */
|
|
|
|
int aof_selected_db; /* Currently selected DB in AOF */
|
2024-05-26 17:41:11 +01:00
|
|
|
mstime_t aof_flush_postponed_start; /* mstime of postponed AOF flush */
|
|
|
|
mstime_t aof_last_fsync; /* mstime of last fsync() */
|
2024-05-28 09:27:51 -07:00
|
|
|
time_t aof_rewrite_time_last; /* Time used by last AOF rewrite run. */
|
|
|
|
time_t aof_rewrite_time_start; /* Current AOF rewrite start time. */
|
|
|
|
time_t aof_cur_timestamp; /* Current record timestamp in AOF */
|
|
|
|
int aof_timestamp_enabled; /* Enable record timestamp in AOF */
|
|
|
|
int aof_lastbgrewrite_status; /* C_OK or C_ERR */
|
|
|
|
unsigned long aof_delayed_fsync; /* delayed AOF fsync() counter */
|
|
|
|
int aof_rewrite_incremental_fsync; /* fsync incrementally while aof rewriting? */
|
|
|
|
int rdb_save_incremental_fsync; /* fsync incrementally while rdb saving? */
|
|
|
|
int aof_last_write_status; /* C_OK or C_ERR */
|
|
|
|
int aof_last_write_errno; /* Valid if aof write/fsync status is ERR */
|
|
|
|
int aof_load_truncated; /* Don't stop on unexpected AOF EOF. */
|
|
|
|
int aof_use_rdb_preamble; /* Specify base AOF to use RDB encoding on AOF rewrites. */
|
|
|
|
_Atomic int aof_bio_fsync_status; /* Status of AOF fsync in bio job. */
|
|
|
|
_Atomic int aof_bio_fsync_errno; /* Errno of AOF fsync in bio job. */
|
|
|
|
aofManifest *aof_manifest; /* Used to track AOFs. */
|
|
|
|
int aof_disable_auto_gc; /* If disable automatically deleting HISTORY type AOFs?
|
|
|
|
default no. (for testings). */
|
Implement Multi Part AOF mechanism to avoid AOFRW overheads. (#9788)
Implement Multi-Part AOF mechanism to avoid overheads during AOFRW.
Introducing a folder with multiple AOF files tracked by a manifest file.
The main issues with the the original AOFRW mechanism are:
* buffering of commands that are processed during rewrite (consuming a lot of RAM)
* freezes of the main process when the AOFRW completes to drain the remaining part of the buffer and fsync it.
* double disk IO for the data that arrives during AOFRW (had to be written to both the old and new AOF files)
The main modifications of this PR:
1. Remove the AOF rewrite buffer and related code.
2. Divide the AOF into multiple files, they are classified as two types, one is the the `BASE` type,
it represents the full amount of data (Maybe AOF or RDB format) after each AOFRW, there is only
one `BASE` file at most. The second is `INCR` type, may have more than one. They represent the
incremental commands since the last AOFRW.
3. Use a AOF manifest file to record and manage these AOF files mentioned above.
4. The original configuration of `appendfilename` will be the base part of the new file name, for example:
`appendonly.aof.1.base.rdb` and `appendonly.aof.2.incr.aof`
5. Add manifest-related TCL tests, and modified some existing tests that depend on the `appendfilename`
6. Remove the `aof_rewrite_buffer_length` field in info.
7. Add `aof-disable-auto-gc` configuration. By default we're automatically deleting HISTORY type AOFs.
It also gives users the opportunity to preserve the history AOFs. just for testing use now.
8. Add AOFRW limiting measure. When the AOFRW failures reaches the threshold (3 times now),
we will delay the execution of the next AOFRW by 1 minute. If the next AOFRW also fails, it will be
delayed by 2 minutes. The next is 4, 8, 16, the maximum delay is 60 minutes (1 hour). During the limit
period, we can still use the 'bgrewriteaof' command to execute AOFRW immediately.
9. Support upgrade (load) data from old version redis.
10. Add `appenddirname` configuration, as the directory name of the append only files. All AOF files and
manifest file will be placed in this directory.
11. Only the last AOF file (BASE or INCR) can be truncated. Otherwise redis will exit even if
`aof-load-truncated` is enabled.
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-04 01:14:13 +08:00
|
|
|
|
2011-12-21 10:05:32 +01:00
|
|
|
/* RDB persistence */
|
2024-05-22 23:24:12 -07:00
|
|
|
long long dirty; /* Changes to DB from the last save */
|
|
|
|
long long dirty_before_bgsave; /* Used to restore dirty on failed BGSAVE */
|
|
|
|
long long rdb_last_load_keys_expired; /* number of expired keys when loading RDB */
|
|
|
|
long long rdb_last_load_keys_loaded; /* number of loaded keys when loading RDB */
|
|
|
|
struct saveparam *saveparams; /* Save points array for RDB */
|
|
|
|
int saveparamslen; /* Number of saving points */
|
|
|
|
char *rdb_filename; /* Name of RDB file */
|
|
|
|
int rdb_compression; /* Use compression in RDB? */
|
|
|
|
int rdb_checksum; /* Use RDB checksum? */
|
|
|
|
int rdb_del_sync_files; /* Remove RDB files used only for SYNC if
|
|
|
|
the instance does not use persistence. */
|
|
|
|
time_t lastsave; /* Unix time of last successful save */
|
|
|
|
time_t lastbgsave_try; /* Unix time of last attempted bgsave */
|
|
|
|
time_t rdb_save_time_last; /* Time used by last RDB save run. */
|
|
|
|
time_t rdb_save_time_start; /* Current RDB save start time. */
|
|
|
|
int rdb_bgsave_scheduled; /* BGSAVE when possible if true. */
|
|
|
|
int rdb_child_type; /* Type of save by active child. */
|
|
|
|
int lastbgsave_status; /* C_OK or C_ERR */
|
|
|
|
int stop_writes_on_bgsave_err; /* Don't allow writes if can't BGSAVE */
|
|
|
|
int rdb_pipe_read; /* RDB pipe used to transfer the rdb data */
|
|
|
|
/* to the parent process in diskless repl. */
|
|
|
|
int rdb_child_exit_pipe; /* Used by the diskless parent allow child exit. */
|
|
|
|
connection **rdb_pipe_conns; /* Connections which are currently the */
|
|
|
|
int rdb_pipe_numconns; /* target of diskless rdb fork child. */
|
|
|
|
int rdb_pipe_numconns_writing; /* Number of rdb conns with pending writes. */
|
|
|
|
char *rdb_pipe_buff; /* In diskless replication, this buffer holds data */
|
|
|
|
int rdb_pipe_bufflen; /* that was read from the rdb pipe. */
|
|
|
|
int rdb_key_save_delay; /* Delay in microseconds between keys while
|
|
|
|
* writing aof or rdb. (for testings). negative
|
|
|
|
* value means fractions of microseconds (on average). */
|
|
|
|
int key_load_delay; /* Delay in microseconds between keys while
|
|
|
|
* loading aof or rdb. (for testings). negative
|
|
|
|
* value means fractions of microseconds (on average). */
|
2016-09-19 13:45:20 +02:00
|
|
|
/* Pipe and data structures for child -> parent info sharing. */
|
2024-05-22 23:24:12 -07:00
|
|
|
int child_info_pipe[2]; /* Pipe used to write the child_info_data. */
|
|
|
|
int child_info_nread; /* Num of bytes of the last read from pipe */
|
2012-02-28 18:03:08 +01:00
|
|
|
/* Propagation of commands in AOF / replication */
|
2024-05-22 23:24:12 -07:00
|
|
|
serverOpArray also_propagate; /* Additional command to propagate. */
|
|
|
|
int replication_allowed; /* Are we allowed to replicate? */
|
2011-01-07 18:15:14 +01:00
|
|
|
/* Logging */
|
2024-05-22 23:24:12 -07:00
|
|
|
char *logfile; /* Path of log file */
|
|
|
|
int syslog_enabled; /* Is syslog enabled? */
|
|
|
|
char *syslog_ident; /* Syslog ident */
|
|
|
|
int syslog_facility; /* Syslog facility */
|
|
|
|
int crashlog_enabled; /* Enable signal handler for crashlog.
|
|
|
|
* disable for clean core dumps. */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
int crashed; /* True if the server has crashed, used in catClientInfoString
|
|
|
|
* to indicate that no wait for IO threads is needed. */
|
2024-05-22 23:24:12 -07:00
|
|
|
int memcheck_enabled; /* Enable memory check on crash. */
|
|
|
|
int use_exit_on_panic; /* Use exit() on panic and assert rather than
|
|
|
|
* abort(). useful for Valgrind. */
|
2022-01-02 08:50:15 +01:00
|
|
|
/* Shutdown */
|
2024-05-22 23:24:12 -07:00
|
|
|
int shutdown_timeout; /* Graceful shutdown time limit in seconds. */
|
|
|
|
int shutdown_on_sigint; /* Shutdown flags configured for SIGINT. */
|
|
|
|
int shutdown_on_sigterm; /* Shutdown flags configured for SIGTERM. */
|
2022-01-02 08:50:15 +01:00
|
|
|
|
2024-06-07 14:21:33 -07:00
|
|
|
/* Replication (primary) */
|
2024-05-28 09:27:51 -07:00
|
|
|
char replid[CONFIG_RUN_ID_SIZE + 1]; /* My current replication ID. */
|
2024-06-07 14:21:33 -07:00
|
|
|
char replid2[CONFIG_RUN_ID_SIZE + 1]; /* replid inherited from primary*/
|
|
|
|
long long primary_repl_offset; /* My current replication offset */
|
2024-05-28 09:27:51 -07:00
|
|
|
long long second_replid_offset; /* Accept offsets up to this for replid2. */
|
|
|
|
_Atomic long long fsynced_reploff_pending; /* Largest replication offset to
|
|
|
|
* potentially have been fsynced, applied to
|
|
|
|
fsynced_reploff only when AOF state is AOF_ON
|
|
|
|
(not during the initial rewrite) */
|
|
|
|
long long fsynced_reploff; /* Largest replication offset that has been confirmed to be fsynced */
|
2024-06-07 14:21:33 -07:00
|
|
|
int replicas_eldb; /* Last SELECTed DB in replication output */
|
|
|
|
int repl_ping_replica_period; /* Primary pings the replica every N seconds */
|
2024-05-28 09:27:51 -07:00
|
|
|
replBacklog *repl_backlog; /* Replication backlog for partial syncs */
|
|
|
|
long long repl_backlog_size; /* Backlog circular buffer size */
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
replDataBuf pending_repl_data; /* Replication data buffer for dual-channel-replication */
|
2024-06-07 14:21:33 -07:00
|
|
|
time_t repl_backlog_time_limit; /* Time without replicas after the backlog
|
2024-05-28 09:27:51 -07:00
|
|
|
gets released. */
|
2024-06-07 14:21:33 -07:00
|
|
|
time_t repl_no_replicas_since; /* We have no replicas since that time.
|
|
|
|
Only valid if server.replicas len is 0. */
|
|
|
|
int repl_min_replicas_to_write; /* Min number of replicas to write. */
|
|
|
|
int repl_min_replicas_max_lag; /* Max lag of <count> replicas to write. */
|
|
|
|
int repl_good_replicas_count; /* Number of replicas with lag <= max_lag. */
|
|
|
|
int repl_diskless_sync; /* Primary send RDB to replicas sockets directly. */
|
|
|
|
int repl_diskless_load; /* Replica parse RDB directly from the socket.
|
2024-05-28 09:27:51 -07:00
|
|
|
* see REPL_DISKLESS_LOAD_* enum */
|
|
|
|
int repl_diskless_sync_delay; /* Delay to start a diskless repl BGSAVE. */
|
|
|
|
int repl_diskless_sync_max_replicas; /* Max replicas for diskless repl BGSAVE
|
|
|
|
* delay (start sooner if they all connect). */
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
int dual_channel_replication; /* Config used to determine if the replica should
|
|
|
|
* use dual channel replication for full syncs. */
|
|
|
|
int wait_before_rdb_client_free; /* Grace period in seconds for replica main channel
|
|
|
|
* to establish psync. */
|
2024-07-25 19:34:39 +03:00
|
|
|
int debug_pause_after_fork; /* Debug param that pauses the main process
|
|
|
|
* after a replication fork() (for bgsave). */
|
2024-05-28 09:27:51 -07:00
|
|
|
size_t repl_buffer_mem; /* The memory of replication buffer. */
|
|
|
|
list *repl_buffer_blocks; /* Replication buffers blocks list
|
|
|
|
* (serving replica clients and repl backlog) */
|
2024-06-07 14:21:33 -07:00
|
|
|
/* Replication (replica) */
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
char *primary_user; /* AUTH with this user and primary_auth with primary */
|
|
|
|
sds primary_auth; /* AUTH with this password with primary */
|
|
|
|
char *primary_host; /* Hostname of primary */
|
|
|
|
int primary_port; /* Port of primary */
|
|
|
|
int repl_timeout; /* Timeout after N seconds of primary idle */
|
|
|
|
client *primary; /* Client that is primary for this replica */
|
|
|
|
uint64_t rdb_client_id; /* Rdb client id as it defined at primary side */
|
|
|
|
struct {
|
|
|
|
connection *conn;
|
|
|
|
char replid[CONFIG_RUN_ID_SIZE + 1];
|
|
|
|
long long reploff;
|
|
|
|
long long read_reploff;
|
|
|
|
int dbid;
|
|
|
|
} repl_provisional_primary;
|
2024-06-07 14:21:33 -07:00
|
|
|
client *cached_primary; /* Cached primary to be reused for PSYNC. */
|
2024-05-22 23:24:12 -07:00
|
|
|
int repl_syncio_timeout; /* Timeout for synchronous I/O calls */
|
2024-06-07 14:21:33 -07:00
|
|
|
int repl_state; /* Replication status if the instance is a replica */
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
int repl_rdb_channel_state; /* State of the replica's rdb channel during dual-channel-replication */
|
2024-06-07 14:21:33 -07:00
|
|
|
off_t repl_transfer_size; /* Size of RDB to read from primary during sync. */
|
|
|
|
off_t repl_transfer_read; /* Amount of RDB read from primary during sync. */
|
2012-08-24 19:28:44 +02:00
|
|
|
off_t repl_transfer_last_fsync_off; /* Offset when we fsync-ed last time. */
|
2024-06-07 14:21:33 -07:00
|
|
|
connection *repl_transfer_s; /* Replica -> Primary SYNC connection */
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
connection *repl_rdb_transfer_s; /* Primary FULL SYNC connection (RDB download) */
|
2024-06-07 14:21:33 -07:00
|
|
|
int repl_transfer_fd; /* Replica -> Primary SYNC temp file descriptor */
|
|
|
|
char *repl_transfer_tmpfile; /* Replica-> Primary SYNC temp file name */
|
2024-05-22 23:24:12 -07:00
|
|
|
time_t repl_transfer_lastio; /* Unix time of the latest read, for timeout */
|
|
|
|
int repl_serve_stale_data; /* Serve stale data when link is down? */
|
2024-06-07 14:21:33 -07:00
|
|
|
int repl_replica_ro; /* Replica is read only? */
|
|
|
|
int repl_replica_ignore_maxmemory; /* If true replicas do not evict. */
|
|
|
|
time_t repl_down_since; /* Unix time at which link with primary went down */
|
2024-05-22 23:24:12 -07:00
|
|
|
int repl_disable_tcp_nodelay; /* Disable TCP_NODELAY after SYNC? */
|
2024-06-07 14:21:33 -07:00
|
|
|
int replica_priority; /* Reported in INFO and used by Sentinel. */
|
2024-05-22 23:24:12 -07:00
|
|
|
int replica_announced; /* If true, replica is announced by Sentinel */
|
2024-06-07 14:21:33 -07:00
|
|
|
int replica_announce_port; /* Give the primary this listening port. */
|
|
|
|
char *replica_announce_ip; /* Give the primary this ip address. */
|
2024-05-22 23:24:12 -07:00
|
|
|
int propagation_error_behavior; /* Configures the behavior of the replica
|
|
|
|
* when it receives an error on the replication stream */
|
Set replicas to panic on disk errors, and optionally panic on replication errors (#10504)
* Till now, replicas that were unable to persist, would still execute the commands
they got from the master, now they'll panic by default, and we add a new
`replica-ignore-disk-errors` config to change that.
* Till now, when a command failed on a replica or AOF-loading, it only logged a
warning and a stat, we add a new `propagation-error-behavior` config to allow
panicking in that state (may become the default one day)
Note that commands that fail on the replica can either indicate a bug that could
cause data inconsistency between the replica and the master, or they could be
in some cases (specifically in previous versions), a result of a command (e.g. EVAL)
that failed on the master, but still had to be propagated to fail on the replica as well.
2022-04-26 03:25:33 -07:00
|
|
|
int repl_ignore_disk_write_error; /* Configures whether replicas panic when unable to
|
|
|
|
* persist writes to AOF. */
|
2024-06-07 14:21:33 -07:00
|
|
|
/* The following two fields is where we store primary PSYNC replid/offset
|
PSYNC2: different improvements to Redis replication.
The gist of the changes is that now, partial resynchronizations between
slaves and masters (without the need of a full resync with RDB transfer
and so forth), work in a number of cases when it was impossible
in the past. For instance:
1. When a slave is promoted to mastrer, the slaves of the old master can
partially resynchronize with the new master.
2. Chained slalves (slaves of slaves) can be moved to replicate to other
slaves or the master itsef, without requiring a full resync.
3. The master itself, after being turned into a slave, is able to
partially resynchronize with the new master, when it joins replication
again.
In order to obtain this, the following main changes were operated:
* Slaves also take a replication backlog, not just masters.
* Same stream replication for all the slaves and sub slaves. The
replication stream is identical from the top level master to its slaves
and is also the same from the slaves to their sub-slaves and so forth.
This means that if a slave is later promoted to master, it has the
same replication backlong, and can partially resynchronize with its
slaves (that were previously slaves of the old master).
* A given replication history is no longer identified by the `runid` of
a Redis node. There is instead a `replication ID` which changes every
time the instance has a new history no longer coherent with the past
one. So, for example, slaves publish the same replication history of
their master, however when they are turned into masters, they publish
a new replication ID, but still remember the old ID, so that they are
able to partially resynchronize with slaves of the old master (up to a
given offset).
* The replication protocol was slightly modified so that a new extended
+CONTINUE reply from the master is able to inform the slave of a
replication ID change.
* REPLCONF CAPA is used in order to notify masters that a slave is able
to understand the new +CONTINUE reply.
* The RDB file was extended with an auxiliary field that is able to
select a given DB after loading in the slave, so that the slave can
continue receiving the replication stream from the point it was
disconnected without requiring the master to insert "SELECT" statements.
This is useful in order to guarantee the "same stream" property, because
the slave must be able to accumulate an identical backlog.
* Slave pings to sub-slaves are now sent in a special form, when the
top-level master is disconnected, in order to don't interfer with the
replication stream. We just use out of band "\n" bytes as in other parts
of the Redis protocol.
An old design document is available here:
https://gist.github.com/antirez/ae068f95c0d084891305
However the implementation is not identical to the description because
during the work to implement it, different changes were needed in order
to make things working well.
2016-11-09 11:31:06 +01:00
|
|
|
* while the PSYNC is in progress. At the end we'll copy the fields into
|
2024-06-07 14:21:33 -07:00
|
|
|
* the server->primary client structure. */
|
|
|
|
char primary_replid[CONFIG_RUN_ID_SIZE + 1]; /* Primary PSYNC runid. */
|
|
|
|
long long primary_initial_offset; /* Primary PSYNC offset. */
|
|
|
|
int repl_replica_lazy_flush; /* Lazy FLUSHALL before loading DB? */
|
2013-12-04 15:52:20 +01:00
|
|
|
/* Synchronous replication. */
|
2024-05-22 23:24:12 -07:00
|
|
|
list *clients_waiting_acks; /* Clients waiting in WAIT or WAITAOF. */
|
2024-06-07 14:21:33 -07:00
|
|
|
int get_ack_from_replicas; /* If true we send REPLCONF GETACK. */
|
2010-11-04 17:29:53 +01:00
|
|
|
/* Limits */
|
2024-05-22 23:24:12 -07:00
|
|
|
unsigned int maxclients; /* Max number of simultaneous clients */
|
|
|
|
unsigned long long maxmemory; /* Max number of memory bytes to use */
|
|
|
|
ssize_t maxmemory_clients; /* Memory limit for total client buffers */
|
|
|
|
int maxmemory_policy; /* Policy for key eviction */
|
|
|
|
int maxmemory_samples; /* Precision of random sampling */
|
|
|
|
int maxmemory_eviction_tenacity; /* Aggressiveness of eviction processing */
|
|
|
|
int lfu_log_factor; /* LFU logarithmic counter factor. */
|
|
|
|
int lfu_decay_time; /* LFU counter decay factor. */
|
|
|
|
long long proto_max_bulk_len; /* Protocol bulk length maximum size. */
|
|
|
|
int oom_score_adj_values[CONFIG_OOM_COUNT]; /* Linux oom_score_adj configuration */
|
|
|
|
int oom_score_adj; /* If true, oom_score_adj is managed */
|
|
|
|
int disable_thp; /* If true, disable THP by syscall */
|
2010-11-04 17:29:53 +01:00
|
|
|
/* Blocked clients */
|
2024-05-22 23:24:12 -07:00
|
|
|
unsigned int blocked_clients; /* # of clients executing a blocking cmd.*/
|
2017-09-09 11:10:59 +02:00
|
|
|
unsigned int blocked_clients_by_type[BLOCKED_NUM];
|
2010-12-29 19:39:42 +01:00
|
|
|
list *unblocked_clients; /* list of clients to unblock before next loop */
|
A reimplementation of blocking operation internals.
Redis provides support for blocking operations such as BLPOP or BRPOP.
This operations are identical to normal LPOP and RPOP operations as long
as there are elements in the target list, but if the list is empty they
block waiting for new data to arrive to the list.
All the clients blocked waiting for th same list are served in a FIFO
way, so the first that blocked is the first to be served when there is
more data pushed by another client into the list.
The previous implementation of blocking operations was conceived to
serve clients in the context of push operations. For for instance:
1) There is a client "A" blocked on list "foo".
2) The client "B" performs `LPUSH foo somevalue`.
3) The client "A" is served in the context of the "B" LPUSH,
synchronously.
Processing things in a synchronous way was useful as if "A" pushes a
value that is served by "B", from the point of view of the database is a
NOP (no operation) thing, that is, nothing is replicated, nothing is
written in the AOF file, and so forth.
However later we implemented two things:
1) Variadic LPUSH that could add multiple values to a list in the
context of a single call.
2) BRPOPLPUSH that was a version of BRPOP that also provided a "PUSH"
side effect when receiving data.
This forced us to make the synchronous implementation more complex. If
client "B" is waiting for data, and "A" pushes three elemnents in a
single call, we needed to propagate an LPUSH with a missing argument
in the AOF and replication link. We also needed to make sure to
replicate the LPUSH side of BRPOPLPUSH, but only if in turn did not
happened to serve another blocking client into another list ;)
This were complex but with a few of mutually recursive functions
everything worked as expected... until one day we introduced scripting
in Redis.
Scripting + synchronous blocking operations = Issue #614.
Basically you can't "rewrite" a script to have just a partial effect on
the replicas and AOF file if the script happened to serve a few blocked
clients.
The solution to all this problems, implemented by this commit, is to
change the way we serve blocked clients. Instead of serving the blocked
clients synchronously, in the context of the command performing the PUSH
operation, it is now an asynchronous and iterative process:
1) If a key that has clients blocked waiting for data is the subject of
a list push operation, We simply mark keys as "ready" and put it into a
queue.
2) Every command pushing stuff on lists, as a variadic LPUSH, a script,
or whatever it is, is replicated verbatim without any rewriting.
3) Every time a Redis command, a MULTI/EXEC block, or a script,
completed its execution, we run the list of keys ready to serve blocked
clients (as more data arrived), and process this list serving the
blocked clients.
4) As a result of "3" maybe more keys are ready again for other clients
(as a result of BRPOPLPUSH we may have push operations), so we iterate
back to step "3" if it's needed.
The new code has a much simpler semantics, and a simpler to understand
implementation, with the disadvantage of not being able to "optmize out"
a PUSH+BPOP as a No OP.
This commit will be tested with care before the final merge, more tests
will be added likely.
2012-09-04 10:37:49 +02:00
|
|
|
list *ready_keys; /* List of readyList structures for BLPOP & co */
|
2019-06-29 20:08:41 -04:00
|
|
|
/* Client side caching. */
|
|
|
|
unsigned int tracking_clients; /* # of clients with tracking enabled.*/
|
2020-02-19 19:00:29 +01:00
|
|
|
size_t tracking_table_max_keys; /* Max number of keys in tracking table. */
|
2024-05-22 23:24:12 -07:00
|
|
|
list *tracking_pending_keys; /* tracking invalidation keys pending to flush */
|
|
|
|
list *pending_push_messages; /* pending publish or other push messages to flush */
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Sort parameters - qsort_r() is only available under BSD so we
|
|
|
|
* have to take this state global, in order to pass it to sortCompare() */
|
|
|
|
int sort_desc;
|
|
|
|
int sort_alpha;
|
|
|
|
int sort_bypattern;
|
2013-07-12 12:02:36 +02:00
|
|
|
int sort_store;
|
2011-12-21 10:05:32 +01:00
|
|
|
/* Zip structure config, see redis.conf for more information */
|
2021-08-10 14:18:49 +08:00
|
|
|
size_t hash_max_listpack_entries;
|
|
|
|
size_t hash_max_listpack_value;
|
2010-07-02 19:57:12 +02:00
|
|
|
size_t set_max_intset_entries;
|
2022-11-09 18:50:07 +01:00
|
|
|
size_t set_max_listpack_entries;
|
|
|
|
size_t set_max_listpack_value;
|
2021-09-09 23:18:53 +08:00
|
|
|
size_t zset_max_listpack_entries;
|
|
|
|
size_t zset_max_listpack_value;
|
2014-04-15 17:46:51 +02:00
|
|
|
size_t hll_sparse_max_bytes;
|
2018-06-07 14:24:45 +02:00
|
|
|
size_t stream_node_max_bytes;
|
2019-11-28 11:24:57 +02:00
|
|
|
long long stream_node_max_entries;
|
2014-12-16 00:49:14 -05:00
|
|
|
/* List parameters */
|
2021-11-24 19:34:13 +08:00
|
|
|
int list_max_listpack_size;
|
2014-12-16 00:49:14 -05:00
|
|
|
int list_compress_depth;
|
|
|
|
/* time cache */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
time_t unixtime; /* Unix time sampled every cron cycle. */
|
2024-05-28 09:27:51 -07:00
|
|
|
time_t timezone; /* Cached timezone. As set by tzset(). */
|
2024-08-14 13:08:20 +08:00
|
|
|
_Atomic int daylight_active; /* Currently in daylight saving time. */
|
2024-05-28 09:27:51 -07:00
|
|
|
mstime_t mstime; /* 'unixtime' in milliseconds. */
|
|
|
|
ustime_t ustime; /* 'unixtime' in microseconds. */
|
|
|
|
mstime_t cmd_time_snapshot; /* Time snapshot of the root execution nesting. */
|
|
|
|
size_t blocking_op_nesting; /* Nesting level of blocking operation, used to reset blocked_last_cron. */
|
2024-05-26 17:41:11 +01:00
|
|
|
long long blocked_last_cron; /* Indicate the mstime of the last time we did cron jobs from a blocking operation */
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Pubsub */
|
2024-05-22 23:24:12 -07:00
|
|
|
kvstore *pubsub_channels; /* Map channels to list of subscribed clients */
|
|
|
|
dict *pubsub_patterns; /* A dict of pubsub_patterns */
|
|
|
|
int notify_keyspace_events; /* Events to propagate via Pub/Sub. This is an
|
|
|
|
xor of NOTIFY_... flags. */
|
|
|
|
kvstore *pubsubshard_channels; /* Map shard channels in every slot to list of subscribed clients */
|
|
|
|
unsigned int pubsub_clients; /* # of clients in Pub/Sub mode */
|
2024-04-01 03:38:22 +08:00
|
|
|
unsigned int watching_clients; /* # of clients are watching keys */
|
2011-04-28 19:00:33 +02:00
|
|
|
/* Cluster */
|
2024-05-22 23:24:12 -07:00
|
|
|
int cluster_enabled; /* Is cluster enabled? */
|
|
|
|
int cluster_port; /* Set the cluster port for a node. */
|
|
|
|
mstime_t cluster_node_timeout; /* Cluster node timeout. */
|
|
|
|
mstime_t cluster_ping_interval; /* A debug configuration for setting how often cluster nodes send ping messages. */
|
|
|
|
char *cluster_configfile; /* Cluster auto-generated config file name. */
|
|
|
|
struct clusterState *cluster; /* State of the cluster */
|
|
|
|
int cluster_migration_barrier; /* Cluster replicas migration barrier. */
|
2024-06-07 14:21:33 -07:00
|
|
|
int cluster_allow_replica_migration; /* Automatic replica migrations to orphaned primaries and from empty primaries */
|
|
|
|
int cluster_replica_validity_factor; /* Replica max data age for failover. */
|
|
|
|
int cluster_require_full_coverage; /* If true, put the cluster down if
|
|
|
|
there is at least an uncovered slot.*/
|
|
|
|
int cluster_replica_no_failover; /* Prevent replica from starting a failover
|
|
|
|
if the primary is in failure state. */
|
|
|
|
char *cluster_announce_ip; /* IP address to announce on cluster bus. */
|
2024-07-10 13:53:52 +02:00
|
|
|
char *cluster_announce_client_ipv4; /* IPv4 for clients, to announce on cluster bus. */
|
|
|
|
char *cluster_announce_client_ipv6; /* IPv6 for clients, to announce on cluster bus. */
|
2024-06-07 14:21:33 -07:00
|
|
|
char *cluster_announce_hostname; /* hostname to announce on cluster bus. */
|
|
|
|
char *cluster_announce_human_nodename; /* Human readable node name assigned to a node. */
|
|
|
|
int cluster_preferred_endpoint_type; /* Use the announced hostname when available. */
|
|
|
|
int cluster_announce_port; /* base port to announce on cluster bus. */
|
|
|
|
int cluster_announce_tls_port; /* TLS port to announce on cluster bus. */
|
|
|
|
int cluster_announce_bus_port; /* bus port to announce on cluster bus. */
|
|
|
|
int cluster_module_flags; /* Set of flags that modules are able
|
|
|
|
to set in order to suppress certain
|
|
|
|
native Redis Cluster features. Check the
|
|
|
|
VALKEYMODULE_CLUSTER_FLAG_*. */
|
|
|
|
int cluster_allow_reads_when_down; /* Are reads allowed when the cluster
|
|
|
|
is down? */
|
|
|
|
int cluster_config_file_lock_fd; /* cluster config fd, will be flocked. */
|
2024-05-22 23:24:12 -07:00
|
|
|
unsigned long long cluster_link_msg_queue_limit_bytes; /* Memory usage limit on individual link msg queue */
|
|
|
|
int cluster_drop_packet_filter; /* Debug config that allows tactically
|
|
|
|
* dropping packets of a specific type */
|
2024-07-13 11:38:25 -07:00
|
|
|
unsigned long cluster_blacklist_ttl; /* Duration in seconds that a node is denied re-entry into
|
|
|
|
* the cluster after it is forgotten with CLUSTER FORGET. */
|
2024-07-22 21:03:28 -04:00
|
|
|
int cluster_slot_stats_enabled; /* Cluster slot usage statistics tracking enabled. */
|
2024-06-16 20:37:09 -07:00
|
|
|
/* Debug config that goes along with cluster_drop_packet_filter. When set, the link is closed on packet drop. */
|
|
|
|
uint32_t debug_cluster_close_link_on_packet_drop : 1;
|
2024-07-10 13:53:52 +02:00
|
|
|
sds cached_cluster_slot_info[CACHE_CONN_TYPE_MAX]; /* Index in array is a bitwise or of CACHE_CONN_TYPE_* */
|
2011-04-30 17:46:52 +02:00
|
|
|
/* Scripting */
|
2022-01-20 09:05:53 +02:00
|
|
|
mstime_t busy_reply_threshold; /* Script / module timeout in milliseconds */
|
2024-05-22 23:24:12 -07:00
|
|
|
int pre_command_oom_state; /* OOM before command (script?) was started */
|
|
|
|
int script_disable_deny_script; /* Allow running commands marked "noscript" inside a script. */
|
2015-10-02 15:27:57 +02:00
|
|
|
/* Lazy free */
|
|
|
|
int lazyfree_lazy_eviction;
|
|
|
|
int lazyfree_lazy_expire;
|
|
|
|
int lazyfree_lazy_server_del;
|
2019-07-17 11:00:51 +08:00
|
|
|
int lazyfree_lazy_user_del;
|
2021-01-15 21:32:58 +08:00
|
|
|
int lazyfree_lazy_user_flush;
|
2014-07-01 11:30:15 +02:00
|
|
|
/* Latency monitor */
|
|
|
|
long long latency_monitor_threshold;
|
|
|
|
dict *latency_events;
|
2019-02-05 10:48:17 +01:00
|
|
|
/* ACLs */
|
Adds pub/sub channel patterns to ACL (#7993)
Fixes #7923.
This PR appropriates the special `&` symbol (because `@` and `*` are taken),
followed by a literal value or pattern for describing the Pub/Sub patterns that
an ACL user can interact with. It is similar to the existing key patterns
mechanism in function (additive) and implementation (copy-pasta). It also adds
the allchannels and resetchannels ACL keywords, naturally.
The default user is given allchannels permissions, whereas new users get
whatever is defined by the acl-pubsub-default configuration directive. For
backward compatibility in 6.2, the default of this directive is allchannels but
this is likely to be changed to resetchannels in the next major version for
stronger default security settings.
Unless allchannels is set for the user, channel access permissions are checked
as follows :
* Calls to both PUBLISH and SUBSCRIBE will fail unless a pattern matching the
argumentative channel name(s) exists for the user.
* Calls to PSUBSCRIBE will fail unless the pattern(s) provided as an argument
literally exist(s) in the user's list.
Such failures are logged to the ACL log.
Runtime changes to channel permissions for a user with existing subscribing
clients cause said clients to disconnect unless the new permissions permit the
connections to continue. Note, however, that PSUBSCRIBErs' patterns are matched
literally, so given the change bar:* -> b*, pattern subscribers to bar:* will be
disconnected.
Notes/questions:
* UNSUBSCRIBE, PUNSUBSCRIBE and PUBSUB remain unprotected due to lack of reasons
for touching them.
2020-12-01 14:21:39 +02:00
|
|
|
char *acl_filename; /* ACL Users file. NULL if not configured. */
|
2020-02-04 13:19:40 +01:00
|
|
|
unsigned long acllog_max_len; /* Maximum length of the ACL LOG list. */
|
Adds pub/sub channel patterns to ACL (#7993)
Fixes #7923.
This PR appropriates the special `&` symbol (because `@` and `*` are taken),
followed by a literal value or pattern for describing the Pub/Sub patterns that
an ACL user can interact with. It is similar to the existing key patterns
mechanism in function (additive) and implementation (copy-pasta). It also adds
the allchannels and resetchannels ACL keywords, naturally.
The default user is given allchannels permissions, whereas new users get
whatever is defined by the acl-pubsub-default configuration directive. For
backward compatibility in 6.2, the default of this directive is allchannels but
this is likely to be changed to resetchannels in the next major version for
stronger default security settings.
Unless allchannels is set for the user, channel access permissions are checked
as follows :
* Calls to both PUBLISH and SUBSCRIBE will fail unless a pattern matching the
argumentative channel name(s) exists for the user.
* Calls to PSUBSCRIBE will fail unless the pattern(s) provided as an argument
literally exist(s) in the user's list.
Such failures are logged to the ACL log.
Runtime changes to channel permissions for a user with existing subscribing
clients cause said clients to disconnect unless the new permissions permit the
connections to continue. Note, however, that PSUBSCRIBErs' patterns are matched
literally, so given the change bar:* -> b*, pattern subscribers to bar:* will be
disconnected.
Notes/questions:
* UNSUBSCRIBE, PUNSUBSCRIBE and PUBSUB remain unprotected due to lack of reasons
for touching them.
2020-12-01 14:21:39 +02:00
|
|
|
sds requirepass; /* Remember the cleartext password set with
|
|
|
|
the old "requirepass" directive for
|
|
|
|
backward compatibility with Redis <= 5. */
|
2024-05-22 23:24:12 -07:00
|
|
|
int acl_pubsub_default; /* Default ACL pub/sub channels flag */
|
|
|
|
aclInfo acl_info; /* ACL info */
|
2013-01-17 01:00:20 +08:00
|
|
|
/* Assert & bug reporting */
|
2024-05-22 23:24:12 -07:00
|
|
|
int watchdog_period; /* Software watchdog period in ms. 0 = off */
|
2014-06-17 23:05:30 -04:00
|
|
|
/* System hardware info */
|
2024-05-22 23:24:12 -07:00
|
|
|
size_t system_memory_size; /* Total memory in system as reported by OS */
|
2019-09-12 10:56:54 +03:00
|
|
|
/* TLS Configuration */
|
|
|
|
int tls_cluster;
|
|
|
|
int tls_replication;
|
|
|
|
int tls_auth_clients;
|
2024-04-04 00:06:01 +07:00
|
|
|
serverTLSContextConfig tls_ctx_config;
|
Support setcpuaffinity on linux/bsd
Currently, there are several types of threads/child processes of a
redis server. Sometimes we need deeply optimise the performance of
redis, so we would like to isolate threads/processes.
There were some discussion about cpu affinity cases in the issue:
https://github.com/antirez/redis/issues/2863
So implement cpu affinity setting by redis.conf in this patch, then
we can config server_cpulist/bio_cpulist/aof_rewrite_cpulist/
bgsave_cpulist by cpu list.
Examples of cpulist in redis.conf:
server_cpulist 0-7:2 means cpu affinity 0,2,4,6
bio_cpulist 1,3 means cpu affinity 1,3
aof_rewrite_cpulist 8-11 means cpu affinity 8,9,10,11
bgsave_cpulist 1,10-11 means cpu affinity 1,10,11
Test on linux/freebsd, both work fine.
Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
2020-05-02 20:05:39 +08:00
|
|
|
/* cpu affinity */
|
2024-05-22 23:24:12 -07:00
|
|
|
char *server_cpulist; /* cpu affinity list of server main/io thread. */
|
|
|
|
char *bio_cpulist; /* cpu affinity list of bio thread. */
|
Support setcpuaffinity on linux/bsd
Currently, there are several types of threads/child processes of a
redis server. Sometimes we need deeply optimise the performance of
redis, so we would like to isolate threads/processes.
There were some discussion about cpu affinity cases in the issue:
https://github.com/antirez/redis/issues/2863
So implement cpu affinity setting by redis.conf in this patch, then
we can config server_cpulist/bio_cpulist/aof_rewrite_cpulist/
bgsave_cpulist by cpu list.
Examples of cpulist in redis.conf:
server_cpulist 0-7:2 means cpu affinity 0,2,4,6
bio_cpulist 1,3 means cpu affinity 1,3
aof_rewrite_cpulist 8-11 means cpu affinity 8,9,10,11
bgsave_cpulist 1,10-11 means cpu affinity 1,10,11
Test on linux/freebsd, both work fine.
Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
2020-05-02 20:05:39 +08:00
|
|
|
char *aof_rewrite_cpulist; /* cpu affinity list of aof rewrite process. */
|
2024-05-22 23:24:12 -07:00
|
|
|
char *bgsave_cpulist; /* cpu affinity list of bgsave process. */
|
Sentinel: Fix Config Dependency and Rewrite Sequence (#8271)
This commit fixes a well known and an annoying issue in Sentinel mode.
Cause of this issue:
Currently, Redis rewrite process works well in server mode, however in sentinel mode,
the sentinel config has variant semantics for different configurations, in example configuration
https://github.com/redis/redis/blob/unstable/sentinel.conf, we put comments on these.
However the rewrite process only treat the sentinel config as a single option. During rewrite
process, it will mess up with the lines and comments.
Approaches:
In order to solve this issue, we need to differentiate different subconfig options in sentinel separately,
for example, sentinel monitor <master-name> <ip> <redis-port> <quorum>
we can treat it as sentinel monitor option, instead of the sentinel option.
This commit also fixes the dependency issue when putting configurations in sentinel.conf.
For example before this commit,we must put
`sentinel monitor <master-name> <ip> <redis-port> <quorum>` before
`sentinel auth-pass <master-name> <password>` for a single master,
otherwise the server cannot start and will return error. This commit fixes this issue, as long as
the monitoring master was configured, no matter the sequence is, the sentinel can start and run properly.
2021-01-26 02:31:54 -05:00
|
|
|
/* Sentinel config */
|
|
|
|
struct sentinelConfig *sentinel_config; /* sentinel config to load at startup time. */
|
2021-01-28 13:18:05 -08:00
|
|
|
/* Coordinate failover info */
|
2024-05-22 23:24:12 -07:00
|
|
|
mstime_t failover_end_time; /* Deadline for failover command. */
|
|
|
|
int force_failover; /* If true then failover will be forced at the
|
|
|
|
* deadline, otherwise failover is aborted. */
|
|
|
|
char *target_replica_host; /* Failover target host. If null during a
|
|
|
|
* failover then any replica can be used. */
|
|
|
|
int target_replica_port; /* Failover target port */
|
|
|
|
int failover_state; /* Failover state */
|
2022-01-03 01:54:47 +01:00
|
|
|
int cluster_allow_pubsubshard_when_down; /* Is pubsubshard allowed when the cluster
|
|
|
|
is down, doesn't affect pubsub global. */
|
2022-02-22 11:19:38 +02:00
|
|
|
long reply_buffer_peak_reset_time; /* The amount of time (in milliseconds) to wait between reply buffer peak resets */
|
2022-03-01 14:40:29 +02:00
|
|
|
int reply_buffer_resizing_enabled; /* Is reply buffer resizing enabled (1 by default) */
|
2024-06-27 15:30:26 -04:00
|
|
|
sds availability_zone; /* When run in a cloud environment we can configure the availability zone it is running in */
|
Support setlocale via CONFIG operation. (#11059)
Till now Redis officially supported tuning it via environment variable see #1074.
But we had other requests to allow changing it at runtime, see #799, and #11041.
Note that `strcoll()` is used as Lua comparison function and also for comparison of
certain string objects in Redis, which leads to a problem that, in different regions,
for some characters, the result may be different. Below is an example.
```
127.0.0.1:6333> SORT test alpha
1) "<"
2) ">"
3) ","
4) "*"
127.0.0.1:6333> CONFIG GET locale-collate
1) "locale-collate"
2) ""
127.0.0.1:6333> CONFIG SET locale-collate 1
(error) ERR CONFIG SET failed (possibly related to argument 'locale')
127.0.0.1:6333> CONFIG SET locale-collate C
OK
127.0.0.1:6333> SORT test alpha
1) "*"
2) ","
3) "<"
4) ">"
```
That will cause accidental code compatibility issues for Lua scripts and some
Redis commands. This commit creates a new config parameter to control the
local environment which only affects `Collate` category. Above shows how it
affects `SORT` command, and below shows the influence on Lua scripts.
```
127.0.0.1:6333> CONFIG GET locale-collate
1) " locale-collate"
2) "C"
127.0.0.1:6333> EVAL "return ',' < '*'" 0
(nil)
127.0.0.1:6333> CONFIG SET locale-collate ""
OK
127.0.0.1:6333> EVAL "return ',' < '*'" 0
(integer) 1
```
Co-authored-by: calvincjli <calvincjli@tencent.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-08-21 22:55:45 +08:00
|
|
|
/* Local environment */
|
|
|
|
char *locale_collate;
|
2024-08-13 02:33:23 +03:00
|
|
|
char *debug_context; /* A free-form string that has no impact on server except being included in a crash report. */
|
2010-06-22 00:07:48 +02:00
|
|
|
};
|
|
|
|
|
2020-10-05 17:03:17 +03:00
|
|
|
#define MAX_KEYS_BUFFER 256
|
|
|
|
|
2022-01-20 13:05:27 -08:00
|
|
|
typedef struct {
|
2024-05-22 23:24:12 -07:00
|
|
|
int pos; /* The position of the key within the client array */
|
2022-01-27 03:03:21 +08:00
|
|
|
int flags; /* The flags associated with the key access, see
|
2022-01-20 13:05:27 -08:00
|
|
|
CMD_KEY_* for more information */
|
|
|
|
} keyReference;
|
|
|
|
|
2020-10-05 17:03:17 +03:00
|
|
|
/* A result structure for the various getkeys function calls. It lists the
|
2022-02-22 01:00:03 -08:00
|
|
|
* keys as indices to the provided argv. This functionality is also re-used
|
|
|
|
* for returning channel information.
|
2020-10-05 17:03:17 +03:00
|
|
|
*/
|
|
|
|
typedef struct {
|
2024-05-22 23:24:12 -07:00
|
|
|
int numkeys; /* Number of key indices return */
|
|
|
|
int size; /* Available array size */
|
2024-06-14 08:42:00 -07:00
|
|
|
keyReference *keys; /* Key indices array, points to keysbuf or heap */
|
|
|
|
keyReference keysbuf[MAX_KEYS_BUFFER]; /* Pre-allocated buffer, to save heap allocations */
|
2020-10-05 17:03:17 +03:00
|
|
|
} getKeysResult;
|
2024-06-14 08:42:00 -07:00
|
|
|
|
|
|
|
static inline void initGetKeysResult(getKeysResult *result) {
|
|
|
|
result->numkeys = 0;
|
|
|
|
result->size = MAX_KEYS_BUFFER;
|
|
|
|
result->keys = NULL;
|
|
|
|
}
|
2020-10-05 17:03:17 +03:00
|
|
|
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
/* Key specs definitions.
|
|
|
|
*
|
|
|
|
* Brief: This is a scheme that tries to describe the location
|
|
|
|
* of key arguments better than the old [first,last,step] scheme
|
|
|
|
* which is limited and doesn't fit many commands.
|
|
|
|
*
|
|
|
|
* There are two steps:
|
2022-01-27 03:03:21 +08:00
|
|
|
* 1. begin_search (BS): in which index should we start searching for keys?
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
* 2. find_keys (FK): relative to the output of BS, how can we will which args are keys?
|
|
|
|
*
|
|
|
|
* There are two types of BS:
|
|
|
|
* 1. index: key args start at a constant index
|
|
|
|
* 2. keyword: key args start just after a specific keyword
|
|
|
|
*
|
|
|
|
* There are two kinds of FK:
|
|
|
|
* 1. range: keys end at a specific index (or relative to the last argument)
|
|
|
|
* 2. keynum: there's an arg that contains the number of key args somewhere before the keys themselves
|
|
|
|
*/
|
|
|
|
|
2024-04-05 16:59:55 -07:00
|
|
|
/* WARNING! Must be synced with generate-command-code.py and ValkeyModuleKeySpecBeginSearchType */
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
typedef enum {
|
|
|
|
KSPEC_BS_INVALID = 0, /* Must be 0 */
|
|
|
|
KSPEC_BS_UNKNOWN,
|
|
|
|
KSPEC_BS_INDEX,
|
|
|
|
KSPEC_BS_KEYWORD
|
|
|
|
} kspec_bs_type;
|
|
|
|
|
2024-04-05 16:59:55 -07:00
|
|
|
/* WARNING! Must be synced with generate-command-code.py and ValkeyModuleKeySpecFindKeysType */
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
typedef enum {
|
|
|
|
KSPEC_FK_INVALID = 0, /* Must be 0 */
|
|
|
|
KSPEC_FK_UNKNOWN,
|
|
|
|
KSPEC_FK_RANGE,
|
|
|
|
KSPEC_FK_KEYNUM
|
|
|
|
} kspec_fk_type;
|
|
|
|
|
2024-04-05 16:59:55 -07:00
|
|
|
/* WARNING! This struct must match ValkeyModuleCommandKeySpec */
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
typedef struct {
|
|
|
|
/* Declarative data */
|
2022-01-30 11:00:03 +01:00
|
|
|
const char *notes;
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
uint64_t flags;
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
kspec_bs_type begin_search_type;
|
|
|
|
union {
|
|
|
|
struct {
|
|
|
|
/* The index from which we start the search for keys */
|
|
|
|
int pos;
|
|
|
|
} index;
|
|
|
|
struct {
|
|
|
|
/* The keyword that indicates the beginning of key args */
|
|
|
|
const char *keyword;
|
|
|
|
/* An index in argv from which to start searching.
|
|
|
|
* Can be negative, which means start search from the end, in reverse
|
2022-01-27 03:03:21 +08:00
|
|
|
* (Example: -2 means to start in reverse from the penultimate arg) */
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
int startfrom;
|
|
|
|
} keyword;
|
|
|
|
} bs;
|
|
|
|
kspec_fk_type find_keys_type;
|
|
|
|
union {
|
|
|
|
/* NOTE: Indices in this struct are relative to the result of the begin_search step!
|
|
|
|
* These are: range.lastkey, keynum.keynumidx, keynum.firstkey */
|
|
|
|
struct {
|
|
|
|
/* Index of the last key.
|
|
|
|
* Can be negative, in which case it's not relative. -1 indicating till the last argument,
|
|
|
|
* -2 one before the last and so on. */
|
|
|
|
int lastkey;
|
|
|
|
/* How many args should we skip after finding a key, in order to find the next one. */
|
|
|
|
int keystep;
|
|
|
|
/* If lastkey is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit.
|
|
|
|
* 2 means 1/2 of the remaining args, 3 means 1/3, and so on. */
|
|
|
|
int limit;
|
|
|
|
} range;
|
|
|
|
struct {
|
|
|
|
/* Index of the argument containing the number of keys to come */
|
|
|
|
int keynumidx;
|
|
|
|
/* Index of the fist key (Usually it's just after keynumidx, in
|
|
|
|
* which case it should be set to keynumidx+1). */
|
|
|
|
int firstkey;
|
|
|
|
/* How many args should we skip after finding a key, in order to find the next one. */
|
|
|
|
int keystep;
|
|
|
|
} keynum;
|
|
|
|
} fk;
|
|
|
|
} keySpec;
|
|
|
|
|
Add reply_schema to command json files (internal for now) (#10273)
Work in progress towards implementing a reply schema as part of COMMAND DOCS, see #9845
Since ironing the details of the reply schema of each and every command can take a long time, we
would like to merge this PR when the infrastructure is ready, and let this mature in the unstable branch.
Meanwhile the changes of this PR are internal, they are part of the repo, but do not affect the produced build.
### Background
In #9656 we add a lot of information about Redis commands, but we are missing information about the replies
### Motivation
1. Documentation. This is the primary goal.
2. It should be possible, based on the output of COMMAND, to be able to generate client code in typed
languages. In order to do that, we need Redis to tell us, in detail, what each reply looks like.
3. We would like to build a fuzzer that verifies the reply structure (for now we use the existing
testsuite, see the "Testing" section)
### Schema
The idea is to supply some sort of schema for the various replies of each command.
The schema will describe the conceptual structure of the reply (for generated clients), as defined in RESP3.
Note that the reply structure itself may change, depending on the arguments (e.g. `XINFO STREAM`, with
and without the `FULL` modifier)
We decided to use the standard json-schema (see https://json-schema.org/) as the reply-schema.
Example for `BZPOPMIN`:
```
"reply_schema": {
"oneOf": [
{
"description": "Timeout reached and no elements were popped.",
"type": "null"
},
{
"description": "The keyname, popped member, and its score.",
"type": "array",
"minItems": 3,
"maxItems": 3,
"items": [
{
"description": "Keyname",
"type": "string"
},
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
]
}
```
#### Notes
1. It is ok that some commands' reply structure depends on the arguments and it's the caller's responsibility
to know which is the relevant one. this comes after looking at other request-reply systems like OpenAPI,
where the reply schema can also be oneOf and the caller is responsible to know which schema is the relevant one.
2. The reply schemas will describe RESP3 replies only. even though RESP3 is structured, we want to use reply
schema for documentation (and possibly to create a fuzzer that validates the replies)
3. For documentation, the description field will include an explanation of the scenario in which the reply is sent,
including any relation to arguments. for example, for `ZRANGE`'s two schemas we will need to state that one
is with `WITHSCORES` and the other is without.
4. For documentation, there will be another optional field "notes" in which we will add a short description of
the representation in RESP2, in case it's not trivial (RESP3's `ZRANGE`'s nested array vs. RESP2's flat
array, for example)
Given the above:
1. We can generate the "return" section of all commands in [redis-doc](https://redis.io/commands/)
(given that "description" and "notes" are comprehensive enough)
2. We can generate a client in a strongly typed language (but the return type could be a conceptual
`union` and the caller needs to know which schema is relevant). see the section below for RESP2 support.
3. We can create a fuzzer for RESP3.
### Limitations (because we are using the standard json-schema)
The problem is that Redis' replies are more diverse than what the json format allows. This means that,
when we convert the reply to a json (in order to validate the schema against it), we lose information (see
the "Testing" section below).
The other option would have been to extend the standard json-schema (and json format) to include stuff
like sets, bulk-strings, error-string, etc. but that would mean also extending the schema-validator - and that
seemed like too much work, so we decided to compromise.
Examples:
1. We cannot tell the difference between an "array" and a "set"
2. We cannot tell the difference between simple-string and bulk-string
3. we cannot verify true uniqueness of items in commands like ZRANGE: json-schema doesn't cover the
case of two identical members with different scores (e.g. `[["m1",6],["m1",7]]`) because `uniqueItems`
compares (member,score) tuples and not just the member name.
### Testing
This commit includes some changes inside Redis in order to verify the schemas (existing and future ones)
are indeed correct (i.e. describe the actual response of Redis).
To do that, we added a debugging feature to Redis that causes it to produce a log of all the commands
it executed and their replies.
For that, Redis needs to be compiled with `-DLOG_REQ_RES` and run with
`--reg-res-logfile <file> --client-default-resp 3` (the testsuite already does that if you run it with
`--log-req-res --force-resp3`)
You should run the testsuite with the above args (and `--dont-clean`) in order to make Redis generate
`.reqres` files (same dir as the `stdout` files) which contain request-response pairs.
These files are later on processed by `./utils/req-res-log-validator.py` which does:
1. Goes over req-res files, generated by redis-servers, spawned by the testsuite (see logreqres.c)
2. For each request-response pair, it validates the response against the request's reply_schema
(obtained from the extended COMMAND DOCS)
5. In order to get good coverage of the Redis commands, and all their different replies, we chose to use
the existing redis test suite, rather than attempt to write a fuzzer.
#### Notes about RESP2
1. We will not be able to use the testing tool to verify RESP2 replies (we are ok with that, it's time to
accept RESP3 as the future RESP)
2. Since the majority of the test suite is using RESP2, and we want the server to reply with RESP3
so that we can validate it, we will need to know how to convert the actual reply to the one expected.
- number and boolean are always strings in RESP2 so the conversion is easy
- objects (maps) are always a flat array in RESP2
- others (nested array in RESP3's `ZRANGE` and others) will need some special per-command
handling (so the client will not be totally auto-generated)
Example for ZRANGE:
```
"reply_schema": {
"anyOf": [
{
"description": "A list of member elements",
"type": "array",
"uniqueItems": true,
"items": {
"type": "string"
}
},
{
"description": "Members and their scores. Returned in case `WITHSCORES` was used.",
"notes": "In RESP2 this is returned as a flat array",
"type": "array",
"uniqueItems": true,
"items": {
"type": "array",
"minItems": 2,
"maxItems": 2,
"items": [
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
}
]
}
```
### Other changes
1. Some tests that behave differently depending on the RESP are now being tested for both RESP,
regardless of the special log-req-res mode ("Pub/Sub PING" for example)
2. Update the history field of CLIENT LIST
3. Added basic tests for commands that were not covered at all by the testsuite
### TODO
- [x] (maybe a different PR) add a "condition" field to anyOf/oneOf schemas that refers to args. e.g.
when `SET` return NULL, the condition is `arguments.get||arguments.condition`, for `OK` the condition
is `!arguments.get`, and for `string` the condition is `arguments.get` - https://github.com/redis/redis/issues/11896
- [x] (maybe a different PR) also run `runtest-cluster` in the req-res logging mode
- [x] add the new tests to GH actions (i.e. compile with `-DLOG_REQ_RES`, run the tests, and run the validator)
- [x] (maybe a different PR) figure out a way to warn about (sub)schemas that are uncovered by the output
of the tests - https://github.com/redis/redis/issues/11897
- [x] (probably a separate PR) add all missing schemas
- [x] check why "SDOWN is triggered by misconfigured instance replying with errors" fails with --log-req-res
- [x] move the response transformers to their own file (run both regular, cluster, and sentinel tests - need to
fight with the tcl including mechanism a bit)
- [x] issue: module API - https://github.com/redis/redis/issues/11898
- [x] (probably a separate PR): improve schemas: add `required` to `object`s - https://github.com/redis/redis/issues/11899
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Hanna Fadida <hanna.fadida@redislabs.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Shaya Potter <shaya@redislabs.com>
2023-03-11 09:14:16 +01:00
|
|
|
#ifdef LOG_REQ_RES
|
|
|
|
|
|
|
|
/* Must be synced with generate-command-code.py */
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
typedef enum {
|
Add reply_schema to command json files (internal for now) (#10273)
Work in progress towards implementing a reply schema as part of COMMAND DOCS, see #9845
Since ironing the details of the reply schema of each and every command can take a long time, we
would like to merge this PR when the infrastructure is ready, and let this mature in the unstable branch.
Meanwhile the changes of this PR are internal, they are part of the repo, but do not affect the produced build.
### Background
In #9656 we add a lot of information about Redis commands, but we are missing information about the replies
### Motivation
1. Documentation. This is the primary goal.
2. It should be possible, based on the output of COMMAND, to be able to generate client code in typed
languages. In order to do that, we need Redis to tell us, in detail, what each reply looks like.
3. We would like to build a fuzzer that verifies the reply structure (for now we use the existing
testsuite, see the "Testing" section)
### Schema
The idea is to supply some sort of schema for the various replies of each command.
The schema will describe the conceptual structure of the reply (for generated clients), as defined in RESP3.
Note that the reply structure itself may change, depending on the arguments (e.g. `XINFO STREAM`, with
and without the `FULL` modifier)
We decided to use the standard json-schema (see https://json-schema.org/) as the reply-schema.
Example for `BZPOPMIN`:
```
"reply_schema": {
"oneOf": [
{
"description": "Timeout reached and no elements were popped.",
"type": "null"
},
{
"description": "The keyname, popped member, and its score.",
"type": "array",
"minItems": 3,
"maxItems": 3,
"items": [
{
"description": "Keyname",
"type": "string"
},
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
]
}
```
#### Notes
1. It is ok that some commands' reply structure depends on the arguments and it's the caller's responsibility
to know which is the relevant one. this comes after looking at other request-reply systems like OpenAPI,
where the reply schema can also be oneOf and the caller is responsible to know which schema is the relevant one.
2. The reply schemas will describe RESP3 replies only. even though RESP3 is structured, we want to use reply
schema for documentation (and possibly to create a fuzzer that validates the replies)
3. For documentation, the description field will include an explanation of the scenario in which the reply is sent,
including any relation to arguments. for example, for `ZRANGE`'s two schemas we will need to state that one
is with `WITHSCORES` and the other is without.
4. For documentation, there will be another optional field "notes" in which we will add a short description of
the representation in RESP2, in case it's not trivial (RESP3's `ZRANGE`'s nested array vs. RESP2's flat
array, for example)
Given the above:
1. We can generate the "return" section of all commands in [redis-doc](https://redis.io/commands/)
(given that "description" and "notes" are comprehensive enough)
2. We can generate a client in a strongly typed language (but the return type could be a conceptual
`union` and the caller needs to know which schema is relevant). see the section below for RESP2 support.
3. We can create a fuzzer for RESP3.
### Limitations (because we are using the standard json-schema)
The problem is that Redis' replies are more diverse than what the json format allows. This means that,
when we convert the reply to a json (in order to validate the schema against it), we lose information (see
the "Testing" section below).
The other option would have been to extend the standard json-schema (and json format) to include stuff
like sets, bulk-strings, error-string, etc. but that would mean also extending the schema-validator - and that
seemed like too much work, so we decided to compromise.
Examples:
1. We cannot tell the difference between an "array" and a "set"
2. We cannot tell the difference between simple-string and bulk-string
3. we cannot verify true uniqueness of items in commands like ZRANGE: json-schema doesn't cover the
case of two identical members with different scores (e.g. `[["m1",6],["m1",7]]`) because `uniqueItems`
compares (member,score) tuples and not just the member name.
### Testing
This commit includes some changes inside Redis in order to verify the schemas (existing and future ones)
are indeed correct (i.e. describe the actual response of Redis).
To do that, we added a debugging feature to Redis that causes it to produce a log of all the commands
it executed and their replies.
For that, Redis needs to be compiled with `-DLOG_REQ_RES` and run with
`--reg-res-logfile <file> --client-default-resp 3` (the testsuite already does that if you run it with
`--log-req-res --force-resp3`)
You should run the testsuite with the above args (and `--dont-clean`) in order to make Redis generate
`.reqres` files (same dir as the `stdout` files) which contain request-response pairs.
These files are later on processed by `./utils/req-res-log-validator.py` which does:
1. Goes over req-res files, generated by redis-servers, spawned by the testsuite (see logreqres.c)
2. For each request-response pair, it validates the response against the request's reply_schema
(obtained from the extended COMMAND DOCS)
5. In order to get good coverage of the Redis commands, and all their different replies, we chose to use
the existing redis test suite, rather than attempt to write a fuzzer.
#### Notes about RESP2
1. We will not be able to use the testing tool to verify RESP2 replies (we are ok with that, it's time to
accept RESP3 as the future RESP)
2. Since the majority of the test suite is using RESP2, and we want the server to reply with RESP3
so that we can validate it, we will need to know how to convert the actual reply to the one expected.
- number and boolean are always strings in RESP2 so the conversion is easy
- objects (maps) are always a flat array in RESP2
- others (nested array in RESP3's `ZRANGE` and others) will need some special per-command
handling (so the client will not be totally auto-generated)
Example for ZRANGE:
```
"reply_schema": {
"anyOf": [
{
"description": "A list of member elements",
"type": "array",
"uniqueItems": true,
"items": {
"type": "string"
}
},
{
"description": "Members and their scores. Returned in case `WITHSCORES` was used.",
"notes": "In RESP2 this is returned as a flat array",
"type": "array",
"uniqueItems": true,
"items": {
"type": "array",
"minItems": 2,
"maxItems": 2,
"items": [
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
}
]
}
```
### Other changes
1. Some tests that behave differently depending on the RESP are now being tested for both RESP,
regardless of the special log-req-res mode ("Pub/Sub PING" for example)
2. Update the history field of CLIENT LIST
3. Added basic tests for commands that were not covered at all by the testsuite
### TODO
- [x] (maybe a different PR) add a "condition" field to anyOf/oneOf schemas that refers to args. e.g.
when `SET` return NULL, the condition is `arguments.get||arguments.condition`, for `OK` the condition
is `!arguments.get`, and for `string` the condition is `arguments.get` - https://github.com/redis/redis/issues/11896
- [x] (maybe a different PR) also run `runtest-cluster` in the req-res logging mode
- [x] add the new tests to GH actions (i.e. compile with `-DLOG_REQ_RES`, run the tests, and run the validator)
- [x] (maybe a different PR) figure out a way to warn about (sub)schemas that are uncovered by the output
of the tests - https://github.com/redis/redis/issues/11897
- [x] (probably a separate PR) add all missing schemas
- [x] check why "SDOWN is triggered by misconfigured instance replying with errors" fails with --log-req-res
- [x] move the response transformers to their own file (run both regular, cluster, and sentinel tests - need to
fight with the tcl including mechanism a bit)
- [x] issue: module API - https://github.com/redis/redis/issues/11898
- [x] (probably a separate PR): improve schemas: add `required` to `object`s - https://github.com/redis/redis/issues/11899
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Hanna Fadida <hanna.fadida@redislabs.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Shaya Potter <shaya@redislabs.com>
2023-03-11 09:14:16 +01:00
|
|
|
JSON_TYPE_STRING,
|
|
|
|
JSON_TYPE_INTEGER,
|
|
|
|
JSON_TYPE_BOOLEAN,
|
|
|
|
JSON_TYPE_OBJECT,
|
|
|
|
JSON_TYPE_ARRAY,
|
|
|
|
} jsonType;
|
|
|
|
|
|
|
|
typedef struct jsonObjectElement {
|
|
|
|
jsonType type;
|
|
|
|
const char *key;
|
|
|
|
union {
|
|
|
|
const char *string;
|
|
|
|
long long integer;
|
|
|
|
int boolean;
|
|
|
|
struct jsonObject *object;
|
|
|
|
struct {
|
|
|
|
struct jsonObject **objects;
|
|
|
|
int length;
|
|
|
|
} array;
|
|
|
|
} value;
|
|
|
|
} jsonObjectElement;
|
|
|
|
|
|
|
|
typedef struct jsonObject {
|
|
|
|
struct jsonObjectElement *elements;
|
|
|
|
int length;
|
|
|
|
} jsonObject;
|
|
|
|
|
|
|
|
#endif
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
|
2024-04-05 16:59:55 -07:00
|
|
|
/* WARNING! This struct must match ValkeyModuleCommandHistoryEntry */
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
typedef struct {
|
|
|
|
const char *since;
|
|
|
|
const char *changes;
|
|
|
|
} commandHistory;
|
|
|
|
|
|
|
|
/* Must be synced with COMMAND_GROUP_STR and generate-command-code.py */
|
|
|
|
typedef enum {
|
|
|
|
COMMAND_GROUP_GENERIC,
|
|
|
|
COMMAND_GROUP_STRING,
|
|
|
|
COMMAND_GROUP_LIST,
|
|
|
|
COMMAND_GROUP_SET,
|
|
|
|
COMMAND_GROUP_SORTED_SET,
|
|
|
|
COMMAND_GROUP_HASH,
|
|
|
|
COMMAND_GROUP_PUBSUB,
|
|
|
|
COMMAND_GROUP_TRANSACTIONS,
|
|
|
|
COMMAND_GROUP_CONNECTION,
|
|
|
|
COMMAND_GROUP_SERVER,
|
|
|
|
COMMAND_GROUP_SCRIPTING,
|
|
|
|
COMMAND_GROUP_HYPERLOGLOG,
|
|
|
|
COMMAND_GROUP_CLUSTER,
|
|
|
|
COMMAND_GROUP_SENTINEL,
|
|
|
|
COMMAND_GROUP_GEO,
|
|
|
|
COMMAND_GROUP_STREAM,
|
|
|
|
COMMAND_GROUP_BITMAP,
|
|
|
|
COMMAND_GROUP_MODULE,
|
2024-04-03 23:29:59 +07:00
|
|
|
} serverCommandGroup;
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
|
2024-04-03 23:33:33 +07:00
|
|
|
typedef void serverCommandProc(client *c);
|
2024-04-03 23:54:33 +07:00
|
|
|
typedef int serverGetKeysProc(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* Command structure.
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
*
|
|
|
|
* Note that the command table is in commands.c and it is auto-generated.
|
|
|
|
*
|
|
|
|
* This is the meaning of the flags:
|
|
|
|
*
|
|
|
|
* CMD_WRITE: Write command (may modify the key space).
|
|
|
|
*
|
|
|
|
* CMD_READONLY: Commands just reading from keys without changing the content.
|
|
|
|
* Note that commands that don't read from the keyspace such as
|
|
|
|
* TIME, SELECT, INFO, administrative commands, and connection
|
|
|
|
* or transaction related commands (multi, exec, discard, ...)
|
|
|
|
* are not flagged as read-only commands, since they affect the
|
|
|
|
* server or the connection in other ways.
|
|
|
|
*
|
|
|
|
* CMD_DENYOOM: May increase memory usage once called. Don't allow if out
|
|
|
|
* of memory.
|
|
|
|
*
|
|
|
|
* CMD_ADMIN: Administrative command, like SAVE or SHUTDOWN.
|
|
|
|
*
|
|
|
|
* CMD_PUBSUB: Pub/Sub related command.
|
|
|
|
*
|
|
|
|
* CMD_NOSCRIPT: Command not allowed in scripts.
|
|
|
|
*
|
2022-01-20 10:32:11 +01:00
|
|
|
* CMD_BLOCKING: The command has the potential to block the client.
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
*
|
|
|
|
* CMD_LOADING: Allow the command while loading the database.
|
|
|
|
*
|
Allow most CONFIG SET during loading, block some commands in async-loading (#9878)
## background
Till now CONFIG SET was blocked during loading.
(In the not so distant past, GET was disallowed too)
We recently (not released yet) added an async-loading mode, see #9323,
and during that time it'll serve CONFIG SET and any other command.
And now we realized (#9770) that some configs, and commands are dangerous
during async-loading.
## changes
* Allow most CONFIG SET during loading (both on async-loading and normal loading)
* Allow CONFIG REWRITE and CONFIG RESETSTAT during loading
* Block a few config during loading (`appendonly`, `repl-diskless-load`, and `dir`)
* Block a few commands during loading (list below)
## the blocked commands:
* SAVE - obviously we don't wanna start a foregreound save during loading 8-)
* BGSAVE - we don't mind to schedule one, but we don't wanna fork now
* BGREWRITEAOF - we don't mind to schedule one, but we don't wanna fork now
* MODULE - we obviously don't wanna unload a module during replication / rdb loading
(MODULE HELP and MODULE LIST are not blocked)
* SYNC / PSYNC - we're in the middle of RDB loading from master, must not allow sync
requests now.
* REPLICAOF / SLAVEOF - we're in the middle of replicating, maybe it makes sense to let
the user abort it, but he couldn't do that so far, i don't wanna take any risk of bugs due to odd state.
* CLUSTER - only allow [HELP, SLOTS, NODES, INFO, MYID, LINKS, KEYSLOT, COUNTKEYSINSLOT,
GETKEYSINSLOT, RESET, REPLICAS, COUNT_FAILURE_REPORTS], for others, preserve the status quo
## other fixes
* processEventsWhileBlocked had an issue when being nested, this could happen with a busy script
during async loading (new), but also in a busy script during AOF loading (old). this lead to a crash in
the scenario described in #6988
2021-12-22 14:11:16 +02:00
|
|
|
* CMD_NO_ASYNC_LOADING: Deny during async loading (when a replica uses diskless
|
2022-01-04 12:37:47 +01:00
|
|
|
* sync swapdb, and allows access to the old dataset)
|
Allow most CONFIG SET during loading, block some commands in async-loading (#9878)
## background
Till now CONFIG SET was blocked during loading.
(In the not so distant past, GET was disallowed too)
We recently (not released yet) added an async-loading mode, see #9323,
and during that time it'll serve CONFIG SET and any other command.
And now we realized (#9770) that some configs, and commands are dangerous
during async-loading.
## changes
* Allow most CONFIG SET during loading (both on async-loading and normal loading)
* Allow CONFIG REWRITE and CONFIG RESETSTAT during loading
* Block a few config during loading (`appendonly`, `repl-diskless-load`, and `dir`)
* Block a few commands during loading (list below)
## the blocked commands:
* SAVE - obviously we don't wanna start a foregreound save during loading 8-)
* BGSAVE - we don't mind to schedule one, but we don't wanna fork now
* BGREWRITEAOF - we don't mind to schedule one, but we don't wanna fork now
* MODULE - we obviously don't wanna unload a module during replication / rdb loading
(MODULE HELP and MODULE LIST are not blocked)
* SYNC / PSYNC - we're in the middle of RDB loading from master, must not allow sync
requests now.
* REPLICAOF / SLAVEOF - we're in the middle of replicating, maybe it makes sense to let
the user abort it, but he couldn't do that so far, i don't wanna take any risk of bugs due to odd state.
* CLUSTER - only allow [HELP, SLOTS, NODES, INFO, MYID, LINKS, KEYSLOT, COUNTKEYSINSLOT,
GETKEYSINSLOT, RESET, REPLICAS, COUNT_FAILURE_REPORTS], for others, preserve the status quo
## other fixes
* processEventsWhileBlocked had an issue when being nested, this could happen with a busy script
during async loading (new), but also in a busy script during AOF loading (old). this lead to a crash in
the scenario described in #6988
2021-12-22 14:11:16 +02:00
|
|
|
*
|
2024-06-07 14:21:33 -07:00
|
|
|
* CMD_STALE: Allow the command while a replica has stale data but is not
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
* allowed to serve this data. Normally no command is accepted
|
|
|
|
* in this condition but just a few.
|
|
|
|
*
|
|
|
|
* CMD_SKIP_MONITOR: Do not automatically propagate the command on MONITOR.
|
|
|
|
*
|
|
|
|
* CMD_SKIP_SLOWLOG: Do not automatically propagate the command to the slowlog.
|
|
|
|
*
|
|
|
|
* CMD_ASKING: Perform an implicit ASKING for this command, so the
|
|
|
|
* command will be accepted in cluster mode if the slot is marked
|
|
|
|
* as 'importing'.
|
|
|
|
*
|
|
|
|
* CMD_FAST: Fast command: O(1) or O(log(N)) command that should never
|
|
|
|
* delay its execution as long as the kernel scheduler is giving
|
|
|
|
* us time. Note that commands that may trigger a DEL as a side
|
|
|
|
* effect (like SET) are not fast commands.
|
|
|
|
*
|
|
|
|
* CMD_NO_AUTH: Command doesn't require authentication
|
|
|
|
*
|
|
|
|
* CMD_MAY_REPLICATE: Command may produce replication traffic, but should be
|
|
|
|
* allowed under circumstances where write commands are disallowed.
|
|
|
|
* Examples include PUBLISH, which replicates pubsub messages,and
|
|
|
|
* EVAL, which may execute write commands, which are replicated,
|
|
|
|
* or may just execute read commands. A command can not be marked
|
|
|
|
* both CMD_WRITE and CMD_MAY_REPLICATE
|
|
|
|
*
|
2022-06-30 21:32:40 +08:00
|
|
|
* CMD_SENTINEL: This command is present in sentinel mode.
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
*
|
2022-06-30 21:32:40 +08:00
|
|
|
* CMD_ONLY_SENTINEL: This command is present only when in sentinel mode.
|
|
|
|
* And should be removed from redis.
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
*
|
|
|
|
* CMD_NO_MANDATORY_KEYS: This key arguments for this command are optional.
|
|
|
|
*
|
2022-07-07 11:31:59 +08:00
|
|
|
* CMD_NO_MULTI: The command is not allowed inside a transaction
|
2022-01-04 12:37:47 +01:00
|
|
|
*
|
2023-02-14 08:33:21 +01:00
|
|
|
* CMD_ALLOW_BUSY: The command can run while another command is running for
|
|
|
|
* a long time (timedout script, module command that yields)
|
|
|
|
*
|
|
|
|
* CMD_TOUCHES_ARBITRARY_KEYS: The command may touch (and cause lazy-expire)
|
|
|
|
* arbitrary key (i.e not provided in argv)
|
|
|
|
*
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
* The following additional flags are only used in order to put commands
|
|
|
|
* in a specific ACL category. Commands can have multiple ACL categories.
|
2024-04-03 10:47:26 -07:00
|
|
|
* See valkey.conf for the exact meaning of each.
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
*
|
|
|
|
* @keyspace, @read, @write, @set, @sortedset, @list, @hash, @string, @bitmap,
|
|
|
|
* @hyperloglog, @stream, @admin, @fast, @slow, @pubsub, @blocking, @dangerous,
|
|
|
|
* @connection, @transaction, @scripting, @geo.
|
|
|
|
*
|
|
|
|
* Note that:
|
|
|
|
*
|
|
|
|
* 1) The read-only flag implies the @read ACL category.
|
|
|
|
* 2) The write flag implies the @write ACL category.
|
|
|
|
* 3) The fast flag implies the @fast ACL category.
|
|
|
|
* 4) The admin flag implies the @admin and @dangerous ACL category.
|
|
|
|
* 5) The pub-sub flag implies the @pubsub ACL category.
|
|
|
|
* 6) The lack of fast flag implies the @slow ACL category.
|
|
|
|
* 7) The non obvious "keyspace" category includes the commands
|
|
|
|
* that interact with keys without having anything to do with
|
|
|
|
* specific data structures, such as: DEL, RENAME, MOVE, SELECT,
|
|
|
|
* TYPE, EXPIRE*, PEXPIRE*, TTL, PTTL, ...
|
|
|
|
*/
|
2024-04-03 23:54:33 +07:00
|
|
|
struct serverCommand {
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
/* Declarative data */
|
2024-05-22 23:24:12 -07:00
|
|
|
const char *declared_name; /* A string representing the command declared_name.
|
|
|
|
* It is a const char * for native commands and SDS for module commands. */
|
|
|
|
const char *summary; /* Summary of the command (optional). */
|
|
|
|
const char *complexity; /* Complexity description (optional). */
|
|
|
|
const char *since; /* Debut version of the command (optional). */
|
|
|
|
int doc_flags; /* Flags for documentation (see CMD_DOC_*). */
|
|
|
|
const char *replaced_by; /* In case the command is deprecated, this is the successor command. */
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
const char *deprecated_since; /* In case the command is deprecated, when did it happen? */
|
2024-05-22 23:24:12 -07:00
|
|
|
serverCommandGroup group; /* Command group */
|
|
|
|
commandHistory *history; /* History of the command */
|
Reimplement cli hints based on command arg docs (#10515)
Now that the command argument specs are available at runtime (#9656), this PR addresses
#8084 by implementing a complete solution for command-line hinting in `redis-cli`.
It correctly handles nearly every case in Redis's complex command argument definitions, including
`BLOCK` and `ONEOF` arguments, reordering of optional arguments, and repeated arguments
(even when followed by mandatory arguments). It also validates numerically-typed arguments.
It may not correctly handle all possible combinations of those, but overall it is quite robust.
Arguments are only matched after the space bar is typed, so partial word matching is not
supported - that proved to be more confusing than helpful. When the user's current input
cannot be matched against the argument specs, hinting is disabled.
Partial support has been implemented for legacy (pre-7.0) servers that do not support
`COMMAND DOCS`, by falling back to a statically-compiled command argument table.
On startup, if the server does not support `COMMAND DOCS`, `redis-cli` will now issue
an `INFO SERVER` command to retrieve the server version (unless `HELLO` has already
been sent, in which case the server version will be extracted from the reply to `HELLO`).
The server version will be used to filter the commands and arguments in the command table,
removing those not supported by that version of the server. However, the static table only
includes core Redis commands, so with a legacy server hinting will not be supported for
module commands. The auto generated help.h and the scripts that generates it are gone.
Command and argument tables for the server and CLI use different structs, due primarily
to the need to support different runtime data. In order to generate code for both, macros
have been added to `commands.def` (previously `commands.c`) to make it possible to
configure the code generation differently for different use cases (one linked with redis-server,
and one with redis-cli).
Also adding a basic testing framework for the command hints based on new (undocumented)
command line options to `redis-cli`: `--test_hint 'INPUT'` prints out the command-line hint for
a given input string, and `--test_hint_file <filename>` runs a suite of test cases for the hinting
mechanism. The test suite is in `tests/assets/test_cli_hint_suite.txt`, and it is run from
`tests/integration/redis-cli.tcl`.
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2023-03-30 19:03:56 +03:00
|
|
|
int num_history;
|
2022-01-20 10:32:11 +01:00
|
|
|
const char **tips; /* An array of strings that are meant to be tips for clients/proxies regarding this command */
|
Reimplement cli hints based on command arg docs (#10515)
Now that the command argument specs are available at runtime (#9656), this PR addresses
#8084 by implementing a complete solution for command-line hinting in `redis-cli`.
It correctly handles nearly every case in Redis's complex command argument definitions, including
`BLOCK` and `ONEOF` arguments, reordering of optional arguments, and repeated arguments
(even when followed by mandatory arguments). It also validates numerically-typed arguments.
It may not correctly handle all possible combinations of those, but overall it is quite robust.
Arguments are only matched after the space bar is typed, so partial word matching is not
supported - that proved to be more confusing than helpful. When the user's current input
cannot be matched against the argument specs, hinting is disabled.
Partial support has been implemented for legacy (pre-7.0) servers that do not support
`COMMAND DOCS`, by falling back to a statically-compiled command argument table.
On startup, if the server does not support `COMMAND DOCS`, `redis-cli` will now issue
an `INFO SERVER` command to retrieve the server version (unless `HELLO` has already
been sent, in which case the server version will be extracted from the reply to `HELLO`).
The server version will be used to filter the commands and arguments in the command table,
removing those not supported by that version of the server. However, the static table only
includes core Redis commands, so with a legacy server hinting will not be supported for
module commands. The auto generated help.h and the scripts that generates it are gone.
Command and argument tables for the server and CLI use different structs, due primarily
to the need to support different runtime data. In order to generate code for both, macros
have been added to `commands.def` (previously `commands.c`) to make it possible to
configure the code generation differently for different use cases (one linked with redis-server,
and one with redis-cli).
Also adding a basic testing framework for the command hints based on new (undocumented)
command line options to `redis-cli`: `--test_hint 'INPUT'` prints out the command-line hint for
a given input string, and `--test_hint_file <filename>` runs a suite of test cases for the hinting
mechanism. The test suite is in `tests/assets/test_cli_hint_suite.txt`, and it is run from
`tests/integration/redis-cli.tcl`.
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2023-03-30 19:03:56 +03:00
|
|
|
int num_tips;
|
2024-04-03 23:33:33 +07:00
|
|
|
serverCommandProc *proc; /* Command implementation */
|
2024-05-22 23:24:12 -07:00
|
|
|
int arity; /* Number of arguments, it is possible to use -N to say >= N */
|
|
|
|
uint64_t flags; /* Command flags, see CMD_*. */
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
uint64_t acl_categories; /* ACl categories, see ACL_CATEGORY_*. */
|
Reimplement cli hints based on command arg docs (#10515)
Now that the command argument specs are available at runtime (#9656), this PR addresses
#8084 by implementing a complete solution for command-line hinting in `redis-cli`.
It correctly handles nearly every case in Redis's complex command argument definitions, including
`BLOCK` and `ONEOF` arguments, reordering of optional arguments, and repeated arguments
(even when followed by mandatory arguments). It also validates numerically-typed arguments.
It may not correctly handle all possible combinations of those, but overall it is quite robust.
Arguments are only matched after the space bar is typed, so partial word matching is not
supported - that proved to be more confusing than helpful. When the user's current input
cannot be matched against the argument specs, hinting is disabled.
Partial support has been implemented for legacy (pre-7.0) servers that do not support
`COMMAND DOCS`, by falling back to a statically-compiled command argument table.
On startup, if the server does not support `COMMAND DOCS`, `redis-cli` will now issue
an `INFO SERVER` command to retrieve the server version (unless `HELLO` has already
been sent, in which case the server version will be extracted from the reply to `HELLO`).
The server version will be used to filter the commands and arguments in the command table,
removing those not supported by that version of the server. However, the static table only
includes core Redis commands, so with a legacy server hinting will not be supported for
module commands. The auto generated help.h and the scripts that generates it are gone.
Command and argument tables for the server and CLI use different structs, due primarily
to the need to support different runtime data. In order to generate code for both, macros
have been added to `commands.def` (previously `commands.c`) to make it possible to
configure the code generation differently for different use cases (one linked with redis-server,
and one with redis-cli).
Also adding a basic testing framework for the command hints based on new (undocumented)
command line options to `redis-cli`: `--test_hint 'INPUT'` prints out the command-line hint for
a given input string, and `--test_hint_file <filename>` runs a suite of test cases for the hinting
mechanism. The test suite is in `tests/assets/test_cli_hint_suite.txt`, and it is run from
`tests/integration/redis-cli.tcl`.
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2023-03-30 19:03:56 +03:00
|
|
|
keySpec *key_specs;
|
|
|
|
int key_specs_num;
|
2011-03-23 18:09:17 +01:00
|
|
|
/* Use a function to determine keys arguments in a command line.
|
2024-04-09 01:24:03 -07:00
|
|
|
* Used for Cluster redirect (may be NULL) */
|
2024-04-03 23:33:33 +07:00
|
|
|
serverGetKeysProc *getkeys_proc;
|
Reimplement cli hints based on command arg docs (#10515)
Now that the command argument specs are available at runtime (#9656), this PR addresses
#8084 by implementing a complete solution for command-line hinting in `redis-cli`.
It correctly handles nearly every case in Redis's complex command argument definitions, including
`BLOCK` and `ONEOF` arguments, reordering of optional arguments, and repeated arguments
(even when followed by mandatory arguments). It also validates numerically-typed arguments.
It may not correctly handle all possible combinations of those, but overall it is quite robust.
Arguments are only matched after the space bar is typed, so partial word matching is not
supported - that proved to be more confusing than helpful. When the user's current input
cannot be matched against the argument specs, hinting is disabled.
Partial support has been implemented for legacy (pre-7.0) servers that do not support
`COMMAND DOCS`, by falling back to a statically-compiled command argument table.
On startup, if the server does not support `COMMAND DOCS`, `redis-cli` will now issue
an `INFO SERVER` command to retrieve the server version (unless `HELLO` has already
been sent, in which case the server version will be extracted from the reply to `HELLO`).
The server version will be used to filter the commands and arguments in the command table,
removing those not supported by that version of the server. However, the static table only
includes core Redis commands, so with a legacy server hinting will not be supported for
module commands. The auto generated help.h and the scripts that generates it are gone.
Command and argument tables for the server and CLI use different structs, due primarily
to the need to support different runtime data. In order to generate code for both, macros
have been added to `commands.def` (previously `commands.c`) to make it possible to
configure the code generation differently for different use cases (one linked with redis-server,
and one with redis-cli).
Also adding a basic testing framework for the command hints based on new (undocumented)
command line options to `redis-cli`: `--test_hint 'INPUT'` prints out the command-line hint for
a given input string, and `--test_hint_file <filename>` runs a suite of test cases for the hinting
mechanism. The test suite is in `tests/assets/test_cli_hint_suite.txt`, and it is run from
`tests/integration/redis-cli.tcl`.
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2023-03-30 19:03:56 +03:00
|
|
|
int num_args; /* Length of args array. */
|
Treat subcommands as commands (#9504)
## Intro
The purpose is to allow having different flags/ACL categories for
subcommands (Example: CONFIG GET is ok-loading but CONFIG SET isn't)
We create a small command table for every command that has subcommands
and each subcommand has its own flags, etc. (same as a "regular" command)
This commit also unites the Redis and the Sentinel command tables
## Affected commands
CONFIG
Used to have "admin ok-loading ok-stale no-script"
Changes:
1. Dropped "ok-loading" in all except GET (this doesn't change behavior since
there were checks in the code doing that)
XINFO
Used to have "read-only random"
Changes:
1. Dropped "random" in all except CONSUMERS
XGROUP
Used to have "write use-memory"
Changes:
1. Dropped "use-memory" in all except CREATE and CREATECONSUMER
COMMAND
No changes.
MEMORY
Used to have "random read-only"
Changes:
1. Dropped "random" in PURGE and USAGE
ACL
Used to have "admin no-script ok-loading ok-stale"
Changes:
1. Dropped "admin" in WHOAMI, GENPASS, and CAT
LATENCY
No changes.
MODULE
No changes.
SLOWLOG
Used to have "admin random ok-loading ok-stale"
Changes:
1. Dropped "random" in RESET
OBJECT
Used to have "read-only random"
Changes:
1. Dropped "random" in ENCODING and REFCOUNT
SCRIPT
Used to have "may-replicate no-script"
Changes:
1. Dropped "may-replicate" in all except FLUSH and LOAD
CLIENT
Used to have "admin no-script random ok-loading ok-stale"
Changes:
1. Dropped "random" in all except INFO and LIST
2. Dropped "admin" in ID, TRACKING, CACHING, GETREDIR, INFO, SETNAME, GETNAME, and REPLY
STRALGO
No changes.
PUBSUB
No changes.
CLUSTER
Changes:
1. Dropped "admin in countkeysinslots, getkeysinslot, info, nodes, keyslot, myid, and slots
SENTINEL
No changes.
(note that DEBUG also fits, but we decided not to convert it since it's for
debugging and anyway undocumented)
## New sub-command
This commit adds another element to the per-command output of COMMAND,
describing the list of subcommands, if any (in the same structure as "regular" commands)
Also, it adds a new subcommand:
```
COMMAND LIST [FILTERBY (MODULE <module-name>|ACLCAT <cat>|PATTERN <pattern>)]
```
which returns a set of all commands (unless filters), but excluding subcommands.
## Module API
A new module API, RM_CreateSubcommand, was added, in order to allow
module writer to define subcommands
## ACL changes:
1. Now, that each subcommand is actually a command, each has its own ACL id.
2. The old mechanism of allowed_subcommands is redundant
(blocking/allowing a subcommand is the same as blocking/allowing a regular command),
but we had to keep it, to support the widespread usage of allowed_subcommands
to block commands with certain args, that aren't subcommands (e.g. "-select +select|0").
3. I have renamed allowed_subcommands to allowed_firstargs to emphasize the difference.
4. Because subcommands are commands in ACL too, you can now use "-" to block subcommands
(e.g. "+client -client|kill"), which wasn't possible in the past.
5. It is also possible to use the allowed_firstargs mechanism with subcommand.
For example: `+config -config|set +config|set|loglevel` will block all CONFIG SET except
for setting the log level.
6. All of the ACL changes above required some amount of refactoring.
## Misc
1. There are two approaches: Either each subcommand has its own function or all
subcommands use the same function, determining what to do according to argv[0].
For now, I took the former approaches only with CONFIG and COMMAND,
while other commands use the latter approach (for smaller blamelog diff).
2. Deleted memoryGetKeys: It is no longer needed because MEMORY USAGE now uses the "range" key spec.
4. Bugfix: GETNAME was missing from CLIENT's help message.
5. Sentinel and Redis now use the same table, with the same function pointer.
Some commands have a different implementation in Sentinel, so we redirect
them (these are ROLE, PUBLISH, and INFO).
6. Command stats now show the stats per subcommand (e.g. instead of stats just
for "config" you will have stats for "config|set", "config|get", etc.)
7. It is now possible to use COMMAND directly on subcommands:
COMMAND INFO CONFIG|GET (The pipeline syntax was inspired from ACL, and
can be used in functions lookupCommandBySds and lookupCommandByCString)
8. STRALGO is now a container command (has "help")
## Breaking changes:
1. Command stats now show the stats per subcommand (see (5) above)
2021-10-20 10:52:57 +02:00
|
|
|
/* Array of subcommands (may be NULL) */
|
2024-04-03 23:54:33 +07:00
|
|
|
struct serverCommand *subcommands;
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
/* Array of arguments (may be NULL) */
|
2024-04-04 01:33:38 +07:00
|
|
|
struct serverCommandArg *args;
|
Add reply_schema to command json files (internal for now) (#10273)
Work in progress towards implementing a reply schema as part of COMMAND DOCS, see #9845
Since ironing the details of the reply schema of each and every command can take a long time, we
would like to merge this PR when the infrastructure is ready, and let this mature in the unstable branch.
Meanwhile the changes of this PR are internal, they are part of the repo, but do not affect the produced build.
### Background
In #9656 we add a lot of information about Redis commands, but we are missing information about the replies
### Motivation
1. Documentation. This is the primary goal.
2. It should be possible, based on the output of COMMAND, to be able to generate client code in typed
languages. In order to do that, we need Redis to tell us, in detail, what each reply looks like.
3. We would like to build a fuzzer that verifies the reply structure (for now we use the existing
testsuite, see the "Testing" section)
### Schema
The idea is to supply some sort of schema for the various replies of each command.
The schema will describe the conceptual structure of the reply (for generated clients), as defined in RESP3.
Note that the reply structure itself may change, depending on the arguments (e.g. `XINFO STREAM`, with
and without the `FULL` modifier)
We decided to use the standard json-schema (see https://json-schema.org/) as the reply-schema.
Example for `BZPOPMIN`:
```
"reply_schema": {
"oneOf": [
{
"description": "Timeout reached and no elements were popped.",
"type": "null"
},
{
"description": "The keyname, popped member, and its score.",
"type": "array",
"minItems": 3,
"maxItems": 3,
"items": [
{
"description": "Keyname",
"type": "string"
},
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
]
}
```
#### Notes
1. It is ok that some commands' reply structure depends on the arguments and it's the caller's responsibility
to know which is the relevant one. this comes after looking at other request-reply systems like OpenAPI,
where the reply schema can also be oneOf and the caller is responsible to know which schema is the relevant one.
2. The reply schemas will describe RESP3 replies only. even though RESP3 is structured, we want to use reply
schema for documentation (and possibly to create a fuzzer that validates the replies)
3. For documentation, the description field will include an explanation of the scenario in which the reply is sent,
including any relation to arguments. for example, for `ZRANGE`'s two schemas we will need to state that one
is with `WITHSCORES` and the other is without.
4. For documentation, there will be another optional field "notes" in which we will add a short description of
the representation in RESP2, in case it's not trivial (RESP3's `ZRANGE`'s nested array vs. RESP2's flat
array, for example)
Given the above:
1. We can generate the "return" section of all commands in [redis-doc](https://redis.io/commands/)
(given that "description" and "notes" are comprehensive enough)
2. We can generate a client in a strongly typed language (but the return type could be a conceptual
`union` and the caller needs to know which schema is relevant). see the section below for RESP2 support.
3. We can create a fuzzer for RESP3.
### Limitations (because we are using the standard json-schema)
The problem is that Redis' replies are more diverse than what the json format allows. This means that,
when we convert the reply to a json (in order to validate the schema against it), we lose information (see
the "Testing" section below).
The other option would have been to extend the standard json-schema (and json format) to include stuff
like sets, bulk-strings, error-string, etc. but that would mean also extending the schema-validator - and that
seemed like too much work, so we decided to compromise.
Examples:
1. We cannot tell the difference between an "array" and a "set"
2. We cannot tell the difference between simple-string and bulk-string
3. we cannot verify true uniqueness of items in commands like ZRANGE: json-schema doesn't cover the
case of two identical members with different scores (e.g. `[["m1",6],["m1",7]]`) because `uniqueItems`
compares (member,score) tuples and not just the member name.
### Testing
This commit includes some changes inside Redis in order to verify the schemas (existing and future ones)
are indeed correct (i.e. describe the actual response of Redis).
To do that, we added a debugging feature to Redis that causes it to produce a log of all the commands
it executed and their replies.
For that, Redis needs to be compiled with `-DLOG_REQ_RES` and run with
`--reg-res-logfile <file> --client-default-resp 3` (the testsuite already does that if you run it with
`--log-req-res --force-resp3`)
You should run the testsuite with the above args (and `--dont-clean`) in order to make Redis generate
`.reqres` files (same dir as the `stdout` files) which contain request-response pairs.
These files are later on processed by `./utils/req-res-log-validator.py` which does:
1. Goes over req-res files, generated by redis-servers, spawned by the testsuite (see logreqres.c)
2. For each request-response pair, it validates the response against the request's reply_schema
(obtained from the extended COMMAND DOCS)
5. In order to get good coverage of the Redis commands, and all their different replies, we chose to use
the existing redis test suite, rather than attempt to write a fuzzer.
#### Notes about RESP2
1. We will not be able to use the testing tool to verify RESP2 replies (we are ok with that, it's time to
accept RESP3 as the future RESP)
2. Since the majority of the test suite is using RESP2, and we want the server to reply with RESP3
so that we can validate it, we will need to know how to convert the actual reply to the one expected.
- number and boolean are always strings in RESP2 so the conversion is easy
- objects (maps) are always a flat array in RESP2
- others (nested array in RESP3's `ZRANGE` and others) will need some special per-command
handling (so the client will not be totally auto-generated)
Example for ZRANGE:
```
"reply_schema": {
"anyOf": [
{
"description": "A list of member elements",
"type": "array",
"uniqueItems": true,
"items": {
"type": "string"
}
},
{
"description": "Members and their scores. Returned in case `WITHSCORES` was used.",
"notes": "In RESP2 this is returned as a flat array",
"type": "array",
"uniqueItems": true,
"items": {
"type": "array",
"minItems": 2,
"maxItems": 2,
"items": [
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
}
]
}
```
### Other changes
1. Some tests that behave differently depending on the RESP are now being tested for both RESP,
regardless of the special log-req-res mode ("Pub/Sub PING" for example)
2. Update the history field of CLIENT LIST
3. Added basic tests for commands that were not covered at all by the testsuite
### TODO
- [x] (maybe a different PR) add a "condition" field to anyOf/oneOf schemas that refers to args. e.g.
when `SET` return NULL, the condition is `arguments.get||arguments.condition`, for `OK` the condition
is `!arguments.get`, and for `string` the condition is `arguments.get` - https://github.com/redis/redis/issues/11896
- [x] (maybe a different PR) also run `runtest-cluster` in the req-res logging mode
- [x] add the new tests to GH actions (i.e. compile with `-DLOG_REQ_RES`, run the tests, and run the validator)
- [x] (maybe a different PR) figure out a way to warn about (sub)schemas that are uncovered by the output
of the tests - https://github.com/redis/redis/issues/11897
- [x] (probably a separate PR) add all missing schemas
- [x] check why "SDOWN is triggered by misconfigured instance replying with errors" fails with --log-req-res
- [x] move the response transformers to their own file (run both regular, cluster, and sentinel tests - need to
fight with the tcl including mechanism a bit)
- [x] issue: module API - https://github.com/redis/redis/issues/11898
- [x] (probably a separate PR): improve schemas: add `required` to `object`s - https://github.com/redis/redis/issues/11899
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Hanna Fadida <hanna.fadida@redislabs.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Shaya Potter <shaya@redislabs.com>
2023-03-11 09:14:16 +01:00
|
|
|
#ifdef LOG_REQ_RES
|
|
|
|
/* Reply schema */
|
|
|
|
struct jsonObject *reply_schema;
|
|
|
|
#endif
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
|
Move doc metadata from COMMAND to COMMAND DOCS (#10056)
Syntax:
`COMMAND DOCS [<command name> ...]`
Background:
Apparently old version of hiredis (and thus also redis-cli) can't
support more than 7 levels of multi-bulk nesting.
The solution is to move all the doc related metadata from COMMAND to a
new COMMAND DOCS sub-command.
The new DOCS sub-command returns a map of commands (not an array like in COMMAND),
And the same goes for the `subcommands` field inside it (also contains a map)
Besides that, the remaining new fields of COMMAND (hints, key-specs, and
sub-commands), are placed in the outer array rather than a nested map.
this was done mainly for consistency with the old format.
Other changes:
---
* Allow COMMAND INFO with no arguments, which returns all commands, so that we can some day deprecated
the plain COMMAND (no args)
* Reduce the amount of deferred replies from both COMMAND and COMMAND
DOCS, especially in the inner loops, since these create many small
reply objects, which lead to many small write syscalls and many small
TCP packets.
To make this easier, when populating the command table, we count the
history, args, and hints so we later know their size in advance.
Additionally, the movablekeys flag was moved into the flags register.
* Update generate-commands-json.py to take the data from both command, it
now executes redis-cli directly, instead of taking input from stdin.
* Sub-commands in both COMMAND (and COMMAND INFO), and also COMMAND DOCS,
show their full name. i.e. CONFIG
* GET will be shown as `config|get` rather than just `get`.
This will be visible both when asking for `COMMAND INFO config` and COMMAND INFO config|get`, but is
especially important for the later.
i.e. imagine someone doing `COMMAND INFO slowlog|get config|get` not being able to distinguish between the two
items in the array response.
2022-01-11 17:16:16 +02:00
|
|
|
/* Runtime populated data */
|
2020-12-31 14:53:43 +00:00
|
|
|
long long microseconds, calls, rejected_calls, failed_calls;
|
2024-05-22 23:24:12 -07:00
|
|
|
int id; /* Command ID. This is a progressive ID starting from 0 that
|
|
|
|
is assigned at runtime, and is used in order to check
|
|
|
|
ACLs. A connection is able to execute a given command if
|
|
|
|
the user associated to the connection has this command
|
|
|
|
bit set in the bitmap of allowed commands. */
|
2022-01-23 16:05:06 +08:00
|
|
|
sds fullname; /* A SDS string representing the command fullname. */
|
2024-05-22 23:24:12 -07:00
|
|
|
struct hdr_histogram
|
|
|
|
*latency_histogram; /*points to the command latency command histogram (unit of time nanosecond) */
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
keySpec legacy_range_key_spec; /* The legacy (first,last,step) key spec is
|
2024-05-22 23:24:12 -07:00
|
|
|
* still maintained (if applicable) so that
|
|
|
|
* we can still support the reply format of
|
|
|
|
* COMMAND INFO and COMMAND GETKEYS */
|
|
|
|
dict *subcommands_dict; /* A dictionary that holds the subcommands, the key is the subcommand sds name
|
|
|
|
* (not the fullname), and the value is the serverCommand structure pointer. */
|
2024-04-03 23:54:33 +07:00
|
|
|
struct serverCommand *parent;
|
2024-04-05 16:59:55 -07:00
|
|
|
struct ValkeyModuleCommand *module_cmd; /* A pointer to the module command data (NULL if native command) */
|
2010-06-22 00:07:48 +02:00
|
|
|
};
|
|
|
|
|
2024-04-04 00:12:34 +07:00
|
|
|
struct serverError {
|
2020-12-31 14:53:43 +00:00
|
|
|
long long count;
|
|
|
|
};
|
|
|
|
|
2024-04-04 01:24:01 +07:00
|
|
|
struct serverFunctionSym {
|
2010-06-22 00:07:48 +02:00
|
|
|
char *name;
|
|
|
|
unsigned long pointer;
|
|
|
|
};
|
|
|
|
|
2024-04-04 01:24:01 +07:00
|
|
|
typedef struct _serverSortObject {
|
2010-06-22 00:07:48 +02:00
|
|
|
robj *obj;
|
|
|
|
union {
|
|
|
|
double score;
|
|
|
|
robj *cmpobj;
|
|
|
|
} u;
|
2024-04-04 01:24:01 +07:00
|
|
|
} serverSortObject;
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2024-04-04 01:24:01 +07:00
|
|
|
typedef struct _serverSortOperation {
|
2010-06-22 00:07:48 +02:00
|
|
|
int type;
|
|
|
|
robj *pattern;
|
2024-04-04 01:24:01 +07:00
|
|
|
} serverSortOperation;
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* Structure to hold list iteration abstraction. */
|
|
|
|
typedef struct {
|
|
|
|
robj *subject;
|
|
|
|
unsigned char encoding;
|
|
|
|
unsigned char direction; /* Iteration direction */
|
Add listpack encoding for list (#11303)
Improve memory efficiency of list keys
## Description of the feature
The new listpack encoding uses the old `list-max-listpack-size` config
to perform the conversion, which we can think it of as a node inside a
quicklist, but without 80 bytes overhead (internal fragmentation included)
of quicklist and quicklistNode structs.
For example, a list key with 5 items of 10 chars each, now takes 128 bytes
instead of 208 it used to take.
## Conversion rules
* Convert listpack to quicklist
When the listpack length or size reaches the `list-max-listpack-size` limit,
it will be converted to a quicklist.
* Convert quicklist to listpack
When a quicklist has only one node, and its length or size is reduced to half
of the `list-max-listpack-size` limit, it will be converted to a listpack.
This is done to avoid frequent conversions when we add or remove at the bounding size or length.
## Interface changes
1. add list entry param to listTypeSetIteratorDirection
When list encoding is listpack, `listTypeIterator->lpi` points to the next entry of current entry,
so when changing the direction, we need to use the current node (listTypeEntry->p) to
update `listTypeIterator->lpi` to the next node in the reverse direction.
## Benchmark
### Listpack VS Quicklist with one node
* LPUSH - roughly 0.3% improvement
* LRANGE - roughly 13% improvement
### Both are quicklist
* LRANGE - roughly 3% improvement
* LRANGE without pipeline - roughly 3% improvement
From the benchmark, as we can see from the results
1. When list is quicklist encoding, LRANGE improves performance by <5%.
2. When list is listpack encoding, LRANGE improves performance by ~13%,
the main enhancement is brought by `addListListpackRangeReply()`.
## Memory usage
1M lists(key:0~key:1000000) with 5 items of 10 chars ("hellohello") each.
shows memory usage down by 35.49%, from 214MB to 138MB.
## Note
1. Add conversion callback to support doing some work before conversion
Since the quicklist iterator decompresses the current node when it is released, we can
no longer decompress the quicklist after we convert the list.
2022-11-17 02:29:46 +08:00
|
|
|
|
2024-05-22 23:24:12 -07:00
|
|
|
unsigned char *lpi; /* listpack iterator */
|
Add listpack encoding for list (#11303)
Improve memory efficiency of list keys
## Description of the feature
The new listpack encoding uses the old `list-max-listpack-size` config
to perform the conversion, which we can think it of as a node inside a
quicklist, but without 80 bytes overhead (internal fragmentation included)
of quicklist and quicklistNode structs.
For example, a list key with 5 items of 10 chars each, now takes 128 bytes
instead of 208 it used to take.
## Conversion rules
* Convert listpack to quicklist
When the listpack length or size reaches the `list-max-listpack-size` limit,
it will be converted to a quicklist.
* Convert quicklist to listpack
When a quicklist has only one node, and its length or size is reduced to half
of the `list-max-listpack-size` limit, it will be converted to a listpack.
This is done to avoid frequent conversions when we add or remove at the bounding size or length.
## Interface changes
1. add list entry param to listTypeSetIteratorDirection
When list encoding is listpack, `listTypeIterator->lpi` points to the next entry of current entry,
so when changing the direction, we need to use the current node (listTypeEntry->p) to
update `listTypeIterator->lpi` to the next node in the reverse direction.
## Benchmark
### Listpack VS Quicklist with one node
* LPUSH - roughly 0.3% improvement
* LRANGE - roughly 13% improvement
### Both are quicklist
* LRANGE - roughly 3% improvement
* LRANGE without pipeline - roughly 3% improvement
From the benchmark, as we can see from the results
1. When list is quicklist encoding, LRANGE improves performance by <5%.
2. When list is listpack encoding, LRANGE improves performance by ~13%,
the main enhancement is brought by `addListListpackRangeReply()`.
## Memory usage
1M lists(key:0~key:1000000) with 5 items of 10 chars ("hellohello") each.
shows memory usage down by 35.49%, from 214MB to 138MB.
## Note
1. Add conversion callback to support doing some work before conversion
Since the quicklist iterator decompresses the current node when it is released, we can
no longer decompress the quicklist after we convert the list.
2022-11-17 02:29:46 +08:00
|
|
|
quicklistIter *iter; /* quicklist iterator */
|
2010-06-22 00:07:48 +02:00
|
|
|
} listTypeIterator;
|
|
|
|
|
|
|
|
/* Structure for an entry while iterating over a list. */
|
|
|
|
typedef struct {
|
|
|
|
listTypeIterator *li;
|
2024-05-22 23:24:12 -07:00
|
|
|
unsigned char *lpe; /* Entry in listpack */
|
2014-11-13 14:11:47 -05:00
|
|
|
quicklistEntry entry; /* Entry in quicklist */
|
2010-06-22 00:07:48 +02:00
|
|
|
} listTypeEntry;
|
|
|
|
|
2010-07-02 19:57:12 +02:00
|
|
|
/* Structure to hold set iteration abstraction. */
|
|
|
|
typedef struct {
|
|
|
|
robj *subject;
|
|
|
|
int encoding;
|
|
|
|
int ii; /* intset iterator */
|
|
|
|
dictIterator *di;
|
2022-11-09 18:50:07 +01:00
|
|
|
unsigned char *lpi; /* listpack iterator */
|
2010-08-21 11:25:13 +02:00
|
|
|
} setTypeIterator;
|
2010-07-02 19:57:12 +02:00
|
|
|
|
2013-01-17 01:00:20 +08:00
|
|
|
/* Structure to hold hash iteration abstraction. Note that iteration over
|
2010-06-22 00:07:48 +02:00
|
|
|
* hashes involves both fields and values. Because it is possible that
|
|
|
|
* not both are required, store pointers in the iterator to avoid
|
|
|
|
* unnecessary memory allocation for fields/values. */
|
|
|
|
typedef struct {
|
2012-01-02 22:14:10 -08:00
|
|
|
robj *subject;
|
2010-06-22 00:07:48 +02:00
|
|
|
int encoding;
|
2012-01-02 22:14:10 -08:00
|
|
|
|
|
|
|
unsigned char *fptr, *vptr;
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
dictIterator *di;
|
|
|
|
dictEntry *de;
|
|
|
|
} hashTypeIterator;
|
|
|
|
|
2024-05-22 23:24:12 -07:00
|
|
|
#include "stream.h" /* Stream data type header file. */
|
2018-03-19 14:16:13 +01:00
|
|
|
|
2015-07-26 15:28:00 +02:00
|
|
|
#define OBJ_HASH_KEY 1
|
|
|
|
#define OBJ_HASH_VALUE 2
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/*-----------------------------------------------------------------------------
|
|
|
|
* Extern declarations
|
|
|
|
*----------------------------------------------------------------------------*/
|
|
|
|
|
2024-04-04 01:34:18 +07:00
|
|
|
extern struct valkeyServer server;
|
2010-06-22 00:07:48 +02:00
|
|
|
extern struct sharedObjectsStruct shared;
|
2015-07-31 18:01:23 +02:00
|
|
|
extern dictType objectKeyPointerValueDictType;
|
2017-09-06 15:43:28 +02:00
|
|
|
extern dictType objectKeyHeapPointerValueDictType;
|
2010-06-22 00:07:48 +02:00
|
|
|
extern dictType setDictType;
|
Make INFO command variadic (#6891)
This is an enhancement for INFO command, previously INFO only support one argument
for different info section , if user want to get more categories information, either perform
INFO all / default or calling INFO for multiple times.
**Description of the feature**
The goal of adding this feature is to let the user retrieve multiple categories via the INFO
command, and still avoid emitting the same section twice.
A use case for this is like Redis Sentinel, which periodically calling INFO command to refresh
info from monitored Master/Slaves, only Server and Replication part categories are used for
parsing information. If the INFO command can return just enough categories that client side
needs, it can save a lot of time for client side parsing it as well as network bandwidth.
**Implementation**
To share code between redis, sentinel, and other users of INFO (DEBUG and modules),
we have a new `genInfoSectionDict` function that returns a dict and some boolean flags
(e.g. `all`) to the caller (built from user input).
Sentinel is later purging unwanted sections from that, and then it is forwarded to the info `genRedisInfoString`.
**Usage Examples**
INFO Server Replication
INFO CPU Memory
INFO default commandstats
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-08 06:14:42 -05:00
|
|
|
extern dictType BenchmarkDictType;
|
2010-06-22 00:07:48 +02:00
|
|
|
extern dictType zsetDictType;
|
2011-07-13 15:38:03 +02:00
|
|
|
extern dictType dbDictType;
|
2010-06-22 00:07:48 +02:00
|
|
|
extern double R_Zero, R_PosInf, R_NegInf, R_Nan;
|
2012-03-27 18:18:57 +02:00
|
|
|
extern dictType hashDictType;
|
Make INFO command variadic (#6891)
This is an enhancement for INFO command, previously INFO only support one argument
for different info section , if user want to get more categories information, either perform
INFO all / default or calling INFO for multiple times.
**Description of the feature**
The goal of adding this feature is to let the user retrieve multiple categories via the INFO
command, and still avoid emitting the same section twice.
A use case for this is like Redis Sentinel, which periodically calling INFO command to refresh
info from monitored Master/Slaves, only Server and Replication part categories are used for
parsing information. If the INFO command can return just enough categories that client side
needs, it can save a lot of time for client side parsing it as well as network bandwidth.
**Implementation**
To share code between redis, sentinel, and other users of INFO (DEBUG and modules),
we have a new `genInfoSectionDict` function that returns a dict and some boolean flags
(e.g. `all`) to the caller (built from user input).
Sentinel is later purging unwanted sections from that, and then it is forwarded to the info `genRedisInfoString`.
**Usage Examples**
INFO Server Replication
INFO CPU Memory
INFO default commandstats
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-08 06:14:42 -05:00
|
|
|
extern dictType stringSetDictType;
|
2022-02-28 23:02:47 -08:00
|
|
|
extern dictType externalStringType;
|
|
|
|
extern dictType sdsHashDictType;
|
2024-01-08 16:32:31 +08:00
|
|
|
extern dictType clientDictType;
|
|
|
|
extern dictType objToDictDictType;
|
Limit the main db and expires dictionaries to expand (#7954)
As we know, redis may reject user's requests or evict some keys if
used memory is over maxmemory. Dictionaries expanding may make
things worse, some big dictionaries, such as main db and expires dict,
may eat huge memory at once for allocating a new big hash table and be
far more than maxmemory after expanding.
There are related issues: #4213 #4583
More details, when expand dict in redis, we will allocate a new big
ht[1] that generally is double of ht[0], The size of ht[1] will be
very big if ht[0] already is big. For db dict, if we have more than
64 million keys, we need to cost 1GB for ht[1] when dict expands.
If the sum of used memory and new hash table of dict needed exceeds
maxmemory, we shouldn't allow the dict to expand. Because, if we
enable keys eviction, we still couldn't add much more keys after
eviction and rehashing, what's worse, redis will keep less keys when
redis only remains a little memory for storing new hash table instead
of users' data. Moreover users can't write data in redis if disable
keys eviction.
What this commit changed ?
Add a new member function expandAllowed for dict type, it provide a way
for caller to allow expand or not. We expose two parameters for this
function: more memory needed for expanding and dict current load factor,
users can implement a function to make a decision by them.
For main db dict and expires dict type, these dictionaries may be very
big and cost huge memory for expanding, so we implement a judgement
function: we can stop dict to expand provisionally if used memory will
be over maxmemory after dict expands, but to guarantee the performance
of redis, we still allow dict to expand if dict load factor exceeds the
safe load factor.
Add test cases to verify we don't allow main db to expand when left
memory is not enough, so that avoid keys eviction.
Other changes:
For new hash table size when expand. Before this commit, the size is
that double used of dict and later _dictNextPower. Actually we aim to
control a dict load factor between 0.5 and 1.0. Now we replace *2 with
+1, since the first check is that used >= size, the outcome of before
will usually be the same as _dictNextPower(used+1). The only case where
it'll differ is when dict_can_resize is false during fork, so that later
the _dictNextPower(used*2) will cause the dict to jump to *4 (i.e.
_dictNextPower(1025*2) will return 4096).
Fix rehash test cases due to changing algorithm of new hash table size
when expand.
2020-12-06 17:53:04 +08:00
|
|
|
extern dictType dbExpiresDictType;
|
2016-03-06 13:44:24 +01:00
|
|
|
extern dictType modulesDictType;
|
2021-01-29 16:47:28 +08:00
|
|
|
extern dictType sdsReplyDictType;
|
Replace slots_to_channels radix tree with slot specific dictionaries for shard channels. (#12804)
We have achieved replacing `slots_to_keys` radix tree with key->slot
linked list (#9356), and then replacing the list with slot specific
dictionaries for keys (#11695).
Shard channels behave just like keys in many ways, and we also need a
slots->channels mapping. Currently this is still done by using a radix
tree. So we should split `server.pubsubshard_channels` into 16384 dicts
and drop the radix tree, just like what we did to DBs.
Some benefits (basically the benefits of what we've done to DBs):
1. Optimize counting channels in a slot. This is currently used only in
removing channels in a slot. But this is potentially more useful:
sometimes we need to know how many channels there are in a specific slot
when doing slot migration. Counting is now implemented by traversing the
radix tree, and with this PR it will be as simple as calling `dictSize`,
from O(n) to O(1).
2. The radix tree in the cluster has been removed. The shard channel
names no longer require additional storage, which can save memory.
3. Potentially useful in slot migration, as shard channels are logically
split by slots, thus making it easier to migrate, remove or add as a
whole.
4. Avoid rehashing a big dict when there is a large number of channels.
Drawbacks:
1. Takes more memory than using radix tree when there are relatively few
shard channels.
What this PR does:
1. in cluster mode, split `server.pubsubshard_channels` into 16384
dicts, in standalone mode, still use only one dict.
2. drop the `slots_to_channels` radix tree.
3. to save memory (to solve the drawback above), all 16384 dicts are
created lazily, which means only when a channel is about to be inserted
to the dict will the dict be initialized, and when all channels are
deleted, the dict would delete itself.
5. use `server.shard_channel_count` to keep track of the number of all
shard channels.
---------
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2023-12-27 17:40:45 +08:00
|
|
|
extern dictType keylistDictType;
|
2018-04-20 21:01:21 +08:00
|
|
|
extern dict *modules;
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/*-----------------------------------------------------------------------------
|
|
|
|
* Functions prototypes
|
|
|
|
*----------------------------------------------------------------------------*/
|
|
|
|
|
Command info module API (#10108)
Adds RM_SetCommandInfo, allowing modules to provide the following command info:
* summary
* complexity
* since
* history
* hints
* arity
* key specs
* args
This information affects the output of `COMMAND`, `COMMAND INFO` and `COMMAND DOCS`,
Cluster, ACL and is used to filter commands with the wrong number of arguments before
the call reaches the module code.
The recently added API functions for key specs (never released) are removed.
A minimalist example would look like so:
```c
RedisModuleCommand *mycmd = RedisModule_GetCommand(ctx,"mymodule.mycommand");
RedisModuleCommandInfo mycmd_info = {
.version = REDISMODULE_COMMAND_INFO_VERSION,
.arity = -5,
.summary = "some description",
};
if (RedisModule_SetCommandInfo(mycmd, &mycmd_info) == REDISMODULE_ERR)
return REDISMODULE_ERR;
````
Notes:
* All the provided information (including strings) is copied, not keeping references to the API input data.
* The version field is actually a static struct that contains the sizes of the the structs used in arrays,
so we can extend these in the future and old version will still be able to take the part they can support.
2022-02-04 20:09:36 +01:00
|
|
|
/* Command metadata */
|
2024-04-03 23:54:33 +07:00
|
|
|
void populateCommandLegacyRangeSpec(struct serverCommand *c);
|
A better approach for COMMAND INFO for movablekeys commands (#8324)
Fix #7297
The problem:
Today, there is no way for a client library or app to know the key name indexes for commands such as
ZUNIONSTORE/EVAL and others with "numkeys", since COMMAND INFO returns no useful info for them.
For cluster-aware redis clients, this requires to 'patch' the client library code specifically for each of these commands or to
resolve each execution of these commands with COMMAND GETKEYS.
The solution:
Introducing key specs other than the legacy "range" (first,last,step)
The 8th element of the command info array, if exists, holds an array of key specs. The array may be empty, which indicates
the command doesn't take any key arguments or may contain one or more key-specs, each one may leads to the discovery
of 0 or more key arguments.
A client library that doesn't support this key-spec feature will keep using the first,last,step and movablekeys flag which will
obviously remain unchanged.
A client that supports this key-specs feature needs only to look at the key-specs array. If it finds an unrecognized spec, it
must resort to using COMMAND GETKEYS if it wishes to get all key name arguments, but if all it needs is one key in order
to know which cluster node to use, then maybe another spec (if the command has several) can supply that, and there's no
need to use GETKEYS.
Each spec is an array of arguments, first one is the spec name, the second is an array of flags, and the third is an array
containing details about the spec (specific meaning for each spec type)
The initial flags we support are "read" and "write" indicating if the keys that this key-spec finds are used for read or for write.
clients should ignore any unfamiliar flags.
In order to easily find the positions of keys in a given array of args we introduce keys specs. There are two logical steps of
key specs:
1. `start_search`: Given an array of args, indicate where we should start searching for keys
2. `find_keys`: Given the output of start_search and an array of args, indicate all possible indices of keys.
### start_search step specs
- `index`: specify an argument index explicitly
- `index`: 0 based index (1 means the first command argument)
- `keyword`: specify a string to match in `argv`. We should start searching for keys just after the keyword appears.
- `keyword`: the string to search for
- `start_search`: an index from which to start the keyword search (can be negative, which means to search from the end)
Examples:
- `SET` has start_search of type `index` with value `1`
- `XREAD` has start_search of type `keyword` with value `[“STREAMS”,1]`
- `MIGRATE` has start_search of type `keyword` with value `[“KEYS”,-2]`
### find_keys step specs
- `range`: specify `[count, step, limit]`.
- `lastkey`: index of the last key. relative to the index returned from begin_search. -1 indicating till the last argument, -2 one before the last
- `step`: how many args should we skip after finding a key, in order to find the next one
- `limit`: if count is -1, we use limit to stop the search by a factor. 0 and 1 mean no limit. 2 means ½ of the remaining args, 3 means ⅓, and so on.
- “keynum”: specify `[keynum_index, first_key_index, step]`.
- `keynum_index`: is relative to the return of the `start_search` spec.
- `first_key_index`: is relative to `keynum_index`.
- `step`: how many args should we skip after finding a key, in order to find the next one
Examples:
- `SET` has `range` of `[0,1,0]`
- `MSET` has `range` of `[-1,2,0]`
- `XREAD` has `range` of `[-1,1,2]`
- `ZUNION` has `start_search` of type `index` with value `1` and `find_keys` of type `keynum` with value `[0,1,1]`
- `AI.DAGRUN` has `start_search` of type `keyword` with value `[“LOAD“,1]` and `find_keys` of type `keynum` with value
`[0,1,1]` (see https://oss.redislabs.com/redisai/master/commands/#aidagrun)
Note: this solution is not perfect as the module writers can come up with anything, but at least we will be able to find the key
args of the vast majority of commands.
If one of the above specs can’t describe the key positions, the module writer can always fall back to the `getkeys-api` option.
Some keys cannot be found easily (`KEYS` in `MIGRATE`: Imagine the argument for `AUTH` is the string “KEYS” - we will
start searching in the wrong index).
The guarantee is that the specs may be incomplete (`incomplete` will be specified in the spec to denote that) but we never
report false information (assuming the command syntax is correct).
For `MIGRATE` we start searching from the end - `startfrom=-1` - and if one of the keys is actually called "keys" we will
report only a subset of all keys - hence the `incomplete` flag.
Some `incomplete` specs can be completely empty (i.e. UNKNOWN begin_search) which should tell the client that
COMMAND GETKEYS (or any other way to get the keys) must be used (Example: For `SORT` there is no way to describe
the STORE keyword spec, as the word "store" can appear anywhere in the command).
We will expose these key specs in the `COMMAND` command so that clients can learn, on startup, where the keys are for
all commands instead of holding hardcoded tables or use `COMMAND GETKEYS` in runtime.
Comments:
1. Redis doesn't internally use the new specs, they are only used for COMMAND output.
2. In order to support the current COMMAND INFO format (reply array indices 4, 5, 6) we created a synthetic range, called
legacy_range, that, if possible, is built according to the new specs.
3. Redis currently uses only getkeys_proc or the legacy_range to get the keys indices (in COMMAND GETKEYS for
example).
"incomplete" specs:
the command we have issues with are MIGRATE, STRALGO, and SORT
for MIGRATE, because the token KEYS, if exists, must be the last token, we can search in reverse. it one of the keys is
actually the string "keys" will return just a subset of the keys (hence, it's "incomplete")
for SORT and STRALGO we can use this heuristic (the keys can be anywhere in the command) and therefore we added a
key spec that is both "incomplete" and of "unknown type"
if a client encounters an "incomplete" spec it means that it must find a different way (either COMMAND GETKEYS or have
its own parser) to retrieve the keys.
please note that all commands, apart from the three mentioned above, have "complete" key specs
2021-09-15 10:10:29 +02:00
|
|
|
|
2016-03-06 13:44:24 +01:00
|
|
|
/* Modules */
|
|
|
|
void moduleInitModulesSystem(void);
|
2020-08-08 19:36:41 +08:00
|
|
|
void moduleInitModulesSystemLast(void);
|
2022-01-11 20:00:56 +03:00
|
|
|
void modulesCron(void);
|
Module Configurations (#10285)
This feature adds the ability to add four different types (Bool, Numeric,
String, Enum) of configurations to a module to be accessed via the redis
config file, and the CONFIG command.
**Configuration Names**:
We impose a restriction that a module configuration always starts with the
module name and contains a '.' followed by the config name. If a module passes
"config1" as the name to a register function, it will be registered as MODULENAME.config1.
**Configuration Persistence**:
Module Configurations exist only as long as a module is loaded. If a module is
unloaded, the configurations are removed.
There is now also a minimal core API for removal of standardConfig objects
from configs by name.
**Get and Set Callbacks**:
Storage of config values is owned by the module that registers them, and provides
callbacks for Redis to access and manipulate the values.
This is exposed through a GET and SET callback.
The get callback returns a typed value of the config to redis. The callback takes
the name of the configuration, and also a privdata pointer. Note that these only
take the CONFIGNAME portion of the config, not the entire MODULENAME.CONFIGNAME.
```
typedef RedisModuleString * (*RedisModuleConfigGetStringFunc)(const char *name, void *privdata);
typedef long long (*RedisModuleConfigGetNumericFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetBoolFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetEnumFunc)(const char *name, void *privdata);
```
Configs must also must specify a set callback, i.e. what to do on a CONFIG SET XYZ 123
or when loading configurations from cli/.conf file matching these typedefs. *name* is
again just the CONFIGNAME portion, *val* is the parsed value from the core,
*privdata* is the registration time privdata pointer, and *err* is for providing errors to a client.
```
typedef int (*RedisModuleConfigSetStringFunc)(const char *name, RedisModuleString *val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetNumericFunc)(const char *name, long long val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetBoolFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetEnumFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
```
Modules can also specify an optional apply callback that will be called after
value(s) have been set via CONFIG SET:
```
typedef int (*RedisModuleConfigApplyFunc)(RedisModuleCtx *ctx, void *privdata, RedisModuleString **err);
```
**Flags:**
We expose 7 new flags to the module, which are used as part of the config registration.
```
#define REDISMODULE_CONFIG_MODIFIABLE 0 /* This is the default for a module config. */
#define REDISMODULE_CONFIG_IMMUTABLE (1ULL<<0) /* Can this value only be set at startup? */
#define REDISMODULE_CONFIG_SENSITIVE (1ULL<<1) /* Does this value contain sensitive information */
#define REDISMODULE_CONFIG_HIDDEN (1ULL<<4) /* This config is hidden in `config get <pattern>` (used for tests/debugging) */
#define REDISMODULE_CONFIG_PROTECTED (1ULL<<5) /* Becomes immutable if enable-protected-configs is enabled. */
#define REDISMODULE_CONFIG_DENY_LOADING (1ULL<<6) /* This config is forbidden during loading. */
/* Numeric Specific Configs */
#define REDISMODULE_CONFIG_MEMORY (1ULL<<7) /* Indicates if this value can be set as a memory value */
```
**Module Registration APIs**:
```
int (*RedisModule_RegisterBoolConfig)(RedisModuleCtx *ctx, char *name, int default_val, unsigned int flags, RedisModuleConfigGetBoolFunc getfn, RedisModuleConfigSetBoolFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterNumericConfig)(RedisModuleCtx *ctx, const char *name, long long default_val, unsigned int flags, long long min, long long max, RedisModuleConfigGetNumericFunc getfn, RedisModuleConfigSetNumericFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterStringConfig)(RedisModuleCtx *ctx, const char *name, const char *default_val, unsigned int flags, RedisModuleConfigGetStringFunc getfn, RedisModuleConfigSetStringFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterEnumConfig)(RedisModuleCtx *ctx, const char *name, int default_val, unsigned int flags, const char **enum_values, const int *int_values, int num_enum_vals, RedisModuleConfigGetEnumFunc getfn, RedisModuleConfigSetEnumFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_LoadConfigs)(RedisModuleCtx *ctx);
```
The module name will be auto appended along with a "." to the front of the name of the config.
**What RM_Register[...]Config does**:
A RedisModule struct now keeps a list of ModuleConfig objects which look like:
```
typedef struct ModuleConfig {
sds name; /* Name of config without the module name appended to the front */
void *privdata; /* Optional data passed into the module config callbacks */
union get_fn { /* The get callback specificed by the module */
RedisModuleConfigGetStringFunc get_string;
RedisModuleConfigGetNumericFunc get_numeric;
RedisModuleConfigGetBoolFunc get_bool;
RedisModuleConfigGetEnumFunc get_enum;
} get_fn;
union set_fn { /* The set callback specified by the module */
RedisModuleConfigSetStringFunc set_string;
RedisModuleConfigSetNumericFunc set_numeric;
RedisModuleConfigSetBoolFunc set_bool;
RedisModuleConfigSetEnumFunc set_enum;
} set_fn;
RedisModuleConfigApplyFunc apply_fn;
RedisModule *module;
} ModuleConfig;
```
It also registers a standardConfig in the configs array, with a pointer to the
ModuleConfig object associated with it.
**What happens on a CONFIG GET/SET MODULENAME.MODULECONFIG:**
For CONFIG SET, we do the same parsing as is done in config.c and pass that
as the argument to the module set callback. For CONFIG GET, we call the
module get callback and return that value to config.c to return to a client.
**CONFIG REWRITE**:
Starting up a server with module configurations in a .conf file but no module load
directive will fail. The flip side is also true, specifying a module load and a bunch
of module configurations will load those configurations in using the module defined
set callbacks on a RM_LoadConfigs call. Configs being rewritten works the same
way as it does for standard configs, as the module has the ability to specify a
default value. If a module is unloaded with configurations specified in the .conf file
those configurations will be commented out from the .conf file on the next config rewrite.
**RM_LoadConfigs:**
`RedisModule_LoadConfigs(RedisModuleCtx *ctx);`
This last API is used to make configs available within the onLoad() after they have
been registered. The expected usage is that a module will register all of its configs,
then call LoadConfigs to trigger all of the set callbacks, and then can error out if any
of them were malformed. LoadConfigs will attempt to set all configs registered to
either a .conf file argument/loadex argument or their default value if an argument is
not specified. **LoadConfigs is a required function if configs are registered.
** Also note that LoadConfigs **does not** call the apply callbacks, but a module
can do that directly after the LoadConfigs call.
**New Command: MODULE LOADEX [CONFIG NAME VALUE] [ARGS ...]:**
This command provides the ability to provide startup context information to a module.
LOADEX stands for "load extended" similar to GETEX. Note that provided config
names need the full MODULENAME.MODULECONFIG name. Any additional
arguments a module might want are intended to be specified after ARGS.
Everything after ARGS is passed to onLoad as RedisModuleString **argv.
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: Madelyn Olson <matolson@amazon.com>
Co-authored-by: sundb <sundbcn@gmail.com>
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Yossi Gottlieb <yossigo@gmail.com>
2022-03-30 05:47:06 -07:00
|
|
|
int moduleLoad(const char *path, void **argv, int argc, int is_loadex);
|
Custom authentication for Modules (#11659)
This change adds new module callbacks that can override the default password based authentication associated with ACLs. With this, Modules can register auth callbacks through which they can implement their own Authentication logic. When `AUTH` and `HELLO AUTH ...` commands are used, Module based authentication is attempted and then normal password based authentication is attempted if needed.
The new Module APIs added in this PR are - `RM_RegisterCustomAuthCallback` and `RM_BlockClientOnAuth` and `RedisModule_ACLAddLogEntryByUserName `.
Module based authentication will be attempted for all Redis users (created through the ACL SETUSER cmd or through Module APIs) even if the Redis user does not exist at the time of the command. This gives a chance for the Module to create the RedisModule user and then authenticate via the RedisModule API - from the custom auth callback.
For the AUTH command, we will support both variations - `AUTH <username> <password>` and `AUTH <password>`. In case of the `AUTH <password>` variation, the custom auth callbacks are triggered with “default” as the username and password as what is provided.
### RedisModule_RegisterCustomAuthCallback
```
void RM_RegisterCustomAuthCallback(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback cb) {
```
This API registers a callback to execute to prior to normal password based authentication. Multiple callbacks can be registered across different modules. These callbacks are responsible for either handling the authentication, each authenticating the user or explicitly denying, or deferring it to other authentication mechanisms. Callbacks are triggered in the order they were registered. When a Module is unloaded, all the auth callbacks registered by it are unregistered. The callbacks are attempted, in the order of most recently registered callbacks, when the AUTH/HELLO (with AUTH field is provided) commands are called. The callbacks will be called with a module context along with a username and a password, and are expected to take one of the following actions:
(1) Authenticate - Use the RM_Authenticate* API successfully and return `REDISMODULE_AUTH_HANDLED`. This will immediately end the auth chain as successful and add the OK reply.
(2) Block a client on authentication - Use the `RM_BlockClientOnAuth` API and return `REDISMODULE_AUTH_HANDLED`. Here, the client will be blocked until the `RM_UnblockClient `API is used which will trigger the auth reply callback (provided earlier through the `RM_BlockClientOnAuth`). In this reply callback, the Module should authenticate, deny or skip handling authentication.
(3) Deny Authentication - Return `REDISMODULE_AUTH_HANDLED` without authenticating or blocking the client. Optionally, `err` can be set to a custom error message. This will immediately end the auth chain as unsuccessful and add the ERR reply.
(4) Skip handling Authentication - Return `REDISMODULE_AUTH_NOT_HANDLED` without blocking the client. This will allow the engine to attempt the next custom auth callback.
If none of the callbacks authenticate or deny auth, then password based auth is attempted and will authenticate or add failure logs and reply to the clients accordingly.
### RedisModule_BlockClientOnAuth
```
RedisModuleBlockedClient *RM_BlockClientOnAuth(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback reply_callback,
void (*free_privdata)(RedisModuleCtx*,void*))
```
This API can only be used from a Module from the custom auth callback. If a client is not in the middle of custom module based authentication, ERROR is returned. Otherwise, the client is blocked and the `RedisModule_BlockedClient` is returned similar to the `RedisModule_BlockClient` API.
### RedisModule_ACLAddLogEntryByUserName
```
int RM_ACLAddLogEntryByUserName(RedisModuleCtx *ctx, RedisModuleString *username, RedisModuleString *object, RedisModuleACLLogEntryReason reason)
```
Adds a new entry in the ACL log with the `username` RedisModuleString provided. This simplifies the Module usage because now, developers do not need to create a Module User just to add an error ACL Log entry. Aside from accepting username (RedisModuleString) instead of a RedisModuleUser, it is the same as the existing `RedisModule_ACLAddLogEntry` API.
### Breaking changes
- HELLO command - Clients can now only set the client name and RESP protocol from the `HELLO` command if they are authenticated. Also, we now finish command arg validation first and return early with a ERR reply if any arg is invalid. This is to avoid mutating the client name / RESP from a command that would have failed on invalid arguments.
### Notable behaviors
- Module unblocking - Now, we will not allow Modules to block the client from inside the context of a reply callback (triggered from the Module unblock flow `moduleHandleBlockedClients`).
---------
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
2023-03-15 15:18:42 -07:00
|
|
|
int moduleUnload(sds name, const char **errmsg);
|
2016-03-06 13:44:24 +01:00
|
|
|
void moduleLoadFromQueue(void);
|
2024-04-03 23:54:33 +07:00
|
|
|
int moduleGetCommandKeysViaAPI(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int moduleGetCommandChannelsViaAPI(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
2016-05-18 11:45:40 +02:00
|
|
|
moduleType *moduleTypeLookupModuleByID(uint64_t id);
|
2023-06-27 21:43:46 +08:00
|
|
|
moduleType *moduleTypeLookupModuleByName(const char *name);
|
|
|
|
moduleType *moduleTypeLookupModuleByNameIgnoreCase(const char *name);
|
2016-05-18 11:45:40 +02:00
|
|
|
void moduleTypeNameByID(char *name, uint64_t moduleid);
|
2021-03-02 09:39:37 +02:00
|
|
|
const char *moduleTypeModuleName(moduleType *mt);
|
2024-04-03 23:54:33 +07:00
|
|
|
const char *moduleNameFromCommand(struct serverCommand *cmd);
|
2024-04-05 16:59:55 -07:00
|
|
|
void moduleFreeContext(struct ValkeyModuleCtx *ctx);
|
Support for RM_Call on blocking commands (#11568)
Allow running blocking commands from within a module using `RM_Call`.
Today, when `RM_Call` is used, the fake client that is used to run command
is marked with `CLIENT_DENY_BLOCKING` flag. This flag tells the command
that it is not allowed to block the client and in case it needs to block, it must
fallback to some alternative (either return error or perform some default behavior).
For example, `BLPOP` fallback to simple `LPOP` if it is not allowed to block.
All the commands must respect the `CLIENT_DENY_BLOCKING` flag (including
module commands). When the command invocation finished, Redis asserts that
the client was not blocked.
This PR introduces the ability to call blocking command using `RM_Call` by
passing a callback that will be called when the client will get unblocked.
In order to do that, the user must explicitly say that he allow to perform blocking
command by passing a new format specifier argument, `K`, to the `RM_Call`
function. This new flag will tell Redis that it is allow to run blocking command
and block the client. In case the command got blocked, Redis will return a new
type of call reply (`REDISMODULE_REPLY_PROMISE`). This call reply indicates
that the command got blocked and the user can set the on_unblocked handler using
`RM_CallReplyPromiseSetUnblockHandler`.
When clients gets unblocked, it eventually reaches `processUnblockedClients` function.
This is where we check if the client is a fake module client and if it is, we call the unblock
callback instead of performing the usual unblock operations.
**Notice**: `RM_CallReplyPromiseSetUnblockHandler` must be called atomically
along side the command invocation (without releasing the Redis lock in between).
In addition, unlike other CallReply types, the promise call reply must be released
by the module when the Redis GIL is acquired.
The module can abort the execution on the blocking command (if it was not yet
executed) using `RM_CallReplyPromiseAbort`. the API will return `REDISMODULE_OK`
on success and `REDISMODULE_ERR` if the operation is already executed.
**Notice** that in case of misbehave module, Abort might finished successfully but the
operation will not really be aborted. This can only happened if the module do not respect
the disconnect callback of the blocked client.
For pure Redis commands this can not happened.
### Atomicity Guarantees
The API promise that the unblock handler will run atomically as an execution unit.
This means that all the operation performed on the unblock handler will be wrapped
with a multi exec transaction when replicated to the replica and AOF.
The API **do not** grantee any other atomicity properties such as when the unblock
handler will be called. This gives us the flexibility to strengthen the grantees (or not)
in the future if we will decide that we need a better guarantees.
That said, the implementation **does** provide a better guarantees when performing
pure Redis blocking command like `BLPOP`. In this case the unblock handler will run
atomically with the operation that got unblocked (for example, in case of `BLPOP`, the
unblock handler will run atomically with the `LPOP` operation that run when the command
got unblocked). This is an implementation detail that might be change in the future and the
module writer should not count on that.
### Calling blocking commands while running on script mode (`S`)
`RM_Call` script mode (`S`) was introduced on #0372. It is used for usecases where the
command that was invoked on `RM_Call` comes from a user input and we want to make
sure the user will not run dangerous commands like `shutdown`. Some command, such
as `BLPOP`, are marked with `NO_SCRIPT` flag, which means they will not be allowed on
script mode. Those commands are marked with `NO_SCRIPT` just because they are
blocking commands and not because they are dangerous. Now that we can run blocking
commands on RM_Call, there is no real reason not to allow such commands on script mode.
The underline problem is that the `NO_SCRIPT` flag is abused to also mark some of the
blocking commands (notice that those commands know not to block the client if it is not
allowed to do so, and have a fallback logic to such cases. So even if those commands
were not marked with `NO_SCRIPT` flag, it would not harm Redis, and today we can
already run those commands within multi exec).
In addition, not all blocking commands are marked with `NO_SCRIPT` flag, for example
`blmpop` are not marked and can run from within a script.
Those facts shows that there are some ambiguity about the meaning of the `NO_SCRIPT`
flag, and its not fully clear where it should be use.
The PR suggest that blocking commands should not be marked with `NO_SCRIPT` flag,
those commands should handle `CLIENT_DENY_BLOCKING` flag and only block when
it's safe (like they already does today). To achieve that, the PR removes the `NO_SCRIPT`
flag from the following commands:
* `blmove`
* `blpop`
* `brpop`
* `brpoplpush`
* `bzpopmax`
* `bzpopmin`
* `wait`
This might be considered a breaking change as now, on scripts, instead of getting
`command is not allowed from script` error, the user will get some fallback behavior
base on the command implementation. That said, the change matches the behavior
of scripts and multi exec with respect to those commands and allow running them on
`RM_Call` even when script mode is used.
### Additional RedisModule API and changes
* `RM_BlockClientSetPrivateData` - Set private data on the blocked client without the
need to unblock the client. This allows up to set the promise CallReply as the private
data of the blocked client and abort it if the client gets disconnected.
* `RM_BlockClientGetPrivateData` - Return the current private data set on a blocked client.
We need it so we will have access to this private data on the disconnect callback.
* On RM_Call, the returned reply will be added to the auto memory context only if auto
memory is enabled, this allows us to keep the call reply for longer time then the context
lifetime and does not force an unneeded borrow relationship between the CallReply and
the RedisModuleContext.
2023-03-16 14:04:31 +02:00
|
|
|
void moduleCallCommandUnblockedHandler(client *c);
|
2024-01-31 19:10:19 +08:00
|
|
|
int isModuleClientUnblocked(client *c);
|
2016-10-07 11:55:35 +02:00
|
|
|
void unblockClientFromModule(client *c);
|
|
|
|
void moduleHandleBlockedClients(void);
|
Fix race condition issues between the main thread and module threads (#12817)
Fix #12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix #12785
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/3fcf959e609e850a114d4016843e4c991066ebac
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
https://github.com/redis/redis/commit/9b01b64430fbc1487429144d2e4e72a4a7fd9db2
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
2024-01-19 21:12:49 +08:00
|
|
|
void moduleBlockedClientTimedOut(client *c, int from_module);
|
Add event loop support to the module API (#10001)
Modules can now register sockets/pipe to the Redis main thread event loop and do network operations asynchronously. Previously, modules had to maintain an event loop and another thread for asynchronous network operations.
Also, if a module is calling API functions after doing some network operations, it had to synchronize its event loop thread's access with Redis main thread by locking the GIL, causing contention on the lock. After this commit, no synchronization is needed as module can operate in Redis main thread context. So, this commit may improve the performance for some use cases.
Added three functions to the module API:
* RedisModule_EventLoopAdd(int fd, int mask, RedisModuleEventLoopFunc func, void *user_data)
* RedisModule_EventLoopDel(int fd, int mask)
* RedisModule_EventLoopAddOneShot(RedisModuleEventLoopOneShotFunc func, void *user_data) - This function can be called from other threads to trigger callback on Redis main thread. Callback will be triggered only once. If Redis main thread is sleeping, this call will wake up the Redis main thread.
Event loop callbacks are called by Redis main thread after locking the GIL. Inside callbacks, modules can operate as if they are holding the GIL.
Added REDISMODULE_EVENT_EVENTLOOP event with two subevents:
* REDISMODULE_SUBEVENT_EVENTLOOP_BEFORE_SLEEP
* REDISMODULE_SUBEVENT_EVENTLOOP_AFTER_SLEEP
These events are for modules that want to participate in the before and after sleep action. e.g It might be useful to implement batching : Read data from the network, write all to a file in one go on BEFORE_SLEEP event.
2022-01-18 14:10:07 +03:00
|
|
|
void modulePipeReadable(aeEventLoop *el, int fd, void *privdata, int mask);
|
2017-05-03 11:26:21 +02:00
|
|
|
size_t moduleCount(void);
|
|
|
|
void moduleAcquireGIL(void);
|
2020-09-09 16:01:16 +03:00
|
|
|
int moduleTryAcquireGIL(void);
|
2017-05-03 11:26:21 +02:00
|
|
|
void moduleReleaseGIL(void);
|
2017-11-27 16:29:55 +02:00
|
|
|
void moduleNotifyKeyspaceEvent(int type, const char *event, robj *key, int dbid);
|
2023-05-02 17:31:32 -07:00
|
|
|
void firePostExecutionUnitJobs(void);
|
2018-02-23 16:19:37 +02:00
|
|
|
void moduleCallCommandFilters(client *c);
|
2023-05-02 17:31:32 -07:00
|
|
|
void modulePostExecutionUnitOperations(void);
|
2019-07-17 08:51:02 +03:00
|
|
|
void ModuleForkDoneHandler(int exitcode, int bysignal);
|
2019-09-27 12:23:07 +02:00
|
|
|
int TerminateModuleForkChild(int child_pid, int wait);
|
2019-07-21 17:41:03 +03:00
|
|
|
ssize_t rdbSaveModulesAux(rio *rdb, int when);
|
2023-05-02 17:31:32 -07:00
|
|
|
int moduleAllDatatypesHandleErrors(void);
|
|
|
|
int moduleAllModulesHandleReplAsyncLoad(void);
|
Make INFO command variadic (#6891)
This is an enhancement for INFO command, previously INFO only support one argument
for different info section , if user want to get more categories information, either perform
INFO all / default or calling INFO for multiple times.
**Description of the feature**
The goal of adding this feature is to let the user retrieve multiple categories via the INFO
command, and still avoid emitting the same section twice.
A use case for this is like Redis Sentinel, which periodically calling INFO command to refresh
info from monitored Master/Slaves, only Server and Replication part categories are used for
parsing information. If the INFO command can return just enough categories that client side
needs, it can save a lot of time for client side parsing it as well as network bandwidth.
**Implementation**
To share code between redis, sentinel, and other users of INFO (DEBUG and modules),
we have a new `genInfoSectionDict` function that returns a dict and some boolean flags
(e.g. `all`) to the caller (built from user input).
Sentinel is later purging unwanted sections from that, and then it is forwarded to the info `genRedisInfoString`.
**Usage Examples**
INFO Server Replication
INFO CPU Memory
INFO default commandstats
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-08 06:14:42 -05:00
|
|
|
sds modulesCollectInfo(sds info, dict *sections_dict, int for_crash_report, int sections);
|
2019-10-22 10:24:50 +02:00
|
|
|
void moduleFireServerEvent(uint64_t eid, int subid, void *data);
|
2019-10-29 17:59:09 +02:00
|
|
|
void processModuleLoadingProgressEvent(int is_aof);
|
2019-10-31 11:35:05 +01:00
|
|
|
int moduleTryServeClientBlockedOnKey(client *c, robj *key);
|
2019-10-30 10:57:44 +01:00
|
|
|
void moduleUnblockClient(client *c);
|
2021-07-01 17:11:27 +03:00
|
|
|
int moduleBlockedClientMayTimeout(client *c);
|
2019-10-31 11:35:05 +01:00
|
|
|
int moduleClientIsBlockedOnKeys(client *c);
|
2019-02-26 01:23:11 +00:00
|
|
|
void moduleNotifyUserChanged(client *c);
|
2022-11-30 17:56:36 +08:00
|
|
|
void moduleNotifyKeyUnlink(robj *key, robj *val, int dbid, int flags);
|
2021-06-16 14:45:49 +08:00
|
|
|
size_t moduleGetFreeEffort(robj *key, robj *val, int dbid);
|
2021-10-17 17:31:06 +03:00
|
|
|
size_t moduleGetMemUsage(robj *key, robj *val, size_t sample_size, int dbid);
|
2021-06-16 14:45:49 +08:00
|
|
|
robj *moduleTypeDupOrReply(client *c, robj *fromkey, robj *tokey, int todb, robj *value);
|
2022-11-21 15:34:38 +01:00
|
|
|
int moduleDefragValue(robj *key, robj *obj, int dbid);
|
|
|
|
int moduleLateDefrag(robj *key, robj *value, unsigned long *cursor, long long endtime, int dbid);
|
|
|
|
void moduleDefragGlobals(void);
|
Treat subcommands as commands (#9504)
## Intro
The purpose is to allow having different flags/ACL categories for
subcommands (Example: CONFIG GET is ok-loading but CONFIG SET isn't)
We create a small command table for every command that has subcommands
and each subcommand has its own flags, etc. (same as a "regular" command)
This commit also unites the Redis and the Sentinel command tables
## Affected commands
CONFIG
Used to have "admin ok-loading ok-stale no-script"
Changes:
1. Dropped "ok-loading" in all except GET (this doesn't change behavior since
there were checks in the code doing that)
XINFO
Used to have "read-only random"
Changes:
1. Dropped "random" in all except CONSUMERS
XGROUP
Used to have "write use-memory"
Changes:
1. Dropped "use-memory" in all except CREATE and CREATECONSUMER
COMMAND
No changes.
MEMORY
Used to have "random read-only"
Changes:
1. Dropped "random" in PURGE and USAGE
ACL
Used to have "admin no-script ok-loading ok-stale"
Changes:
1. Dropped "admin" in WHOAMI, GENPASS, and CAT
LATENCY
No changes.
MODULE
No changes.
SLOWLOG
Used to have "admin random ok-loading ok-stale"
Changes:
1. Dropped "random" in RESET
OBJECT
Used to have "read-only random"
Changes:
1. Dropped "random" in ENCODING and REFCOUNT
SCRIPT
Used to have "may-replicate no-script"
Changes:
1. Dropped "may-replicate" in all except FLUSH and LOAD
CLIENT
Used to have "admin no-script random ok-loading ok-stale"
Changes:
1. Dropped "random" in all except INFO and LIST
2. Dropped "admin" in ID, TRACKING, CACHING, GETREDIR, INFO, SETNAME, GETNAME, and REPLY
STRALGO
No changes.
PUBSUB
No changes.
CLUSTER
Changes:
1. Dropped "admin in countkeysinslots, getkeysinslot, info, nodes, keyslot, myid, and slots
SENTINEL
No changes.
(note that DEBUG also fits, but we decided not to convert it since it's for
debugging and anyway undocumented)
## New sub-command
This commit adds another element to the per-command output of COMMAND,
describing the list of subcommands, if any (in the same structure as "regular" commands)
Also, it adds a new subcommand:
```
COMMAND LIST [FILTERBY (MODULE <module-name>|ACLCAT <cat>|PATTERN <pattern>)]
```
which returns a set of all commands (unless filters), but excluding subcommands.
## Module API
A new module API, RM_CreateSubcommand, was added, in order to allow
module writer to define subcommands
## ACL changes:
1. Now, that each subcommand is actually a command, each has its own ACL id.
2. The old mechanism of allowed_subcommands is redundant
(blocking/allowing a subcommand is the same as blocking/allowing a regular command),
but we had to keep it, to support the widespread usage of allowed_subcommands
to block commands with certain args, that aren't subcommands (e.g. "-select +select|0").
3. I have renamed allowed_subcommands to allowed_firstargs to emphasize the difference.
4. Because subcommands are commands in ACL too, you can now use "-" to block subcommands
(e.g. "+client -client|kill"), which wasn't possible in the past.
5. It is also possible to use the allowed_firstargs mechanism with subcommand.
For example: `+config -config|set +config|set|loglevel` will block all CONFIG SET except
for setting the log level.
6. All of the ACL changes above required some amount of refactoring.
## Misc
1. There are two approaches: Either each subcommand has its own function or all
subcommands use the same function, determining what to do according to argv[0].
For now, I took the former approaches only with CONFIG and COMMAND,
while other commands use the latter approach (for smaller blamelog diff).
2. Deleted memoryGetKeys: It is no longer needed because MEMORY USAGE now uses the "range" key spec.
4. Bugfix: GETNAME was missing from CLIENT's help message.
5. Sentinel and Redis now use the same table, with the same function pointer.
Some commands have a different implementation in Sentinel, so we redirect
them (these are ROLE, PUBLISH, and INFO).
6. Command stats now show the stats per subcommand (e.g. instead of stats just
for "config" you will have stats for "config|set", "config|get", etc.)
7. It is now possible to use COMMAND directly on subcommands:
COMMAND INFO CONFIG|GET (The pipeline syntax was inspired from ACL, and
can be used in functions lookupCommandBySds and lookupCommandByCString)
8. STRALGO is now a container command (has "help")
## Breaking changes:
1. Command stats now show the stats per subcommand (see (5) above)
2021-10-20 10:52:57 +02:00
|
|
|
void *moduleGetHandleByName(char *modulename);
|
2024-04-03 23:54:33 +07:00
|
|
|
int moduleIsModuleCommand(void *module_handle, struct serverCommand *cmd);
|
2016-03-06 13:44:24 +01:00
|
|
|
|
2011-02-25 19:11:25 +01:00
|
|
|
/* Utils */
|
|
|
|
long long ustime(void);
|
2022-10-09 13:18:34 +08:00
|
|
|
mstime_t mstime(void);
|
|
|
|
mstime_t commandTimeSnapshot(void);
|
2018-04-05 13:24:22 +02:00
|
|
|
void getRandomHexChars(char *p, size_t len);
|
|
|
|
void getRandomBytes(unsigned char *p, size_t len);
|
2012-04-09 12:20:47 +02:00
|
|
|
uint64_t crc64(uint64_t crc, const unsigned char *s, uint64_t l);
|
2012-04-07 12:11:23 +02:00
|
|
|
void exitFromChild(int retcode);
|
2024-04-04 01:26:33 +07:00
|
|
|
long long serverPopcount(void *s, long count);
|
|
|
|
int serverSetProcTitle(char *title);
|
2021-01-28 18:17:39 +02:00
|
|
|
int validateProcTitleTemplate(const char *template);
|
2024-04-04 01:26:33 +07:00
|
|
|
int serverCommunicateSystemd(const char *sd_notify_msg);
|
|
|
|
void serverSetCpuAffinity(const char *cpulist);
|
2024-06-07 20:32:27 -07:00
|
|
|
void dictVanillaFree(dict *d, void *val);
|
2011-02-25 19:11:25 +01:00
|
|
|
|
2024-07-14 20:04:47 -07:00
|
|
|
/* ERROR STATS constants */
|
|
|
|
|
|
|
|
/* Once the errors RAX reaches this limit, instead of tracking custom
|
|
|
|
* errors (e.g. LUA), we track the error under the prefix below. */
|
|
|
|
#define ERRORSTATS_LIMIT 128
|
|
|
|
#define ERRORSTATS_OVERFLOW_ERR "ERRORSTATS_OVERFLOW"
|
|
|
|
|
Sort out the mess around Lua error messages and error stats (#10329)
This PR fix 2 issues on Lua scripting:
* Server error reply statistics (some errors were counted twice).
* Error code and error strings returning from scripts (error code was missing / misplaced).
## Statistics
a Lua script user is considered part of the user application, a sophisticated transaction,
so we want to count an error even if handled silently by the script, but when it is
propagated outwards from the script we don't wanna count it twice. on the other hand,
if the script decides to throw an error on its own (using `redis.error_reply`), we wanna
count that too.
Besides, we do count the `calls` in command statistics for the commands the script calls,
we we should certainly also count `failed_calls`.
So when a simple `eval "return redis.call('set','x','y')" 0` fails, it should count the failed call
to both SET and EVAL, but the `errorstats` and `total_error_replies` should be counted only once.
The PR changes the error object that is raised on errors. Instead of raising a simple Lua
string, Redis will raise a Lua table in the following format:
```
{
err='<error message (including error code)>',
source='<User source file name>',
line='<line where the error happned>',
ignore_error_stats_update=true/false,
}
```
The `luaPushError` function was modified to construct the new error table as describe above.
The `luaRaiseError` was renamed to `luaError` and is now simply called `lua_error` to raise
the table on the top of the Lua stack as the error object.
The reason is that since its functionality is changed, in case some Redis branch / fork uses it,
it's better to have a compilation error than a bug.
The `source` and `line` fields are enriched by the error handler (if possible) and the
`ignore_error_stats_update` is optional and if its not present then the default value is `false`.
If `ignore_error_stats_update` is true, the error will not be counted on the error stats.
When parsing Redis call reply, each error is translated to a Lua table on the format describe
above and the `ignore_error_stats_update` field is set to `true` so we will not count errors
twice (we counted this error when we invoke the command).
The changes in this PR might have been considered as a breaking change for users that used
Lua `pcall` function. Before, the error was a string and now its a table. To keep backward
comparability the PR override the `pcall` implementation and extract the error message from
the error table and return it.
Example of the error stats update:
```
127.0.0.1:6379> lpush l 1
(integer) 2
127.0.0.1:6379> eval "return redis.call('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value. script: e471b73f1ef44774987ab00bdf51f21fd9f7974a, on @user_script:1.
127.0.0.1:6379> info Errorstats
# Errorstats
errorstat_WRONGTYPE:count=1
127.0.0.1:6379> info commandstats
# Commandstats
cmdstat_eval:calls=1,usec=341,usec_per_call=341.00,rejected_calls=0,failed_calls=1
cmdstat_info:calls=1,usec=35,usec_per_call=35.00,rejected_calls=0,failed_calls=0
cmdstat_lpush:calls=1,usec=14,usec_per_call=14.00,rejected_calls=0,failed_calls=0
cmdstat_get:calls=1,usec=10,usec_per_call=10.00,rejected_calls=0,failed_calls=1
```
## error message
We can now construct the error message (sent as a reply to the user) from the error table,
so this solves issues where the error message was malformed and the error code appeared
in the middle of the error message:
```diff
127.0.0.1:6379> eval "return redis.call('set','x','y')" 0
-(error) ERR Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479): @user_script:1: OOM command not allowed when used memory > 'maxmemory'.
+(error) OOM command not allowed when used memory > 'maxmemory' @user_script:1. Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479)
```
```diff
127.0.0.1:6379> eval "redis.call('get', 'l')" 0
-(error) ERR Error running script (call to f_8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1): @user_script:1: WRONGTYPE Operation against a key holding the wrong kind of value
+(error) WRONGTYPE Operation against a key holding the wrong kind of value script: 8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1, on @user_script:1.
```
Notica that `redis.pcall` was not change:
```
127.0.0.1:6379> eval "return redis.pcall('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value
```
## other notes
Notice that Some commands (like GEOADD) changes the cmd variable on the client stats so we
can not count on it to update the command stats. In order to be able to update those stats correctly
we needed to promote `realcmd` variable to be located on the client struct.
Tests was added and modified to verify the changes.
Related PR's: #10279, #10218, #10278, #10309
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-27 13:40:57 +02:00
|
|
|
/* afterErrorReply flags */
|
2024-07-14 20:04:47 -07:00
|
|
|
|
|
|
|
/* Indicating that we should not update error stats after sending error reply. */
|
|
|
|
#define ERR_REPLY_FLAG_NO_STATS_UPDATE (1ULL << 0)
|
|
|
|
/* Indicates the error message is custom (e.g. from LUA). */
|
|
|
|
#define ERR_REPLY_FLAG_CUSTOM (1ULL << 1)
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* networking.c -- Networking and Client related operations */
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
|
|
|
|
/* Read flags for various read errors and states */
|
|
|
|
#define READ_FLAGS_QB_LIMIT_REACHED (1 << 0)
|
|
|
|
#define READ_FLAGS_ERROR_BIG_INLINE_REQUEST (1 << 1)
|
|
|
|
#define READ_FLAGS_ERROR_BIG_MULTIBULK (1 << 2)
|
|
|
|
#define READ_FLAGS_ERROR_INVALID_MULTIBULK_LEN (1 << 3)
|
|
|
|
#define READ_FLAGS_ERROR_UNAUTHENTICATED_MULTIBULK_LEN (1 << 4)
|
|
|
|
#define READ_FLAGS_ERROR_UNAUTHENTICATED_BULK_LEN (1 << 5)
|
|
|
|
#define READ_FLAGS_ERROR_BIG_BULK_COUNT (1 << 6)
|
|
|
|
#define READ_FLAGS_ERROR_MBULK_UNEXPECTED_CHARACTER (1 << 7)
|
|
|
|
#define READ_FLAGS_ERROR_MBULK_INVALID_BULK_LEN (1 << 8)
|
|
|
|
#define READ_FLAGS_ERROR_UNEXPECTED_INLINE_FROM_PRIMARY (1 << 9)
|
|
|
|
#define READ_FLAGS_ERROR_UNBALANCED_QUOTES (1 << 10)
|
|
|
|
#define READ_FLAGS_INLINE_ZERO_QUERY_LEN (1 << 11)
|
|
|
|
#define READ_FLAGS_PARSING_NEGATIVE_MBULK_LEN (1 << 12)
|
|
|
|
#define READ_FLAGS_PARSING_COMPLETED (1 << 13)
|
|
|
|
#define READ_FLAGS_PRIMARY (1 << 14)
|
|
|
|
#define READ_FLAGS_DONT_PARSE (1 << 15)
|
|
|
|
#define READ_FLAGS_AUTH_REQUIRED (1 << 16)
|
|
|
|
|
|
|
|
/* Write flags for various write errors and states */
|
|
|
|
#define WRITE_FLAGS_WRITE_ERROR (1 << 0)
|
|
|
|
|
|
|
|
|
2019-09-12 10:56:54 +03:00
|
|
|
client *createClient(connection *conn);
|
2015-07-26 15:20:46 +02:00
|
|
|
void freeClient(client *c);
|
|
|
|
void freeClientAsync(client *c);
|
2022-01-10 14:21:16 +08:00
|
|
|
void logInvalidUseAndFreeClientAsync(client *c, const char *fmt, ...);
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
void beforeNextClient(client *c);
|
2022-01-11 20:00:56 +03:00
|
|
|
void clearClientConnectionState(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void resetClient(client *c);
|
2020-11-10 13:50:03 -08:00
|
|
|
void freeClientOriginalArgv(client *c);
|
acl check api for functions and eval (#10220)
Changes:
1. Adds the `redis.acl_check_cmd()` api to lua scripts. It can be used to check if the
current user has permissions to execute a given command. The new function receives
the command to check as an argument exactly like `redis.call()` receives the command
to execute as an argument.
2. In the PR I unified the code used to convert lua arguments to redis argv arguments from
both the new `redis.acl_check_cmd()` API and the `redis.[p]call()` API. This cleans up
potential duplicate code.
3. While doing the refactoring in 2 I noticed there's an optimization to reduce allocation calls
when parsing lua arguments into an `argv` array in the `redis.[p]call()` implementation.
These optimizations were introduced years ago in 48c49c485155ba9e4a7851fd1644c171674c6f0f
and 4f686555ce962e6632235d824512ea8fdeda003c. It is unclear why this was added.
The original commit message claims a 4% performance increase which I couldn't recreate
and might not be worth it even if it did recreate. This PR removes that optimization.
Following are details of the benchmark I did that couldn't reveal any performance
improvements due to this optimization:
```
benchmark 1: src/redis-benchmark -P 500 -n 10000000 eval 'return redis.call("ping")' 0
benchmark 2: src/redis-benchmark -P 500 -r 1000 -n 1000000 eval 'return redis.call("mset","k1__rand_int__","v1__rand_int__","k2__rand_int__","v2__rand_int__","k3__rand_int__","v3__rand_int__","k4__rand_int__","v4__rand_int__")' 0
benchmark 3: src/redis-benchmark -P 500 -r 1000 -n 100000 eval "for i=1,100,1 do redis.call('set','kk'..i,'vv'..__rand_int__) end return redis.call('get','kk5')" 0
benchmark 4: src/redis-benchmark -P 500 -r 1000 -n 1000000 eval 'return redis.call("mset","k1__rand_int__","v1__rand_int__","k2__rand_int__","v2__rand_int__","k3__rand_int__","v3__rand_int__","k4__rand_int__","v4__rand_int__xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx")'
```
I ran the benchmark on this branch with and without commit 68b71680a4d3bb8f0509e06578a9f15d05b92a47
Results in requests per second:
cmd | without optimization | without optimization 2nd run | with original optimization | with original optimization 2nd run
-- | -- | -- | -- | --
1 | 461233.34 | 477395.31 | 471098.16 | 469946.91
2 | 34774.14 | 35469.8 | 35149.38 | 34464.93
3 | 6390.59 | 6281.41 | 6146.28 | 6464.12
4 | 28005.71 | | 27965.77 |
As you can see, different use cases showed identical or negligible performance differences.
So finally I decided to chuck the original optimization and simplify the code.
2022-02-07 08:04:01 +02:00
|
|
|
void freeClientArgv(client *c);
|
2019-09-12 10:56:54 +03:00
|
|
|
void sendReplyToClient(connection *conn);
|
2018-11-08 12:28:56 +01:00
|
|
|
void *addReplyDeferredLen(client *c);
|
|
|
|
void setDeferredArrayLen(client *c, void *node, long length);
|
|
|
|
void setDeferredMapLen(client *c, void *node, long length);
|
|
|
|
void setDeferredSetLen(client *c, void *node, long length);
|
|
|
|
void setDeferredAttributeLen(client *c, void *node, long length);
|
|
|
|
void setDeferredPushLen(client *c, void *node, long length);
|
2021-09-26 17:45:02 +03:00
|
|
|
int processInputBuffer(client *c);
|
2024-07-01 02:33:10 +08:00
|
|
|
void acceptCommonHandler(connection *conn, struct ClientFlags flags, char *ip);
|
2019-09-12 10:56:54 +03:00
|
|
|
void readQueryFromClient(connection *conn);
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
int prepareClientToWrite(client *c);
|
2018-11-27 11:58:55 +01:00
|
|
|
void addReplyNull(client *c);
|
2018-11-30 16:31:02 +01:00
|
|
|
void addReplyNullArray(client *c);
|
2018-12-04 18:00:35 +01:00
|
|
|
void addReplyBool(client *c, int b);
|
2018-12-10 16:55:20 +01:00
|
|
|
void addReplyVerbatim(client *c, const char *s, size_t len, const char *ext);
|
2018-12-17 16:59:19 +01:00
|
|
|
void addReplyProto(client *c, const char *s, size_t len);
|
2019-03-24 13:10:55 +02:00
|
|
|
void AddReplyFromClient(client *c, client *src);
|
2015-07-26 15:20:46 +02:00
|
|
|
void addReplyBulk(client *c, robj *obj);
|
|
|
|
void addReplyBulkCString(client *c, const char *s);
|
|
|
|
void addReplyBulkCBuffer(client *c, const void *p, size_t len);
|
|
|
|
void addReplyBulkLongLong(client *c, long long ll);
|
|
|
|
void addReply(client *c, robj *obj);
|
2022-11-30 20:08:12 +00:00
|
|
|
void addReplyStatusLength(client *c, const char *s, size_t len);
|
2015-07-26 15:20:46 +02:00
|
|
|
void addReplySds(client *c, sds s);
|
|
|
|
void addReplyBulkSds(client *c, sds s);
|
2021-01-06 10:34:27 +02:00
|
|
|
void setDeferredReplyBulkSds(client *c, void *node, sds s);
|
2020-08-27 12:54:01 +03:00
|
|
|
void addReplyErrorObject(client *c, robj *err);
|
2021-08-10 10:19:21 +03:00
|
|
|
void addReplyOrErrorObject(client *c, robj *reply);
|
Sort out the mess around Lua error messages and error stats (#10329)
This PR fix 2 issues on Lua scripting:
* Server error reply statistics (some errors were counted twice).
* Error code and error strings returning from scripts (error code was missing / misplaced).
## Statistics
a Lua script user is considered part of the user application, a sophisticated transaction,
so we want to count an error even if handled silently by the script, but when it is
propagated outwards from the script we don't wanna count it twice. on the other hand,
if the script decides to throw an error on its own (using `redis.error_reply`), we wanna
count that too.
Besides, we do count the `calls` in command statistics for the commands the script calls,
we we should certainly also count `failed_calls`.
So when a simple `eval "return redis.call('set','x','y')" 0` fails, it should count the failed call
to both SET and EVAL, but the `errorstats` and `total_error_replies` should be counted only once.
The PR changes the error object that is raised on errors. Instead of raising a simple Lua
string, Redis will raise a Lua table in the following format:
```
{
err='<error message (including error code)>',
source='<User source file name>',
line='<line where the error happned>',
ignore_error_stats_update=true/false,
}
```
The `luaPushError` function was modified to construct the new error table as describe above.
The `luaRaiseError` was renamed to `luaError` and is now simply called `lua_error` to raise
the table on the top of the Lua stack as the error object.
The reason is that since its functionality is changed, in case some Redis branch / fork uses it,
it's better to have a compilation error than a bug.
The `source` and `line` fields are enriched by the error handler (if possible) and the
`ignore_error_stats_update` is optional and if its not present then the default value is `false`.
If `ignore_error_stats_update` is true, the error will not be counted on the error stats.
When parsing Redis call reply, each error is translated to a Lua table on the format describe
above and the `ignore_error_stats_update` field is set to `true` so we will not count errors
twice (we counted this error when we invoke the command).
The changes in this PR might have been considered as a breaking change for users that used
Lua `pcall` function. Before, the error was a string and now its a table. To keep backward
comparability the PR override the `pcall` implementation and extract the error message from
the error table and return it.
Example of the error stats update:
```
127.0.0.1:6379> lpush l 1
(integer) 2
127.0.0.1:6379> eval "return redis.call('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value. script: e471b73f1ef44774987ab00bdf51f21fd9f7974a, on @user_script:1.
127.0.0.1:6379> info Errorstats
# Errorstats
errorstat_WRONGTYPE:count=1
127.0.0.1:6379> info commandstats
# Commandstats
cmdstat_eval:calls=1,usec=341,usec_per_call=341.00,rejected_calls=0,failed_calls=1
cmdstat_info:calls=1,usec=35,usec_per_call=35.00,rejected_calls=0,failed_calls=0
cmdstat_lpush:calls=1,usec=14,usec_per_call=14.00,rejected_calls=0,failed_calls=0
cmdstat_get:calls=1,usec=10,usec_per_call=10.00,rejected_calls=0,failed_calls=1
```
## error message
We can now construct the error message (sent as a reply to the user) from the error table,
so this solves issues where the error message was malformed and the error code appeared
in the middle of the error message:
```diff
127.0.0.1:6379> eval "return redis.call('set','x','y')" 0
-(error) ERR Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479): @user_script:1: OOM command not allowed when used memory > 'maxmemory'.
+(error) OOM command not allowed when used memory > 'maxmemory' @user_script:1. Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479)
```
```diff
127.0.0.1:6379> eval "redis.call('get', 'l')" 0
-(error) ERR Error running script (call to f_8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1): @user_script:1: WRONGTYPE Operation against a key holding the wrong kind of value
+(error) WRONGTYPE Operation against a key holding the wrong kind of value script: 8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1, on @user_script:1.
```
Notica that `redis.pcall` was not change:
```
127.0.0.1:6379> eval "return redis.pcall('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value
```
## other notes
Notice that Some commands (like GEOADD) changes the cmd variable on the client stats so we
can not count on it to update the command stats. In order to be able to update those stats correctly
we needed to promote `realcmd` variable to be located on the client struct.
Tests was added and modified to verify the changes.
Related PR's: #10279, #10218, #10278, #10309
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-27 13:40:57 +02:00
|
|
|
void afterErrorReply(client *c, const char *s, size_t len, int flags);
|
2023-04-12 15:11:29 +08:00
|
|
|
void addReplyErrorFormatInternal(client *c, int flags, const char *fmt, va_list ap);
|
Sort out the mess around Lua error messages and error stats (#10329)
This PR fix 2 issues on Lua scripting:
* Server error reply statistics (some errors were counted twice).
* Error code and error strings returning from scripts (error code was missing / misplaced).
## Statistics
a Lua script user is considered part of the user application, a sophisticated transaction,
so we want to count an error even if handled silently by the script, but when it is
propagated outwards from the script we don't wanna count it twice. on the other hand,
if the script decides to throw an error on its own (using `redis.error_reply`), we wanna
count that too.
Besides, we do count the `calls` in command statistics for the commands the script calls,
we we should certainly also count `failed_calls`.
So when a simple `eval "return redis.call('set','x','y')" 0` fails, it should count the failed call
to both SET and EVAL, but the `errorstats` and `total_error_replies` should be counted only once.
The PR changes the error object that is raised on errors. Instead of raising a simple Lua
string, Redis will raise a Lua table in the following format:
```
{
err='<error message (including error code)>',
source='<User source file name>',
line='<line where the error happned>',
ignore_error_stats_update=true/false,
}
```
The `luaPushError` function was modified to construct the new error table as describe above.
The `luaRaiseError` was renamed to `luaError` and is now simply called `lua_error` to raise
the table on the top of the Lua stack as the error object.
The reason is that since its functionality is changed, in case some Redis branch / fork uses it,
it's better to have a compilation error than a bug.
The `source` and `line` fields are enriched by the error handler (if possible) and the
`ignore_error_stats_update` is optional and if its not present then the default value is `false`.
If `ignore_error_stats_update` is true, the error will not be counted on the error stats.
When parsing Redis call reply, each error is translated to a Lua table on the format describe
above and the `ignore_error_stats_update` field is set to `true` so we will not count errors
twice (we counted this error when we invoke the command).
The changes in this PR might have been considered as a breaking change for users that used
Lua `pcall` function. Before, the error was a string and now its a table. To keep backward
comparability the PR override the `pcall` implementation and extract the error message from
the error table and return it.
Example of the error stats update:
```
127.0.0.1:6379> lpush l 1
(integer) 2
127.0.0.1:6379> eval "return redis.call('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value. script: e471b73f1ef44774987ab00bdf51f21fd9f7974a, on @user_script:1.
127.0.0.1:6379> info Errorstats
# Errorstats
errorstat_WRONGTYPE:count=1
127.0.0.1:6379> info commandstats
# Commandstats
cmdstat_eval:calls=1,usec=341,usec_per_call=341.00,rejected_calls=0,failed_calls=1
cmdstat_info:calls=1,usec=35,usec_per_call=35.00,rejected_calls=0,failed_calls=0
cmdstat_lpush:calls=1,usec=14,usec_per_call=14.00,rejected_calls=0,failed_calls=0
cmdstat_get:calls=1,usec=10,usec_per_call=10.00,rejected_calls=0,failed_calls=1
```
## error message
We can now construct the error message (sent as a reply to the user) from the error table,
so this solves issues where the error message was malformed and the error code appeared
in the middle of the error message:
```diff
127.0.0.1:6379> eval "return redis.call('set','x','y')" 0
-(error) ERR Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479): @user_script:1: OOM command not allowed when used memory > 'maxmemory'.
+(error) OOM command not allowed when used memory > 'maxmemory' @user_script:1. Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479)
```
```diff
127.0.0.1:6379> eval "redis.call('get', 'l')" 0
-(error) ERR Error running script (call to f_8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1): @user_script:1: WRONGTYPE Operation against a key holding the wrong kind of value
+(error) WRONGTYPE Operation against a key holding the wrong kind of value script: 8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1, on @user_script:1.
```
Notica that `redis.pcall` was not change:
```
127.0.0.1:6379> eval "return redis.pcall('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value
```
## other notes
Notice that Some commands (like GEOADD) changes the cmd variable on the client stats so we
can not count on it to update the command stats. In order to be able to update those stats correctly
we needed to promote `realcmd` variable to be located on the client struct.
Tests was added and modified to verify the changes.
Related PR's: #10279, #10218, #10278, #10309
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-27 13:40:57 +02:00
|
|
|
void addReplyErrorSdsEx(client *c, sds err, int flags);
|
2020-08-27 12:54:01 +03:00
|
|
|
void addReplyErrorSds(client *c, sds err);
|
2022-09-22 16:29:00 +03:00
|
|
|
void addReplyErrorSdsSafe(client *c, sds err);
|
2015-07-26 15:20:46 +02:00
|
|
|
void addReplyError(client *c, const char *err);
|
2022-01-23 16:05:06 +08:00
|
|
|
void addReplyErrorArity(client *c);
|
|
|
|
void addReplyErrorExpireTime(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void addReplyStatus(client *c, const char *status);
|
|
|
|
void addReplyDouble(client *c, double d);
|
2024-05-22 23:24:12 -07:00
|
|
|
void addReplyBigNum(client *c, const char *num, size_t len);
|
2016-02-18 22:08:47 +01:00
|
|
|
void addReplyHumanLongDouble(client *c, long double d);
|
2015-07-26 15:20:46 +02:00
|
|
|
void addReplyLongLong(client *c, long long ll);
|
2018-11-08 13:05:50 +01:00
|
|
|
void addReplyArrayLen(client *c, long length);
|
|
|
|
void addReplyMapLen(client *c, long length);
|
|
|
|
void addReplySetLen(client *c, long length);
|
|
|
|
void addReplyAttributeLen(client *c, long length);
|
|
|
|
void addReplyPushLen(client *c, long length);
|
2017-11-27 17:57:44 +02:00
|
|
|
void addReplyHelp(client *c, const char **help);
|
2023-10-30 17:08:30 +02:00
|
|
|
void addExtendedReplyHelp(client *c, const char **help, const char **extended_help);
|
2018-07-02 18:49:34 +02:00
|
|
|
void addReplySubcommandSyntaxError(client *c);
|
2018-12-04 12:46:16 +01:00
|
|
|
void addReplyLoadedModules(client *c);
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
void copyReplicaOutputBuffer(client *dst, client *src);
|
2021-09-09 17:02:33 +08:00
|
|
|
void addListRangeReply(client *c, robj *o, long start, long end, int reverse);
|
2022-02-13 18:37:32 +02:00
|
|
|
void deferredAfterErrorReply(client *c, list *errors);
|
2016-05-18 08:08:43 +03:00
|
|
|
size_t sdsZmallocSize(sds s);
|
|
|
|
size_t getStringObjectSdsUsedMemory(robj *o);
|
2018-02-21 20:18:34 +02:00
|
|
|
void freeClientReplyValue(void *o);
|
2010-06-22 00:07:48 +02:00
|
|
|
void *dupClientReplyValue(void *o);
|
2015-07-26 15:20:46 +02:00
|
|
|
char *getClientPeerId(client *client);
|
2020-10-28 21:13:44 +02:00
|
|
|
char *getClientSockName(client *client);
|
2024-07-10 13:53:52 +02:00
|
|
|
int isClientConnIpV6(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
sds catClientInfoString(sds s, client *client);
|
2018-06-28 17:43:05 +08:00
|
|
|
sds getAllClientsInfoString(int type);
|
Custom authentication for Modules (#11659)
This change adds new module callbacks that can override the default password based authentication associated with ACLs. With this, Modules can register auth callbacks through which they can implement their own Authentication logic. When `AUTH` and `HELLO AUTH ...` commands are used, Module based authentication is attempted and then normal password based authentication is attempted if needed.
The new Module APIs added in this PR are - `RM_RegisterCustomAuthCallback` and `RM_BlockClientOnAuth` and `RedisModule_ACLAddLogEntryByUserName `.
Module based authentication will be attempted for all Redis users (created through the ACL SETUSER cmd or through Module APIs) even if the Redis user does not exist at the time of the command. This gives a chance for the Module to create the RedisModule user and then authenticate via the RedisModule API - from the custom auth callback.
For the AUTH command, we will support both variations - `AUTH <username> <password>` and `AUTH <password>`. In case of the `AUTH <password>` variation, the custom auth callbacks are triggered with “default” as the username and password as what is provided.
### RedisModule_RegisterCustomAuthCallback
```
void RM_RegisterCustomAuthCallback(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback cb) {
```
This API registers a callback to execute to prior to normal password based authentication. Multiple callbacks can be registered across different modules. These callbacks are responsible for either handling the authentication, each authenticating the user or explicitly denying, or deferring it to other authentication mechanisms. Callbacks are triggered in the order they were registered. When a Module is unloaded, all the auth callbacks registered by it are unregistered. The callbacks are attempted, in the order of most recently registered callbacks, when the AUTH/HELLO (with AUTH field is provided) commands are called. The callbacks will be called with a module context along with a username and a password, and are expected to take one of the following actions:
(1) Authenticate - Use the RM_Authenticate* API successfully and return `REDISMODULE_AUTH_HANDLED`. This will immediately end the auth chain as successful and add the OK reply.
(2) Block a client on authentication - Use the `RM_BlockClientOnAuth` API and return `REDISMODULE_AUTH_HANDLED`. Here, the client will be blocked until the `RM_UnblockClient `API is used which will trigger the auth reply callback (provided earlier through the `RM_BlockClientOnAuth`). In this reply callback, the Module should authenticate, deny or skip handling authentication.
(3) Deny Authentication - Return `REDISMODULE_AUTH_HANDLED` without authenticating or blocking the client. Optionally, `err` can be set to a custom error message. This will immediately end the auth chain as unsuccessful and add the ERR reply.
(4) Skip handling Authentication - Return `REDISMODULE_AUTH_NOT_HANDLED` without blocking the client. This will allow the engine to attempt the next custom auth callback.
If none of the callbacks authenticate or deny auth, then password based auth is attempted and will authenticate or add failure logs and reply to the clients accordingly.
### RedisModule_BlockClientOnAuth
```
RedisModuleBlockedClient *RM_BlockClientOnAuth(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback reply_callback,
void (*free_privdata)(RedisModuleCtx*,void*))
```
This API can only be used from a Module from the custom auth callback. If a client is not in the middle of custom module based authentication, ERROR is returned. Otherwise, the client is blocked and the `RedisModule_BlockedClient` is returned similar to the `RedisModule_BlockClient` API.
### RedisModule_ACLAddLogEntryByUserName
```
int RM_ACLAddLogEntryByUserName(RedisModuleCtx *ctx, RedisModuleString *username, RedisModuleString *object, RedisModuleACLLogEntryReason reason)
```
Adds a new entry in the ACL log with the `username` RedisModuleString provided. This simplifies the Module usage because now, developers do not need to create a Module User just to add an error ACL Log entry. Aside from accepting username (RedisModuleString) instead of a RedisModuleUser, it is the same as the existing `RedisModule_ACLAddLogEntry` API.
### Breaking changes
- HELLO command - Clients can now only set the client name and RESP protocol from the `HELLO` command if they are authenticated. Also, we now finish command arg validation first and return early with a ERR reply if any arg is invalid. This is to avoid mutating the client name / RESP from a command that would have failed on invalid arguments.
### Notable behaviors
- Module unblocking - Now, we will not allow Modules to block the client from inside the context of a reply callback (triggered from the Module unblock flow `moduleHandleBlockedClients`).
---------
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
2023-03-15 15:18:42 -07:00
|
|
|
int clientSetName(client *c, robj *name, const char **err);
|
2015-07-26 15:20:46 +02:00
|
|
|
void rewriteClientCommandVector(client *c, int argc, ...);
|
|
|
|
void rewriteClientCommandArgument(client *c, int i, robj *newval);
|
|
|
|
void replaceClientCommandVector(client *c, int argc, robj **argv);
|
2021-05-19 08:23:54 -07:00
|
|
|
void redactClientCommandArgument(client *c, int argc);
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
size_t getClientOutputBufferMemoryUsage(client *c);
|
|
|
|
size_t getClientMemoryUsage(client *c, size_t *output_buffer_mem_usage);
|
2020-05-14 10:02:57 +02:00
|
|
|
int freeClientsInAsyncFreeQueue(void);
|
2021-05-04 13:45:08 +03:00
|
|
|
int closeClientOnOutputBufferLimitReached(client *c, int async);
|
2015-07-26 15:20:46 +02:00
|
|
|
int getClientType(client *c);
|
2014-06-16 10:43:05 +02:00
|
|
|
int getClientTypeByName(char *name);
|
|
|
|
char *getClientTypeName(int class);
|
2024-06-07 14:21:33 -07:00
|
|
|
void flushReplicasOutputBuffers(void);
|
|
|
|
void disconnectReplicas(void);
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
void evictClients(void);
|
Introduce .listen into connection type
Introduce listen method into connection type, this allows no hard code
of listen logic. Originally, we initialize server during startup like
this:
if (server.port)
listenToPort(server.port,&server.ipfd);
if (server.tls_port)
listenToPort(server.port,&server.tlsfd);
if (server.unixsocket)
anetUnixServer(...server.unixsocket...);
...
if (createSocketAcceptHandler(&server.ipfd, acceptTcpHandler) != C_OK)
if (createSocketAcceptHandler(&server.tlsfd, acceptTcpHandler) != C_OK)
if (createSocketAcceptHandler(&server.sofd, acceptTcpHandler) != C_OK)
...
If a new connection type gets supported, we have to add more hard code
to setup listener.
Introduce .listen and refactor listener, and Unix socket supports this.
this allows to setup listener arguments and create listener in a loop.
What's more, '.listen' is defined in connection.h, so we should include
server.h to import 'struct socketFds', but server.h has already include
'connection.h'. To avoid including loop(also to make code reasonable),
define 'struct connListener' in connection.h instead of 'struct socketFds'
in server.h. This leads this commit to get more changes.
There are more fields in 'struct connListener', hence it's possible to
simplify changeBindAddr & applyTLSPort() & updatePort() into a single
logic: update the listener config from the server.xxx, and re-create
the listener.
Because of the new field 'priv' in struct connListener, we expect to pass
this to the accept handler(even it's not used currently), this may be used
in the future.
Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
2022-07-27 12:18:28 +08:00
|
|
|
int listenToPort(connListener *fds);
|
2022-10-27 11:57:04 +03:00
|
|
|
void pauseActions(pause_purpose purpose, mstime_t end, uint32_t actions_bitmask);
|
|
|
|
void unpauseActions(pause_purpose purpose);
|
|
|
|
uint32_t isPausedActions(uint32_t action_bitmask);
|
|
|
|
uint32_t isPausedActionsWithUpdate(uint32_t action_bitmask);
|
|
|
|
void updatePausedActions(void);
|
2023-05-02 17:31:32 -07:00
|
|
|
void unblockPostponedClients(void);
|
2020-05-14 10:02:57 +02:00
|
|
|
void processEventsWhileBlocked(void);
|
2023-05-02 17:31:32 -07:00
|
|
|
void whileBlockedCron(void);
|
|
|
|
void blockingOperationStarts(void);
|
|
|
|
void blockingOperationEnds(void);
|
2015-09-30 17:23:34 +02:00
|
|
|
int handleClientsWithPendingWrites(void);
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
void adjustThreadedIOIfNeeded(void);
|
2015-09-30 16:41:48 +02:00
|
|
|
int clientHasPendingReplies(client *c);
|
2022-12-06 22:26:56 -08:00
|
|
|
int updateClientMemUsageAndBucket(client *c);
|
|
|
|
void removeClientFromMemUsageBucket(client *c, int allow_eviction);
|
2015-09-30 16:56:02 +02:00
|
|
|
void unlinkClient(client *c);
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
void removeFromServerClientList(client *c);
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
int writeToClient(client *c);
|
2017-12-05 15:59:56 +01:00
|
|
|
void linkClient(client *c);
|
2018-10-09 13:15:41 +02:00
|
|
|
void protectClient(client *c);
|
|
|
|
void unprotectClient(client *c);
|
2024-05-28 21:09:37 +03:00
|
|
|
void initSharedQueryBuf(void);
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
void freeSharedQueryBuf(void);
|
2019-07-03 19:16:20 +02:00
|
|
|
client *lookupClientByID(uint64_t id);
|
2021-10-04 12:10:31 +03:00
|
|
|
int authRequired(client *c);
|
Durability enhancement for appendfsync=always policy (#9678)
Durability of database is a big and old topic, in this regard Redis use AOF to
support it, and `appendfsync=alwasys` policy is the most strict level, guarantee
all data is both written and synced on disk before reply success to client.
But there are some cases have been overlooked, and could lead to durability broken.
1. The most clear one is about threaded-io mode
we should also set client's write handler with `ae_barrier` in
`handleClientsWithPendingWritesUsingThreads`, or the write handler would be
called after read handler in the next event loop, it means the write command result
could be replied to client before flush to AOF.
2. About blocked client (mostly by module)
in `beforeSleep()`, `handleClientsBlockedOnKeys()` should be called before
`flushAppendOnlyFile()`, in case the unblocked clients modify data without persistence
but send reply.
3. When handling `ProcessingEventsWhileBlocked`
normally it takes place when lua/function/module timeout, and we give a chance to users
to kill the slow operation, but we should call `flushAppendOnlyFile()` before
`handleClientsWithPendingWrites()`, in case the other clients in the last event loop get
acknowledge before data persistence.
for a instance:
```
in the same event loop
client A executes set foo bar
client B executes eval "for var=1,10000000,1 do end" 0
```
after the script timeout, client A will get `OK` but lose data after restart (kill redis when
timeout) if we don't flush the write command to AOF.
4. A more complex case about `ProcessingEventsWhileBlocked`
it is lua timeout in transaction, for example
`MULTI; set foo bar; eval "for var=1,10000000,1 do end" 0; EXEC`, then client will get set
command's result before the whole transaction done, that breaks atomicity too.
fortunately, it's already fixed by #5428 (although it's not the original purpose just a side
effect : )), but module timeout should be fixed too.
case 1, 2, 3 are fixed in this commit, the module issue in case 4 needs a followup PR.
2022-04-11 16:08:39 +08:00
|
|
|
void putClientInPendingWriteQueue(client *c);
|
2024-06-28 14:56:13 +08:00
|
|
|
client *createCachedResponseClient(int resp);
|
2024-05-23 02:51:41 +05:30
|
|
|
void deleteCachedResponseClient(client *recording_client);
|
Async IO threads (#758)
This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests
per second, as detailed by [dan touitou](https://github.com/touitou-dan)
in https://github.com/valkey-io/valkey/issues/22. This PR modifies the
IO threads to be fully asynchronous, which is a first and necessary step
to allow more work offloading and better utilization of the IO threads.
### Current IO threads state:
Valkey IO threads were introduced in Redis 6.0 to allow better
utilization of multi-core machines. Before this, Redis was
single-threaded and could only use one CPU core for network and command
processing. The introduction of IO threads helps in offloading the IO
operations to multiple threads.
**Current IO Threads flow:**
1. Initialization: When Redis starts, it initializes a specified number
of IO threads. These threads are in addition to the main thread, each
thread starts with an empty list, the main thread will populate that
list in each event-loop with pending-read-clients or
pending-write-clients.
2. Read Phase: The main thread accepts incoming connections and reads
requests from clients. The reading of requests are offloaded to IO
threads. The main thread puts the clients ready-to-read in a list and
set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick
the clients up, perform the read operation and parse the first incoming
command.
3. Command Processing: After reading the requests, command processing is
still single-threaded and handled by the main thread.
4. Write Phase: Similar to the read phase, the write phase is also be
offloaded to IO threads. The main thread prepares the response in the
clients’ output buffer then the main thread puts the client in the list,
and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO
threads then pick the clients up and perform the write operation to send
the responses back to clients.
5. Synchronization: The main-thread communicate with the threads on how
many jobs left per each thread with atomic counter. The main-thread
doesn’t access the clients while being handled by the IO threads.
**Issues with current implementation:**
* Underutilized Cores: The current implementation of IO-threads leads to
the underutilization of CPU cores.
* The main thread remains responsible for a significant portion of
IO-related tasks that could be offloaded to IO-threads.
* When the main-thread is processing client’s commands, the IO threads
are idle for a considerable amount of time.
* Notably, the main thread's performance during the IO-related tasks is
constrained by the speed of the slowest IO-thread.
* Limited Offloading: Currently, Since the Main-threads waits
synchronously for the IO threads, the Threads perform only read-parse,
and write operations, with parsing done only for the first command. If
the threads can do work asynchronously we may offload more work to the
threads reducing the load from the main-thread.
* TLS: Currently, we don't support IO threads with TLS (where offloading
IO would be more beneficial) since TLS read/write operations are not
thread-safe with the current implementation.
### Suggested change
Non-blocking main thread - The main thread and IO threads will operate
in parallel to maximize efficiency. The main thread will not be blocked
by IO operations. It will continue to process commands independently of
the IO thread's activities.
**Implementation details**
**Inter-thread communication.**
* We use a static, lock-free ring buffer of fixed size (2048 jobs) for
the main thread to send jobs and for the IO to receive them. If the ring
buffer fills up, the main thread will handle the task itself, acting as
back pressure (in case IO operations are more expensive than command
processing). A static ring buffer is a better candidate than a dynamic
job queue as it eliminates the need for allocation/freeing per job.
* An IO job will be in the format: ` [void* function-call-back | void
*data] `where data is either a client to read/write from and the
function-ptr is the function to be called with the data for example
readQueryFromClient using this format we can use it later to offload
other types of works to the IO threads.
* The Ring buffer is one way from the main-thread to the IO thread, Upon
read/write event the main thread will send a read/write job then in
before sleep it will iterate over the pending read/write clients to
checking for each client if the IO threads has already finished handling
it. The IO thread signals it has finished handling a client read/write
by toggling an atomic flag read_state / write_state on the client
struct.
**Thread Safety**
As suggested in this solution, the IO threads are reading from and
writing to the clients' buffers while the main thread may access those
clients.
We must ensure no race conditions or unsafe access occurs while keeping
the Valkey code simple and lock free.
Minimal Action in the IO Threads
The main change is to limit the IO thread operations to the bare
minimum. The IO thread will access only the client's struct and only the
necessary fields in this struct.
The IO threads will be responsible for the following:
* Read Operation: The IO thread will only read and parse a single
command. It will not update the server stats, handle read errors, or
parsing errors. These tasks will be taken care of by the main thread.
* Write Operation: The IO thread will only write the available data. It
will not free the client's replies, handle write errors, or update the
server statistics.
To achieve this without code duplication, the read/write code has been
refactored into smaller, independent components:
* Functions that perform only the read/parse/write calls.
* Functions that handle the read/parse/write results.
This refactor accounts for the majority of the modifications in this PR.
**Client Struct Safe Access**
As we ensure that the IO threads access memory only within the client
struct, we need to ensure thread safety only for the client's struct's
shared fields.
* Query Buffer
* Command parsing - The main thread will not try to parse a command from
the query buffer when a client is offloaded to the IO thread.
* Client's memory checks in client-cron - The main thread will not
access the client query buffer if it is offloaded and will handle the
querybuf grow/shrink when the client is back.
* CLIENT LIST command - The main thread will busy-wait for the IO thread
to finish handling the client, falling back to the current behavior
where the main thread waits for the IO thread to finish their
processing.
* Output Buffer
* The IO thread will not change the client's bufpos and won't free the
client's reply lists. These actions will be done by the main thread on
the client's return from the IO thread.
* bufpos / block→used: As the main thread may change the bufpos, the
reply-block→used, or add/delete blocks to the reply list while the IO
thread writes, we add two fields to the client struct: io_last_bufpos
and io_last_reply_block. The IO thread will write until the
io_last_bufpos, which was set by the main-thread before sending the
client to the IO thread. If more data has been added to the cob in
between, it will be written in the next write-job. In addition, the main
thread will not trim or merge reply blocks while the client is
offloaded.
* Parsing Fields
* Client's cmd, argc, argv, reqtype, etc., are set during parsing.
* The main thread will indicate to the IO thread not to parse a cmd if
the client is not reset. In this case, the IO thread will only read from
the network and won't attempt to parse a new command.
* The main thread won't access the c→cmd/c→argv in the CLIENT LIST
command as stated before it will busy wait for the IO threads.
* Client Flags
* c→flags, which may be changed by the main thread in multiple places,
won't be accessed by the IO thread. Instead, the main thread will set
the c→io_flags with the information necessary for the IO thread to know
the client's state.
* Client Close
* On freeClient, the main thread will busy wait for the IO thread to
finish processing the client's read/write before proceeding to free the
client.
* Client's Memory Limits
* The IO thread won't handle the qb/cob limits. In case a client crosses
the qb limit, the IO thread will stop reading for it, letting the main
thread know that the client crossed the limit.
**TLS**
TLS is currently not supported with IO threads for the following
reasons:
1. Pending reads - If SSL has pending data that has already been read
from the socket, there is a risk of not calling the read handler again.
To handle this, a list is used to hold the pending clients. With IO
threads, multiple threads can access the list concurrently.
2. Event loop modification - Currently, the TLS code
registers/unregisters the file descriptor from the event loop depending
on the read/write results. With IO threads, multiple threads can modify
the event loop struct simultaneously.
3. The same client can be sent to 2 different threads concurrently
(https://github.com/redis/redis/issues/12540).
Those issues were handled in the current PR:
1. The IO thread only performs the read operation. The main thread will
check for pending reads after the client returns from the IO thread and
will be the only one to access the pending list.
2. The registering/unregistering of events will be similarly postponed
and handled by the main thread only.
3. Each client is being sent to the same dedicated thread (c→id %
num_of_threads).
**Sending Replies Immediately with IO threads.**
Currently, after processing a command, we add the client to the
pending_writes_list. Only after processing all the clients do we send
all the replies. Since the IO threads are now working asynchronously, we
can send the reply immediately after processing the client’s requests,
reducing the command latency. However, if we are using AOF=always, we
must wait for the AOF buffer to be written, in which case we revert to
the current behavior.
**IO threads dynamic adjustment**
Currently, we use an all-or-nothing approach when activating the IO
threads. The current logic is as follows: if the number of pending write
clients is greater than twice the number of threads (including the main
thread), we enable all threads; otherwise, we enable none. For example,
if 8 IO threads are defined, we enable all 8 threads if there are 16
pending clients; else, we enable none.
It makes more sense to enable partial activation of the IO threads. If
we have 10 pending clients, we will enable 5 threads, and so on. This
approach allows for a more granular and efficient allocation of
resources based on the current workload.
In addition, the user will now be able to change the number of I/O
threads at runtime. For example, when decreasing the number of threads
from 4 to 2, threads 3 and 4 will be closed after flushing their job
queues.
**Tests**
Currently, we run the io-threads tests with 4 IO threads
(https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353).
This means that we will not activate the IO threads unless there are 8
(threads * 2) pending write clients per single loop, which is unlikely
to happened in most of tests, meaning the IO threads are not currently
being tested.
To enforce the main thread to always offload work to the IO threads,
regardless of the number of pending events, we add an
events-per-io-thread configuration with a default value of 2. When set
to 0, this configuration will force the main thread to always offload
work to the IO threads.
When we offload every single read/write operation to the IO threads, the
IO-threads are running with 100% CPU when running multiple tests
concurrently some tests fail as a result of larger than expected command
latencies. To address this issue, we have to add some after or wait_for
calls to some of the tests to ensure they pass with IO threads as well.
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
|
|
|
void waitForClientIO(client *c);
|
|
|
|
void ioThreadReadQueryFromClient(void *data);
|
|
|
|
void ioThreadWriteToClient(void *data);
|
|
|
|
int canParseCommand(client *c);
|
|
|
|
int processIOThreadsReadDone(void);
|
|
|
|
int processIOThreadsWriteDone(void);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
Add reply_schema to command json files (internal for now) (#10273)
Work in progress towards implementing a reply schema as part of COMMAND DOCS, see #9845
Since ironing the details of the reply schema of each and every command can take a long time, we
would like to merge this PR when the infrastructure is ready, and let this mature in the unstable branch.
Meanwhile the changes of this PR are internal, they are part of the repo, but do not affect the produced build.
### Background
In #9656 we add a lot of information about Redis commands, but we are missing information about the replies
### Motivation
1. Documentation. This is the primary goal.
2. It should be possible, based on the output of COMMAND, to be able to generate client code in typed
languages. In order to do that, we need Redis to tell us, in detail, what each reply looks like.
3. We would like to build a fuzzer that verifies the reply structure (for now we use the existing
testsuite, see the "Testing" section)
### Schema
The idea is to supply some sort of schema for the various replies of each command.
The schema will describe the conceptual structure of the reply (for generated clients), as defined in RESP3.
Note that the reply structure itself may change, depending on the arguments (e.g. `XINFO STREAM`, with
and without the `FULL` modifier)
We decided to use the standard json-schema (see https://json-schema.org/) as the reply-schema.
Example for `BZPOPMIN`:
```
"reply_schema": {
"oneOf": [
{
"description": "Timeout reached and no elements were popped.",
"type": "null"
},
{
"description": "The keyname, popped member, and its score.",
"type": "array",
"minItems": 3,
"maxItems": 3,
"items": [
{
"description": "Keyname",
"type": "string"
},
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
]
}
```
#### Notes
1. It is ok that some commands' reply structure depends on the arguments and it's the caller's responsibility
to know which is the relevant one. this comes after looking at other request-reply systems like OpenAPI,
where the reply schema can also be oneOf and the caller is responsible to know which schema is the relevant one.
2. The reply schemas will describe RESP3 replies only. even though RESP3 is structured, we want to use reply
schema for documentation (and possibly to create a fuzzer that validates the replies)
3. For documentation, the description field will include an explanation of the scenario in which the reply is sent,
including any relation to arguments. for example, for `ZRANGE`'s two schemas we will need to state that one
is with `WITHSCORES` and the other is without.
4. For documentation, there will be another optional field "notes" in which we will add a short description of
the representation in RESP2, in case it's not trivial (RESP3's `ZRANGE`'s nested array vs. RESP2's flat
array, for example)
Given the above:
1. We can generate the "return" section of all commands in [redis-doc](https://redis.io/commands/)
(given that "description" and "notes" are comprehensive enough)
2. We can generate a client in a strongly typed language (but the return type could be a conceptual
`union` and the caller needs to know which schema is relevant). see the section below for RESP2 support.
3. We can create a fuzzer for RESP3.
### Limitations (because we are using the standard json-schema)
The problem is that Redis' replies are more diverse than what the json format allows. This means that,
when we convert the reply to a json (in order to validate the schema against it), we lose information (see
the "Testing" section below).
The other option would have been to extend the standard json-schema (and json format) to include stuff
like sets, bulk-strings, error-string, etc. but that would mean also extending the schema-validator - and that
seemed like too much work, so we decided to compromise.
Examples:
1. We cannot tell the difference between an "array" and a "set"
2. We cannot tell the difference between simple-string and bulk-string
3. we cannot verify true uniqueness of items in commands like ZRANGE: json-schema doesn't cover the
case of two identical members with different scores (e.g. `[["m1",6],["m1",7]]`) because `uniqueItems`
compares (member,score) tuples and not just the member name.
### Testing
This commit includes some changes inside Redis in order to verify the schemas (existing and future ones)
are indeed correct (i.e. describe the actual response of Redis).
To do that, we added a debugging feature to Redis that causes it to produce a log of all the commands
it executed and their replies.
For that, Redis needs to be compiled with `-DLOG_REQ_RES` and run with
`--reg-res-logfile <file> --client-default-resp 3` (the testsuite already does that if you run it with
`--log-req-res --force-resp3`)
You should run the testsuite with the above args (and `--dont-clean`) in order to make Redis generate
`.reqres` files (same dir as the `stdout` files) which contain request-response pairs.
These files are later on processed by `./utils/req-res-log-validator.py` which does:
1. Goes over req-res files, generated by redis-servers, spawned by the testsuite (see logreqres.c)
2. For each request-response pair, it validates the response against the request's reply_schema
(obtained from the extended COMMAND DOCS)
5. In order to get good coverage of the Redis commands, and all their different replies, we chose to use
the existing redis test suite, rather than attempt to write a fuzzer.
#### Notes about RESP2
1. We will not be able to use the testing tool to verify RESP2 replies (we are ok with that, it's time to
accept RESP3 as the future RESP)
2. Since the majority of the test suite is using RESP2, and we want the server to reply with RESP3
so that we can validate it, we will need to know how to convert the actual reply to the one expected.
- number and boolean are always strings in RESP2 so the conversion is easy
- objects (maps) are always a flat array in RESP2
- others (nested array in RESP3's `ZRANGE` and others) will need some special per-command
handling (so the client will not be totally auto-generated)
Example for ZRANGE:
```
"reply_schema": {
"anyOf": [
{
"description": "A list of member elements",
"type": "array",
"uniqueItems": true,
"items": {
"type": "string"
}
},
{
"description": "Members and their scores. Returned in case `WITHSCORES` was used.",
"notes": "In RESP2 this is returned as a flat array",
"type": "array",
"uniqueItems": true,
"items": {
"type": "array",
"minItems": 2,
"maxItems": 2,
"items": [
{
"description": "Member",
"type": "string"
},
{
"description": "Score",
"type": "number"
}
]
}
}
]
}
```
### Other changes
1. Some tests that behave differently depending on the RESP are now being tested for both RESP,
regardless of the special log-req-res mode ("Pub/Sub PING" for example)
2. Update the history field of CLIENT LIST
3. Added basic tests for commands that were not covered at all by the testsuite
### TODO
- [x] (maybe a different PR) add a "condition" field to anyOf/oneOf schemas that refers to args. e.g.
when `SET` return NULL, the condition is `arguments.get||arguments.condition`, for `OK` the condition
is `!arguments.get`, and for `string` the condition is `arguments.get` - https://github.com/redis/redis/issues/11896
- [x] (maybe a different PR) also run `runtest-cluster` in the req-res logging mode
- [x] add the new tests to GH actions (i.e. compile with `-DLOG_REQ_RES`, run the tests, and run the validator)
- [x] (maybe a different PR) figure out a way to warn about (sub)schemas that are uncovered by the output
of the tests - https://github.com/redis/redis/issues/11897
- [x] (probably a separate PR) add all missing schemas
- [x] check why "SDOWN is triggered by misconfigured instance replying with errors" fails with --log-req-res
- [x] move the response transformers to their own file (run both regular, cluster, and sentinel tests - need to
fight with the tcl including mechanism a bit)
- [x] issue: module API - https://github.com/redis/redis/issues/11898
- [x] (probably a separate PR): improve schemas: add `required` to `object`s - https://github.com/redis/redis/issues/11899
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Hanna Fadida <hanna.fadida@redislabs.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Shaya Potter <shaya@redislabs.com>
2023-03-11 09:14:16 +01:00
|
|
|
/* logreqres.c - logging of requests and responses */
|
|
|
|
void reqresReset(client *c, int free_buf);
|
|
|
|
void reqresSaveClientReplyOffset(client *c);
|
|
|
|
size_t reqresAppendRequest(client *c);
|
|
|
|
size_t reqresAppendResponse(client *c);
|
|
|
|
|
2010-09-02 19:52:24 +02:00
|
|
|
#ifdef __GNUC__
|
2024-05-22 23:24:12 -07:00
|
|
|
void addReplyErrorFormatEx(client *c, int flags, const char *fmt, ...) __attribute__((format(printf, 3, 4)));
|
|
|
|
void addReplyErrorFormat(client *c, const char *fmt, ...) __attribute__((format(printf, 2, 3)));
|
|
|
|
void addReplyStatusFormat(client *c, const char *fmt, ...) __attribute__((format(printf, 2, 3)));
|
2010-09-02 19:52:24 +02:00
|
|
|
#else
|
Sort out the mess around Lua error messages and error stats (#10329)
This PR fix 2 issues on Lua scripting:
* Server error reply statistics (some errors were counted twice).
* Error code and error strings returning from scripts (error code was missing / misplaced).
## Statistics
a Lua script user is considered part of the user application, a sophisticated transaction,
so we want to count an error even if handled silently by the script, but when it is
propagated outwards from the script we don't wanna count it twice. on the other hand,
if the script decides to throw an error on its own (using `redis.error_reply`), we wanna
count that too.
Besides, we do count the `calls` in command statistics for the commands the script calls,
we we should certainly also count `failed_calls`.
So when a simple `eval "return redis.call('set','x','y')" 0` fails, it should count the failed call
to both SET and EVAL, but the `errorstats` and `total_error_replies` should be counted only once.
The PR changes the error object that is raised on errors. Instead of raising a simple Lua
string, Redis will raise a Lua table in the following format:
```
{
err='<error message (including error code)>',
source='<User source file name>',
line='<line where the error happned>',
ignore_error_stats_update=true/false,
}
```
The `luaPushError` function was modified to construct the new error table as describe above.
The `luaRaiseError` was renamed to `luaError` and is now simply called `lua_error` to raise
the table on the top of the Lua stack as the error object.
The reason is that since its functionality is changed, in case some Redis branch / fork uses it,
it's better to have a compilation error than a bug.
The `source` and `line` fields are enriched by the error handler (if possible) and the
`ignore_error_stats_update` is optional and if its not present then the default value is `false`.
If `ignore_error_stats_update` is true, the error will not be counted on the error stats.
When parsing Redis call reply, each error is translated to a Lua table on the format describe
above and the `ignore_error_stats_update` field is set to `true` so we will not count errors
twice (we counted this error when we invoke the command).
The changes in this PR might have been considered as a breaking change for users that used
Lua `pcall` function. Before, the error was a string and now its a table. To keep backward
comparability the PR override the `pcall` implementation and extract the error message from
the error table and return it.
Example of the error stats update:
```
127.0.0.1:6379> lpush l 1
(integer) 2
127.0.0.1:6379> eval "return redis.call('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value. script: e471b73f1ef44774987ab00bdf51f21fd9f7974a, on @user_script:1.
127.0.0.1:6379> info Errorstats
# Errorstats
errorstat_WRONGTYPE:count=1
127.0.0.1:6379> info commandstats
# Commandstats
cmdstat_eval:calls=1,usec=341,usec_per_call=341.00,rejected_calls=0,failed_calls=1
cmdstat_info:calls=1,usec=35,usec_per_call=35.00,rejected_calls=0,failed_calls=0
cmdstat_lpush:calls=1,usec=14,usec_per_call=14.00,rejected_calls=0,failed_calls=0
cmdstat_get:calls=1,usec=10,usec_per_call=10.00,rejected_calls=0,failed_calls=1
```
## error message
We can now construct the error message (sent as a reply to the user) from the error table,
so this solves issues where the error message was malformed and the error code appeared
in the middle of the error message:
```diff
127.0.0.1:6379> eval "return redis.call('set','x','y')" 0
-(error) ERR Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479): @user_script:1: OOM command not allowed when used memory > 'maxmemory'.
+(error) OOM command not allowed when used memory > 'maxmemory' @user_script:1. Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479)
```
```diff
127.0.0.1:6379> eval "redis.call('get', 'l')" 0
-(error) ERR Error running script (call to f_8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1): @user_script:1: WRONGTYPE Operation against a key holding the wrong kind of value
+(error) WRONGTYPE Operation against a key holding the wrong kind of value script: 8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1, on @user_script:1.
```
Notica that `redis.pcall` was not change:
```
127.0.0.1:6379> eval "return redis.pcall('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value
```
## other notes
Notice that Some commands (like GEOADD) changes the cmd variable on the client stats so we
can not count on it to update the command stats. In order to be able to update those stats correctly
we needed to promote `realcmd` variable to be located on the client struct.
Tests was added and modified to verify the changes.
Related PR's: #10279, #10218, #10278, #10309
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-27 13:40:57 +02:00
|
|
|
void addReplyErrorFormatEx(client *c, int flags, const char *fmt, ...);
|
2015-07-26 15:20:46 +02:00
|
|
|
void addReplyErrorFormat(client *c, const char *fmt, ...);
|
|
|
|
void addReplyStatusFormat(client *c, const char *fmt, ...);
|
2010-09-02 19:52:24 +02:00
|
|
|
#endif
|
|
|
|
|
2019-06-29 20:08:41 -04:00
|
|
|
/* Client side caching (tracking mode) */
|
2024-07-01 02:33:10 +08:00
|
|
|
void enableTracking(client *c, uint64_t redirect_to, struct ClientFlags options, robj **prefix, size_t numprefix);
|
2019-06-29 20:08:41 -04:00
|
|
|
void disableTracking(client *c);
|
2023-02-16 08:07:35 +02:00
|
|
|
void trackingRememberKeys(client *tracking, client *executing);
|
2021-10-07 20:13:42 +08:00
|
|
|
void trackingInvalidateKey(client *c, robj *keyobj, int bcast);
|
|
|
|
void trackingScheduleKeyInvalidation(uint64_t client_id, robj *keyobj);
|
|
|
|
void trackingHandlePendingKeyInvalidations(void);
|
2020-12-23 19:13:12 -08:00
|
|
|
void trackingInvalidateKeysOnFlush(int async);
|
2021-04-19 22:16:27 -07:00
|
|
|
void freeTrackingRadixTree(rax *rt);
|
2020-12-23 19:13:12 -08:00
|
|
|
void freeTrackingRadixTreeAsync(rax *rt);
|
2024-03-19 14:18:22 +08:00
|
|
|
void freeErrorsRadixTreeAsync(rax *errors);
|
2019-07-22 18:59:53 +02:00
|
|
|
void trackingLimitUsedSlots(void);
|
2020-02-07 14:03:43 +01:00
|
|
|
uint64_t trackingGetTotalItems(void);
|
2020-02-07 18:12:45 +01:00
|
|
|
uint64_t trackingGetTotalKeys(void);
|
2020-05-02 19:20:44 -04:00
|
|
|
uint64_t trackingGetTotalPrefixes(void);
|
2020-02-11 18:11:59 +01:00
|
|
|
void trackingBroadcastInvalidationMessages(void);
|
2021-01-08 00:00:35 -08:00
|
|
|
int checkPrefixCollisionsOrReply(client *c, robj **prefix, size_t numprefix);
|
2019-06-29 20:08:41 -04:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* List data type */
|
|
|
|
void listTypePush(robj *subject, robj *value, int where);
|
|
|
|
robj *listTypePop(robj *subject, int where);
|
2016-06-20 23:08:06 +03:00
|
|
|
unsigned long listTypeLength(const robj *subject);
|
2012-01-31 10:35:52 +01:00
|
|
|
listTypeIterator *listTypeInitIterator(robj *subject, long index, unsigned char direction);
|
2010-06-22 00:07:48 +02:00
|
|
|
void listTypeReleaseIterator(listTypeIterator *li);
|
Add listpack encoding for list (#11303)
Improve memory efficiency of list keys
## Description of the feature
The new listpack encoding uses the old `list-max-listpack-size` config
to perform the conversion, which we can think it of as a node inside a
quicklist, but without 80 bytes overhead (internal fragmentation included)
of quicklist and quicklistNode structs.
For example, a list key with 5 items of 10 chars each, now takes 128 bytes
instead of 208 it used to take.
## Conversion rules
* Convert listpack to quicklist
When the listpack length or size reaches the `list-max-listpack-size` limit,
it will be converted to a quicklist.
* Convert quicklist to listpack
When a quicklist has only one node, and its length or size is reduced to half
of the `list-max-listpack-size` limit, it will be converted to a listpack.
This is done to avoid frequent conversions when we add or remove at the bounding size or length.
## Interface changes
1. add list entry param to listTypeSetIteratorDirection
When list encoding is listpack, `listTypeIterator->lpi` points to the next entry of current entry,
so when changing the direction, we need to use the current node (listTypeEntry->p) to
update `listTypeIterator->lpi` to the next node in the reverse direction.
## Benchmark
### Listpack VS Quicklist with one node
* LPUSH - roughly 0.3% improvement
* LRANGE - roughly 13% improvement
### Both are quicklist
* LRANGE - roughly 3% improvement
* LRANGE without pipeline - roughly 3% improvement
From the benchmark, as we can see from the results
1. When list is quicklist encoding, LRANGE improves performance by <5%.
2. When list is listpack encoding, LRANGE improves performance by ~13%,
the main enhancement is brought by `addListListpackRangeReply()`.
## Memory usage
1M lists(key:0~key:1000000) with 5 items of 10 chars ("hellohello") each.
shows memory usage down by 35.49%, from 214MB to 138MB.
## Note
1. Add conversion callback to support doing some work before conversion
Since the quicklist iterator decompresses the current node when it is released, we can
no longer decompress the quicklist after we convert the list.
2022-11-17 02:29:46 +08:00
|
|
|
void listTypeSetIteratorDirection(listTypeIterator *li, listTypeEntry *entry, unsigned char direction);
|
2010-06-22 00:07:48 +02:00
|
|
|
int listTypeNext(listTypeIterator *li, listTypeEntry *entry);
|
|
|
|
robj *listTypeGet(listTypeEntry *entry);
|
Add listpack encoding for list (#11303)
Improve memory efficiency of list keys
## Description of the feature
The new listpack encoding uses the old `list-max-listpack-size` config
to perform the conversion, which we can think it of as a node inside a
quicklist, but without 80 bytes overhead (internal fragmentation included)
of quicklist and quicklistNode structs.
For example, a list key with 5 items of 10 chars each, now takes 128 bytes
instead of 208 it used to take.
## Conversion rules
* Convert listpack to quicklist
When the listpack length or size reaches the `list-max-listpack-size` limit,
it will be converted to a quicklist.
* Convert quicklist to listpack
When a quicklist has only one node, and its length or size is reduced to half
of the `list-max-listpack-size` limit, it will be converted to a listpack.
This is done to avoid frequent conversions when we add or remove at the bounding size or length.
## Interface changes
1. add list entry param to listTypeSetIteratorDirection
When list encoding is listpack, `listTypeIterator->lpi` points to the next entry of current entry,
so when changing the direction, we need to use the current node (listTypeEntry->p) to
update `listTypeIterator->lpi` to the next node in the reverse direction.
## Benchmark
### Listpack VS Quicklist with one node
* LPUSH - roughly 0.3% improvement
* LRANGE - roughly 13% improvement
### Both are quicklist
* LRANGE - roughly 3% improvement
* LRANGE without pipeline - roughly 3% improvement
From the benchmark, as we can see from the results
1. When list is quicklist encoding, LRANGE improves performance by <5%.
2. When list is listpack encoding, LRANGE improves performance by ~13%,
the main enhancement is brought by `addListListpackRangeReply()`.
## Memory usage
1M lists(key:0~key:1000000) with 5 items of 10 chars ("hellohello") each.
shows memory usage down by 35.49%, from 214MB to 138MB.
## Note
1. Add conversion callback to support doing some work before conversion
Since the quicklist iterator decompresses the current node when it is released, we can
no longer decompress the quicklist after we convert the list.
2022-11-17 02:29:46 +08:00
|
|
|
unsigned char *listTypeGetValue(listTypeEntry *entry, size_t *vlen, long long *lval);
|
2010-06-22 00:07:48 +02:00
|
|
|
void listTypeInsert(listTypeEntry *entry, robj *value, int where);
|
2021-09-14 16:48:06 +02:00
|
|
|
void listTypeReplace(listTypeEntry *entry, robj *value);
|
2010-06-22 00:07:48 +02:00
|
|
|
int listTypeEqual(listTypeEntry *entry, robj *o);
|
2014-11-13 14:11:47 -05:00
|
|
|
void listTypeDelete(listTypeIterator *iter, listTypeEntry *entry);
|
2020-11-17 19:03:05 +09:00
|
|
|
robj *listTypeDup(robj *o);
|
Add listpack encoding for list (#11303)
Improve memory efficiency of list keys
## Description of the feature
The new listpack encoding uses the old `list-max-listpack-size` config
to perform the conversion, which we can think it of as a node inside a
quicklist, but without 80 bytes overhead (internal fragmentation included)
of quicklist and quicklistNode structs.
For example, a list key with 5 items of 10 chars each, now takes 128 bytes
instead of 208 it used to take.
## Conversion rules
* Convert listpack to quicklist
When the listpack length or size reaches the `list-max-listpack-size` limit,
it will be converted to a quicklist.
* Convert quicklist to listpack
When a quicklist has only one node, and its length or size is reduced to half
of the `list-max-listpack-size` limit, it will be converted to a listpack.
This is done to avoid frequent conversions when we add or remove at the bounding size or length.
## Interface changes
1. add list entry param to listTypeSetIteratorDirection
When list encoding is listpack, `listTypeIterator->lpi` points to the next entry of current entry,
so when changing the direction, we need to use the current node (listTypeEntry->p) to
update `listTypeIterator->lpi` to the next node in the reverse direction.
## Benchmark
### Listpack VS Quicklist with one node
* LPUSH - roughly 0.3% improvement
* LRANGE - roughly 13% improvement
### Both are quicklist
* LRANGE - roughly 3% improvement
* LRANGE without pipeline - roughly 3% improvement
From the benchmark, as we can see from the results
1. When list is quicklist encoding, LRANGE improves performance by <5%.
2. When list is listpack encoding, LRANGE improves performance by ~13%,
the main enhancement is brought by `addListListpackRangeReply()`.
## Memory usage
1M lists(key:0~key:1000000) with 5 items of 10 chars ("hellohello") each.
shows memory usage down by 35.49%, from 214MB to 138MB.
## Note
1. Add conversion callback to support doing some work before conversion
Since the quicklist iterator decompresses the current node when it is released, we can
no longer decompress the quicklist after we convert the list.
2022-11-17 02:29:46 +08:00
|
|
|
void listTypeDelRange(robj *o, long start, long stop);
|
2015-07-26 15:20:46 +02:00
|
|
|
void popGenericCommand(client *c, int where);
|
2022-09-29 02:07:38 +08:00
|
|
|
void listElementsRemoved(client *c, robj *key, int where, robj *o, long count, int signal, int *deleted);
|
Add listpack encoding for list (#11303)
Improve memory efficiency of list keys
## Description of the feature
The new listpack encoding uses the old `list-max-listpack-size` config
to perform the conversion, which we can think it of as a node inside a
quicklist, but without 80 bytes overhead (internal fragmentation included)
of quicklist and quicklistNode structs.
For example, a list key with 5 items of 10 chars each, now takes 128 bytes
instead of 208 it used to take.
## Conversion rules
* Convert listpack to quicklist
When the listpack length or size reaches the `list-max-listpack-size` limit,
it will be converted to a quicklist.
* Convert quicklist to listpack
When a quicklist has only one node, and its length or size is reduced to half
of the `list-max-listpack-size` limit, it will be converted to a listpack.
This is done to avoid frequent conversions when we add or remove at the bounding size or length.
## Interface changes
1. add list entry param to listTypeSetIteratorDirection
When list encoding is listpack, `listTypeIterator->lpi` points to the next entry of current entry,
so when changing the direction, we need to use the current node (listTypeEntry->p) to
update `listTypeIterator->lpi` to the next node in the reverse direction.
## Benchmark
### Listpack VS Quicklist with one node
* LPUSH - roughly 0.3% improvement
* LRANGE - roughly 13% improvement
### Both are quicklist
* LRANGE - roughly 3% improvement
* LRANGE without pipeline - roughly 3% improvement
From the benchmark, as we can see from the results
1. When list is quicklist encoding, LRANGE improves performance by <5%.
2. When list is listpack encoding, LRANGE improves performance by ~13%,
the main enhancement is brought by `addListListpackRangeReply()`.
## Memory usage
1M lists(key:0~key:1000000) with 5 items of 10 chars ("hellohello") each.
shows memory usage down by 35.49%, from 214MB to 138MB.
## Note
1. Add conversion callback to support doing some work before conversion
Since the quicklist iterator decompresses the current node when it is released, we can
no longer decompress the quicklist after we convert the list.
2022-11-17 02:29:46 +08:00
|
|
|
typedef enum {
|
|
|
|
LIST_CONV_AUTO,
|
|
|
|
LIST_CONV_GROWING,
|
|
|
|
LIST_CONV_SHRINKING,
|
|
|
|
} list_conv_type;
|
|
|
|
typedef void (*beforeConvertCB)(void *data);
|
|
|
|
void listTypeTryConversion(robj *o, list_conv_type lct, beforeConvertCB fn, void *data);
|
|
|
|
void listTypeTryConversionAppend(robj *o, robj **argv, int start, int end, beforeConvertCB fn, void *data);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* MULTI/EXEC/WATCH... */
|
2015-07-26 15:20:46 +02:00
|
|
|
void unwatchAllKeys(client *c);
|
|
|
|
void initClientMultiState(client *c);
|
|
|
|
void freeClientMultiState(client *c);
|
Expose script flags to processCommand for better handling (#10744)
The important part is that read-only scripts (not just EVAL_RO
and FCALL_RO, but also ones with `no-writes` executed by normal EVAL or
FCALL), will now be permitted to run during CLIENT PAUSE WRITE (unlike
before where only the _RO commands would be processed).
Other than that, some errors like OOM, READONLY, MASTERDOWN are now
handled by processCommand, rather than the command itself affects the
error string (and even error code in some cases), and command stats.
Besides that, now the `may-replicate` commands, PFCOUNT and PUBLISH, will
be considered `write` commands in scripts and will be blocked in all
read-only scripts just like other write commands.
They'll also be blocked in EVAL_RO (i.e. even for scripts without the
`no-writes` shebang flag.
This commit also hides the `may_replicate` flag from the COMMAND command
output. this is a **breaking change**.
background about may_replicate:
We don't want to expose a no-may-replicate flag or alike to scripts, since we
consider the may-replicate thing an internal concern of redis, that we may
some day get rid of.
In fact, the may-replicate flag was initially introduced to flag EVAL: since
we didn't know what it's gonna do ahead of execution, before function-flags
existed). PUBLISH and PFCOUNT, both of which because they have side effects
which may some day be fixed differently.
code changes:
The changes in eval.c are mostly code re-ordering:
- evalCalcFunctionName is extracted out of evalGenericCommand
- evalExtractShebangFlags is extracted luaCreateFunction
- evalGetCommandFlags is new code
2022-06-01 14:09:40 +03:00
|
|
|
void queueMultiCommand(client *c, uint64_t cmd_flags);
|
Client eviction (#8687)
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-09-23 14:02:16 +03:00
|
|
|
size_t multiStateMemOverhead(client *c);
|
2024-04-03 10:02:43 +07:00
|
|
|
void touchWatchedKey(serverDb *db, robj *key);
|
2021-07-11 13:17:23 +03:00
|
|
|
int isWatchedKeyExpired(client *c);
|
2024-04-03 10:02:43 +07:00
|
|
|
void touchAllWatchedKeysInDb(serverDb *emptied, serverDb *replaced_with);
|
2015-07-26 15:20:46 +02:00
|
|
|
void discardTransaction(client *c);
|
|
|
|
void flagTransaction(client *c);
|
EXEC always fails with EXECABORT and multi-state is cleared
In order to support the use of multi-exec in pipeline, it is important that
MULTI and EXEC are never rejected and it is easy for the client to know if the
connection is still in multi state.
It was easy to make sure MULTI and DISCARD never fail (done by previous
commits) since these only change the client state and don't do any actual
change in the server, but EXEC is a different story.
Since in the past, it was possible for clients to handle some EXEC errors and
retry the EXEC, we now can't affort to return any error on EXEC other than
EXECABORT, which now carries with it the real reason for the abort too.
Other fixes in this commit:
- Some checks that where performed at the time of queuing need to be re-
validated when EXEC runs, for instance if the transaction contains writes
commands, it needs to be aborted. there was one check that was already done
in execCommand (-READONLY), but other checks where missing: -OOM, -MISCONF,
-NOREPLICAS, -MASTERDOWN
- When a command is rejected by processCommand it was rejected with addReply,
which was not recognized as an error in case the bad command came from the
master. this will enable to count or MONITOR these errors in the future.
- make it easier for tests to create additional (non deferred) clients.
- add tests for the fixes of this commit.
2020-06-11 21:09:35 +03:00
|
|
|
void execCommandAbort(client *c, sds error);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* Object implementation */
|
2013-01-24 11:27:10 +01:00
|
|
|
void decrRefCount(robj *o);
|
|
|
|
void decrRefCountVoid(void *o);
|
2010-06-22 00:07:48 +02:00
|
|
|
void incrRefCount(robj *o);
|
2015-07-30 11:46:31 +02:00
|
|
|
robj *makeObjectShared(robj *o);
|
2010-06-22 00:07:48 +02:00
|
|
|
void freeStringObject(robj *o);
|
|
|
|
void freeListObject(robj *o);
|
|
|
|
void freeSetObject(robj *o);
|
|
|
|
void freeZsetObject(robj *o);
|
|
|
|
void freeHashObject(robj *o);
|
Use madvise(MADV_DONTNEED) to release memory to reduce COW (#8974)
## Backgroud
As we know, after `fork`, one process will copy pages when writing data to these
pages(CoW), and another process still keep old pages, they totally cost more memory.
For redis, we suffered that redis consumed much memory when the fork child is serializing
key/values, even that maybe cause OOM.
But actually we find, in redis fork child process, the child process don't need to keep some
memory and parent process may write or update that, for example, child process will never
access the key-value that is serialized but users may update it in parent process.
So we think it may reduce COW if the child process release memory that it is not needed.
## Implementation
For releasing key value in child process, we may think we call `decrRefCount` to free memory,
but i find the fork child process still use much memory when we don't write any data to redis,
and it costs much more time that slows down bgsave. Maybe because memory allocator doesn't
really release memory to OS, and it may modify some inner data for this free operation, especially
when we free small objects.
Moreover, CoW is based on pages, so it is a easy way that we only free the memory bulk that is
not less than kernel page size. madvise(MADV_DONTNEED) can quickly release specified region
pages to OS bypassing memory allocator, and allocator still consider that this memory still is used
and don't change its inner data.
There are some buffers we can release in the fork child process:
- **Serialized key-values**
the fork child process never access serialized key-values, so we try to free them.
Because we only can release big bulk memory, and it is time consumed to iterate all
items/members/fields/entries of complex data type. So we decide to iterate them and
try to release them only when their average size of item/member/field/entry is more
than page size of OS.
- **Replication backlog**
Because replication backlog is a cycle buffer, it will be changed quickly if redis has heavy
write traffic, but in fork child process, we don't need to access that.
- **Client buffers**
If clients have requests during having the fork child process, clients' buffer also be changed
frequently. The memory includes client query buffer, output buffer, and client struct used memory.
To get child process peak private dirty memory, we need to count peak memory instead
of last used memory, because the child process may continue to release memory (since
COW used to only grow till now, the last was equivalent to the peak).
Also we're adding a new `current_cow_peak` info variable (to complement the existing
`current_cow_size`)
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-08-05 04:01:46 +08:00
|
|
|
void dismissObject(robj *o, size_t dump_size);
|
2010-06-22 00:07:48 +02:00
|
|
|
robj *createObject(int type, void *ptr);
|
2023-05-24 14:40:11 +08:00
|
|
|
void initObjectLRUOrLFU(robj *o);
|
2015-03-11 16:59:56 +01:00
|
|
|
robj *createStringObject(const char *ptr, size_t len);
|
|
|
|
robj *createRawStringObject(const char *ptr, size_t len);
|
|
|
|
robj *createEmbeddedStringObject(const char *ptr, size_t len);
|
2021-08-05 22:56:14 +03:00
|
|
|
robj *tryCreateRawStringObject(const char *ptr, size_t len);
|
|
|
|
robj *tryCreateStringObject(const char *ptr, size_t len);
|
2016-06-22 20:57:24 +03:00
|
|
|
robj *dupStringObject(const robj *o);
|
2015-07-31 18:01:23 +02:00
|
|
|
int isSdsRepresentableAsLongLong(sds s, long long *llval);
|
2011-04-27 13:24:52 +02:00
|
|
|
int isObjectRepresentableAsLongLong(robj *o, long long *llongval);
|
2010-06-22 00:07:48 +02:00
|
|
|
robj *tryObjectEncoding(robj *o);
|
2023-05-30 10:43:25 +03:00
|
|
|
robj *tryObjectEncodingEx(robj *o, int try_trim);
|
2010-06-22 00:07:48 +02:00
|
|
|
robj *getDecodedObject(robj *o);
|
|
|
|
size_t stringObjectLen(robj *o);
|
|
|
|
robj *createStringObjectFromLongLong(long long value);
|
2018-06-18 16:54:13 +02:00
|
|
|
robj *createStringObjectFromLongLongForValue(long long value);
|
2023-06-20 20:14:44 +08:00
|
|
|
robj *createStringObjectFromLongLongWithSds(long long value);
|
2014-12-02 18:19:30 +01:00
|
|
|
robj *createStringObjectFromLongDouble(long double value, int humanfriendly);
|
2024-02-08 20:36:11 +08:00
|
|
|
robj *createQuicklistObject(int fill, int compress);
|
Add listpack encoding for list (#11303)
Improve memory efficiency of list keys
## Description of the feature
The new listpack encoding uses the old `list-max-listpack-size` config
to perform the conversion, which we can think it of as a node inside a
quicklist, but without 80 bytes overhead (internal fragmentation included)
of quicklist and quicklistNode structs.
For example, a list key with 5 items of 10 chars each, now takes 128 bytes
instead of 208 it used to take.
## Conversion rules
* Convert listpack to quicklist
When the listpack length or size reaches the `list-max-listpack-size` limit,
it will be converted to a quicklist.
* Convert quicklist to listpack
When a quicklist has only one node, and its length or size is reduced to half
of the `list-max-listpack-size` limit, it will be converted to a listpack.
This is done to avoid frequent conversions when we add or remove at the bounding size or length.
## Interface changes
1. add list entry param to listTypeSetIteratorDirection
When list encoding is listpack, `listTypeIterator->lpi` points to the next entry of current entry,
so when changing the direction, we need to use the current node (listTypeEntry->p) to
update `listTypeIterator->lpi` to the next node in the reverse direction.
## Benchmark
### Listpack VS Quicklist with one node
* LPUSH - roughly 0.3% improvement
* LRANGE - roughly 13% improvement
### Both are quicklist
* LRANGE - roughly 3% improvement
* LRANGE without pipeline - roughly 3% improvement
From the benchmark, as we can see from the results
1. When list is quicklist encoding, LRANGE improves performance by <5%.
2. When list is listpack encoding, LRANGE improves performance by ~13%,
the main enhancement is brought by `addListListpackRangeReply()`.
## Memory usage
1M lists(key:0~key:1000000) with 5 items of 10 chars ("hellohello") each.
shows memory usage down by 35.49%, from 214MB to 138MB.
## Note
1. Add conversion callback to support doing some work before conversion
Since the quicklist iterator decompresses the current node when it is released, we can
no longer decompress the quicklist after we convert the list.
2022-11-17 02:29:46 +08:00
|
|
|
robj *createListListpackObject(void);
|
2010-06-22 00:07:48 +02:00
|
|
|
robj *createSetObject(void);
|
2010-07-02 19:57:12 +02:00
|
|
|
robj *createIntsetObject(void);
|
2022-11-09 18:50:07 +01:00
|
|
|
robj *createSetListpackObject(void);
|
2010-06-22 00:07:48 +02:00
|
|
|
robj *createHashObject(void);
|
|
|
|
robj *createZsetObject(void);
|
2021-09-09 23:18:53 +08:00
|
|
|
robj *createZsetListpackObject(void);
|
2017-08-30 12:40:27 +02:00
|
|
|
robj *createStreamObject(void);
|
2016-05-18 11:45:40 +02:00
|
|
|
robj *createModuleObject(moduleType *mt, void *value);
|
2015-07-26 15:20:46 +02:00
|
|
|
int getLongFromObjectOrReply(client *c, robj *o, long *target, const char *msg);
|
2020-11-05 12:58:54 -05:00
|
|
|
int getPositiveLongFromObjectOrReply(client *c, robj *o, long *target, const char *msg);
|
|
|
|
int getRangeLongFromObjectOrReply(client *c, robj *o, long min, long max, long *target, const char *msg);
|
2015-07-26 15:20:46 +02:00
|
|
|
int checkType(client *c, robj *o, int type);
|
|
|
|
int getLongLongFromObjectOrReply(client *c, robj *o, long long *target, const char *msg);
|
|
|
|
int getDoubleFromObjectOrReply(client *c, robj *o, double *target, const char *msg);
|
2016-06-20 23:08:06 +03:00
|
|
|
int getDoubleFromObject(const robj *o, double *target);
|
2010-06-22 00:07:48 +02:00
|
|
|
int getLongLongFromObject(robj *o, long long *target);
|
2011-11-12 19:27:35 +01:00
|
|
|
int getLongDoubleFromObject(robj *o, long double *target);
|
2015-07-26 15:20:46 +02:00
|
|
|
int getLongDoubleFromObjectOrReply(client *c, robj *o, long double *target, const char *msg);
|
Improve dbid range check for SELECT, MOVE, COPY (#8085)
SELECT used to read the index into a `long` variable, and then pass it to a function
that takes an `int`, possibly causing an overflow before the range check.
Now all these commands use better and cleaner range check, and that also results in
a slight change of the error response in case of an invalid database index.
SELECT:
in the past it would have returned either `-ERR invalid DB index` (if not a number),
or `-ERR DB index is out of range` (if not between 1..16 or alike).
now it'll return either `-ERR value is out of range` (if not a number), or
`-ERR value is out of range, value must between -2147483648 and 2147483647`
(if not in the range for an int), or `-ERR DB index is out of range`
(if not between 0..16 or alike)
MOVE:
in the past it would only fail with `-ERR index out of range` no matter the reason.
now return the same errors as the new ones for SELECT mentioned above.
(i.e. unlike for SELECT even for a value like 17 we changed the error message)
COPY:
doesn't really matter how it behaved in the past (new command), new behavior is
like the above two.
2020-12-02 03:41:26 +08:00
|
|
|
int getIntFromObjectOrReply(client *c, robj *o, int *target, const char *msg);
|
2010-06-22 00:07:48 +02:00
|
|
|
char *strEncoding(int encoding);
|
2022-07-19 13:59:39 +08:00
|
|
|
int compareStringObjects(const robj *a, const robj *b);
|
|
|
|
int collateStringObjects(const robj *a, const robj *b);
|
2010-06-22 00:07:48 +02:00
|
|
|
int equalStringObjects(robj *a, robj *b);
|
2014-03-20 11:47:12 +01:00
|
|
|
unsigned long long estimateObjectIdleTime(robj *o);
|
2023-02-28 19:38:58 +02:00
|
|
|
void trimStringObjectIfNeeded(robj *o, int trim_small_values);
|
Unshare object to avoid LRU/LFU being messed up (#250)
When LRU/LFU enabled, Valkey does not allow using shared objects, as
value objects may be shared among many different keys and they can't
share LRU/LFU information.
However `maxmemory-policy` is modifiable at runtime. If LRU/LFU is not
enabled at start, but then enabled when some shared objects are already
used, there could be some confusion in LRU/LFU information.
For `set` command it is OK since it is going to create a new object when
LRU/LFU enabled, but `get` command will not unshare the object and just
update LRU/LFU information.
So we may duplicate the object in this case. It is a one-time task for
each key using shared objects, unless this is the case for so many keys,
there should be no serious performance degradation.
Still, LRU will be updated anyway, no matter LRU/LFU is enabled or not,
because `OBJECT IDLETIME` needs it, unless `maxmemory-policy` is set to
LFU. So idle time of a key may still be messed up.
---------
Signed-off-by: chentianjie.ctj <chentianjie.ctj@alibaba-inc.com>
Signed-off-by: Chen Tianjie <TJ_Chen@outlook.com>
2024-06-01 16:09:20 +08:00
|
|
|
static inline int canUseSharedObject(void) {
|
|
|
|
return server.maxmemory == 0 || !(server.maxmemory_policy & MAXMEMORY_FLAG_NO_SHARED_INTEGERS);
|
|
|
|
}
|
2015-07-26 15:28:00 +02:00
|
|
|
#define sdsEncodedObject(objptr) (objptr->encoding == OBJ_ENCODING_RAW || objptr->encoding == OBJ_ENCODING_EMBSTR)
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2010-10-24 16:22:52 +02:00
|
|
|
/* Synchronous I/O with timeout */
|
2012-03-31 11:21:45 +02:00
|
|
|
ssize_t syncWrite(int fd, char *ptr, ssize_t size, long long timeout);
|
|
|
|
ssize_t syncRead(int fd, char *ptr, ssize_t size, long long timeout);
|
|
|
|
ssize_t syncReadLine(int fd, char *ptr, ssize_t size, long long timeout);
|
2010-10-24 16:22:52 +02:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Replication */
|
2024-06-07 14:21:33 -07:00
|
|
|
void replicationFeedReplicas(int dictid, robj **argv, int argc);
|
|
|
|
void replicationFeedStreamFromPrimaryStream(char *buf, size_t buflen);
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
void resetReplicationBuffer(void);
|
|
|
|
void feedReplicationBuffer(char *buf, size_t len);
|
|
|
|
void freeReplicaReferencedReplBuffer(client *replica);
|
2015-07-26 15:20:46 +02:00
|
|
|
void replicationFeedMonitors(client *c, list *monitors, int dictid, robj **argv, int argc);
|
2024-06-07 14:21:33 -07:00
|
|
|
void updateReplicasWaitingBgsave(int bgsaveerr, int type);
|
2010-11-04 17:29:53 +01:00
|
|
|
void replicationCron(void);
|
Accelerate diskless master connections, and general re-connections (#6271)
Diskless master has some inherent latencies.
1) fork starts with delay from cron rather than immediately
2) replica is put online only after an ACK. but the ACK
was sent only once a second.
3) but even if it would arrive immediately, it will not
register in case cron didn't yet detect that the fork is done.
Besides that, when a replica disconnects, it doesn't immediately
attempts to re-connect, it waits for replication cron (one per second).
in case it was already online, it may be important to try to re-connect
as soon as possible, so that the backlog at the master doesn't vanish.
In case it disconnected during rdb transfer, one can argue that it's
not very important to re-connect immediately, but this is needed for the
"diskless loading short read" test to be able to run 100 iterations in 5
seconds, rather than 3 (waiting for replication cron re-connection)
changes in this commit:
1) sync command starts a fork immediately if no sync_delay is configured
2) replica sends REPLCONF ACK when done reading the rdb (rather than on 1s cron)
3) when a replica unexpectedly disconnets, it immediately tries to
re-connect rather than waiting 1s
4) when when a child exits, if there is another replica waiting, we spawn a new
one right away, instead of waiting for 1s replicationCron.
5) added a call to connectWithMaster from replicationSetMaster. which is called
from the REPLICAOF command but also in 3 places in cluster.c, in all of
these the connection attempt will now be immediate instead of delayed by 1
second.
side note:
we can add a call to rdbPipeReadHandler in replconfCommand when getting
a REPLCONF ACK from the replica to solve a race where the replica got
the entire rdb and EOF marker before we detected that the pipe was
closed.
in the test i did see this race happens in one about of some 300 runs,
but i concluded that this race is unlikely in real life (where the
replica is on another host and we're more likely to first detect the
pipe was closed.
the test runs 100 iterations in 3 seconds, so in some cases it'll take 4
seconds instead (waiting for another REPLCONF ACK).
Removing unneeded startBgsaveForReplication from updateSlavesWaitingForBgsave
Now that CheckChildrenDone is calling the new replicationStartPendingFork
(extracted from serverCron) there's actually no need to call
startBgsaveForReplication from updateSlavesWaitingForBgsave anymore,
since as soon as updateSlavesWaitingForBgsave returns, CheckChildrenDone is
calling replicationStartPendingFork that handles that anyway.
The code in updateSlavesWaitingForBgsave had a bug in which it ignored
repl-diskless-sync-delay, but removing that code shows that this bug was
hiding another bug, which is that the max_idle should have used >= and
not >, this one second delay has a big impact on my new test.
2020-08-06 16:53:06 +03:00
|
|
|
void replicationStartPendingFork(void);
|
2024-06-07 14:21:33 -07:00
|
|
|
void replicationHandlePrimaryDisconnection(void);
|
|
|
|
void replicationCachePrimary(client *c);
|
2023-05-02 17:31:32 -07:00
|
|
|
void resizeReplicationBacklog(void);
|
Fix data loss when replica do a failover with a old history repl offset (#885)
Our current replica can initiate a failover without restriction when
it detects that the primary node is offline. This is generally not a
problem. However, consider the following scenarios:
1. In slot migration, a primary loses its last slot and then becomes
a replica. When it is fully synchronized with the new primary, the new
primary downs.
2. In CLUSTER REPLICATE command, a replica becomes a replica of another
primary. When it is fully synchronized with the new primary, the new
primary downs.
In the above scenario, case 1 may cause the empty primary to be elected
as the new primary, resulting in primary data loss. Case 2 may cause the
non-empty replica to be elected as the new primary, resulting in data
loss and confusion.
The reason is that we have cached primary logic, which is used for psync.
In the above scenario, when clusterSetPrimary is called, myself will cache
server.primary in server.cached_primary for psync. In replicationGetReplicaOffset,
we get server.cached_primary->reploff for offset, gossip it and rank it,
which causes the replica to use the old historical offset to initiate
failover, and it get a good rank, initiates election first, and then is
elected as the new primary.
The main problem here is that when the replica has not completed full
sync, it may get the historical offset in replicationGetReplicaOffset.
The fix is to clear cached_primary in these places where full sync is
obviously needed, and let the replica use offset == 0 to participate
in the election. In this way, this unhealthy replica has a worse rank
and is not easy to be elected.
Of course, it is possible that it will be elected with offset == 0.
In the future, we may need to prohibit the replica with offset == 0
from having the right to initiate elections.
Another point worth mentioning, in above cases:
1. In the ROLE command, the replica status will be handshake, and the
offset will be -1.
2. Before this PR, in the CLUSTER SHARD command, the replica status will
be online, and the offset will be the old cached value (which is wrong).
3. After this PR, in the CLUSTER SHARD, the replica status will be loading,
and the offset will be 0.
Signed-off-by: Binbin <binloveplay1314@qq.com>
2024-08-21 12:42:59 +08:00
|
|
|
void replicationSetPrimary(char *ip, int port, int full_sync_required);
|
2024-06-07 14:21:33 -07:00
|
|
|
void replicationUnsetPrimary(void);
|
|
|
|
void refreshGoodReplicasCount(void);
|
Add new RM_Call flags for script mode, no writes, and error replies. (#10372)
The PR extends RM_Call with 3 new capabilities using new flags that
are given to RM_Call as part of the `fmt` argument.
It aims to assist modules that are getting a list of commands to be
executed from the user (not hard coded as part of the module logic),
think of a module that implements a new scripting language...
* `S` - Run the command in a script mode, this means that it will raise an
error if a command which are not allowed inside a script (flaged with the
`deny-script` flag) is invoked (like SHUTDOWN). In addition, on script mode,
write commands are not allowed if there is not enough good replicas (as
configured with `min-replicas-to-write`) and/or a disk error happened.
* `W` - no writes mode, Redis will reject any command that is marked with `write`
flag. Again can be useful to modules that implement a new scripting language
and wants to prevent any write commands.
* `E` - Return errors as RedisModuleCallReply. Today the errors that happened
before the command was invoked (like unknown commands or acl error) return
a NULL reply and set errno. This might be missing important information about
the failure and it is also impossible to just pass the error to the user using
RM_ReplyWithCallReply. This new flag allows you to get a RedisModuleCallReply
object with the relevant error message and treat it as if it was an error that was
raised by the command invocation.
Tests were added to verify the new code paths.
In addition small refactoring was done to share some code between modules,
scripts, and `processCommand` function:
1. `getAclErrorMessage` was added to `acl.c` to unified to log message extraction
from the acl result
2. `checkGoodReplicasStatus` was added to `replication.c` to check the status of
good replicas. It is used on `scriptVerifyWriteCommandAllow`, `RM_Call`, and
`processCommand`.
3. `writeCommandsGetDiskErrorMessage` was added to `server.c` to get the error
message on persistence failure. Again it is used on `scriptVerifyWriteCommandAllow`,
`RM_Call`, and `processCommand`.
2022-03-22 14:13:28 +02:00
|
|
|
int checkGoodReplicasStatus(void);
|
2013-12-04 15:52:20 +01:00
|
|
|
void processClientsWaitingReplicas(void);
|
2015-07-26 15:20:46 +02:00
|
|
|
void unblockClientWaitingReplicas(client *c);
|
2013-12-04 15:52:20 +01:00
|
|
|
int replicationCountAcksByOffset(long long offset);
|
Implementing the WAITAOF command (issue #10505) (#11713)
Implementing the WAITAOF functionality which would allow the user to
block until a specified number of Redises have fsynced all previous write
commands to the AOF.
Syntax: `WAITAOF <num_local> <num_replicas> <timeout>`
Response: Array containing two elements: num_local, num_replicas
num_local is always either 0 or 1 representing the local AOF on the master.
num_replicas is the number of replicas that acknowledged the a replication
offset of the last write being fsynced to the AOF.
Returns an error when called on replicas, or when called with non-zero
num_local on a master with AOF disabled, in all other cases the response
just contains number of fsync copies.
Main changes:
* Added code to keep track of replication offsets that are confirmed to have
been fsynced to disk.
* Keep advancing master_repl_offset even when replication is disabled (and
there's no replication backlog, only if there's an AOF enabled).
This way we can use this command and it's mechanisms even when replication
is disabled.
* Extend REPLCONF ACK to `REPLCONF ACK <ofs> FACK <ofs>`, the FACK
will be appended only if there's an AOF on the replica, and already ignored on
old masters (thus backwards compatible)
* WAIT now no longer wait for the replication offset after your last command, but
rather the replication offset after your last write (or read command that caused
propagation, e.g. lazy expiry).
Unrelated changes:
* WAIT command respects CLIENT_DENY_BLOCKING (not just CLIENT_MULTI)
Implementation details:
* Add an atomic var named `fsynced_reploff_pending` that's updated
(usually by the bio thread) and later copied to the main `fsynced_reploff`
variable (only if the AOF base file exists).
I.e. during the initial AOF rewrite it will not be used as the fsynced offset
since the AOF base is still missing.
* Replace close+fsync bio job with new BIO_CLOSE_AOF (AOF specific)
job that will also update fsync offset the field.
* Handle all AOF jobs (BIO_CLOSE_AOF, BIO_AOF_FSYNC) in the same bio
worker thread, to impose ordering on their execution. This solves a
race condition where a job could set `fsynced_reploff_pending` to a higher
value than another pending fsync job, resulting in indicating an offset
for which parts of the data have not yet actually been fsynced.
Imposing an ordering on the jobs guarantees that fsync jobs are executed
in increasing order of replication offset.
* Drain bio jobs when switching `appendfsync` to "always"
This should prevent a write race between updates to `fsynced_reploff_pending`
in the main thread (`flushAppendOnlyFile` when set to ALWAYS fsync), and
those done in the bio thread.
* Drain the pending fsync when starting over a new AOF to avoid race conditions
with the previous AOF offsets overriding the new one (e.g. after switching to
replicate from a new master).
* Make sure to update the fsynced offset at the end of the initial AOF rewrite.
a must in case there are no additional writes that trigger a periodic fsync,
specifically for a replica that does a full sync.
Limitations:
It is possible to write a module and a Lua script that propagate to the AOF and doesn't
propagate to the replication stream. see REDISMODULE_ARGV_NO_REPLICAS and luaRedisSetReplCommand.
These features are incompatible with the WAITAOF command, and can result
in two bad cases. The scenario is that the user executes command that only
propagates to AOF, and then immediately
issues a WAITAOF, and there's no further writes on the replication stream after that.
1. if the the last thing that happened on the replication stream is a PING
(which increased the replication offset but won't trigger an fsync on the replica),
then the client would hang forever (will wait for an fack that the replica will never
send sine it doesn't trigger any fsyncs).
2. if the last thing that happened is a write command that got propagated properly,
then WAITAOF will be released immediately, without waiting for an fsync (since
the offset didn't change)
Refactoring:
* Plumbing to allow bio worker to handle multiple job types
This introduces infrastructure necessary to allow BIO workers to
not have a 1-1 mapping of worker to job-type. This allows in the
future to assign multiple job types to a single worker, either as
a performance/resource optimization, or as a way of enforcing
ordering between specific classes of jobs.
Co-authored-by: Oran Agra <oran@redislabs.com>
2023-03-14 20:26:21 +02:00
|
|
|
int replicationCountAOFAcksByOffset(long long offset);
|
2024-06-07 14:21:33 -07:00
|
|
|
void replicationSendNewlineToPrimary(void);
|
|
|
|
long long replicationGetReplicaOffset(void);
|
|
|
|
char *replicationGetReplicaName(client *c);
|
2015-08-04 16:56:00 +02:00
|
|
|
long long getPsyncInitialOffset(void);
|
2024-06-07 14:21:33 -07:00
|
|
|
int replicationSetupReplicaForFullResync(client *replica, long long offset);
|
PSYNC2: different improvements to Redis replication.
The gist of the changes is that now, partial resynchronizations between
slaves and masters (without the need of a full resync with RDB transfer
and so forth), work in a number of cases when it was impossible
in the past. For instance:
1. When a slave is promoted to mastrer, the slaves of the old master can
partially resynchronize with the new master.
2. Chained slalves (slaves of slaves) can be moved to replicate to other
slaves or the master itsef, without requiring a full resync.
3. The master itself, after being turned into a slave, is able to
partially resynchronize with the new master, when it joins replication
again.
In order to obtain this, the following main changes were operated:
* Slaves also take a replication backlog, not just masters.
* Same stream replication for all the slaves and sub slaves. The
replication stream is identical from the top level master to its slaves
and is also the same from the slaves to their sub-slaves and so forth.
This means that if a slave is later promoted to master, it has the
same replication backlong, and can partially resynchronize with its
slaves (that were previously slaves of the old master).
* A given replication history is no longer identified by the `runid` of
a Redis node. There is instead a `replication ID` which changes every
time the instance has a new history no longer coherent with the past
one. So, for example, slaves publish the same replication history of
their master, however when they are turned into masters, they publish
a new replication ID, but still remember the old ID, so that they are
able to partially resynchronize with slaves of the old master (up to a
given offset).
* The replication protocol was slightly modified so that a new extended
+CONTINUE reply from the master is able to inform the slave of a
replication ID change.
* REPLCONF CAPA is used in order to notify masters that a slave is able
to understand the new +CONTINUE reply.
* The RDB file was extended with an auxiliary field that is able to
select a given DB after loading in the slave, so that the slave can
continue receiving the replication stream from the point it was
disconnected without requiring the master to insert "SELECT" statements.
This is useful in order to guarantee the "same stream" property, because
the slave must be able to accumulate an identical backlog.
* Slave pings to sub-slaves are now sent in a special form, when the
top-level master is disconnected, in order to don't interfer with the
replication stream. We just use out of band "\n" bytes as in other parts
of the Redis protocol.
An old design document is available here:
https://gist.github.com/antirez/ae068f95c0d084891305
However the implementation is not identical to the description because
during the work to implement it, different changes were needed in order
to make things working well.
2016-11-09 11:31:06 +01:00
|
|
|
void changeReplicationId(void);
|
|
|
|
void clearReplicationId2(void);
|
2021-09-13 15:39:11 +08:00
|
|
|
void createReplicationBacklog(void);
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
void freeReplicationBacklog(void);
|
2024-06-07 14:21:33 -07:00
|
|
|
void replicationCachePrimaryUsingMyself(void);
|
Fix replication of SLAVEOF inside transaction.
In Redis 4.0 replication, with the introduction of PSYNC2, masters and
slaves replicate commands to cascading slaves and to the replication
backlog itself in a different way compared to the past.
Masters actually replicate the effects of client commands.
Slaves just propagate what they receive from masters.
This mechanism can cause problems when the configuration of an instance
is changed from master to slave inside a transaction. For instance
we could send to a master instance the following sequence:
MULTI
SLAVEOF 127.0.0.1 0
EXEC
SLAVEOF NO ONE
Before the fixes in this commit, the MULTI command used to be propagated
into the replication backlog, however after the SLAVEOF command the
instance is a slave, so the EXEC implementation failed to also propagate
the EXEC command. When the slaves of the above instance reconnected,
they were incrementally synchronized just sending a "MULTI". This put
the master client (in the slaves) into MULTI state, breaking the
replication.
Notably even Redis Sentinel uses the above approach in order to guarantee
that configuration changes are always performed together with rewrites
of the configuration and with clients disconnection. Sentiel does:
MULTI
SLAVEOF ...
CONFIG REWRITE
CLIENT KILL TYPE normal
EXEC
So this was a really problematic issue. However even with the fix in
this commit, that will add the final EXEC to the replication stream in
case the instance was switched from master to slave during the
transaction, the result would be to increment the slave replication
offset, so a successive reconnection with the new master, will not
permit a successful partial resynchronization: no way the new master can
provide us with the backlog needed, we incremented our offset to a value
that the new master cannot have.
However the EXEC implementation waits to emit the MULTI, so that if the
commands inside the transaction actually do not need to be replicated,
no commands propagation happens at all. From multi.c:
if (!must_propagate && !(c->cmd->flags & (CMD_READONLY|CMD_ADMIN))) {
execCommandPropagateMulti(c);
must_propagate = 1;
}
The above code is already modified by this commit you are reading.
Now also ADMIN commands do not trigger the emission of MULTI. It is actually
not clear why we do not just check for CMD_WRITE... Probably I wrote it this
way in order to make the code more reliable: better to over-emit MULTI
than not emitting it in time.
So this commit should indeed fix issue #3836 (verified), however it looks
like some reconsideration of this code path is needed in the long term.
BONUS POINT: The reverse bug.
Even in a read only slave "B", in a replication setup like:
A -> B -> C
There are commands without the READONLY nor the ADMIN flag, that are also
not flagged as WRITE commands. An example is just the PING command.
So if we send B the following sequence:
MULTI
PING
SLAVEOF NO ONE
EXEC
The result will be the reverse bug, where only EXEC is emitted, but not the
previous MULTI. However this apparently does not create problems in practice
but it is yet another acknowledge of the fact some work is needed here
in order to make this code path less surprising.
Note that there are many different approaches we could follow. For instance
MULTI/EXEC blocks containing administrative commands may be allowed ONLY
if all the commands are administrative ones, otherwise they could be
denined. When allowed, the commands could simply never be replicated at all.
2017-07-12 11:07:28 +02:00
|
|
|
void feedReplicationBacklog(void *ptr, size_t len);
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
void incrementalTrimReplicationBacklog(size_t blocks);
|
|
|
|
int canFeedReplicaReplBuffer(client *replica);
|
2021-11-02 16:53:52 +08:00
|
|
|
void rebaseReplicationBuffer(long long base_repl_offset);
|
2020-05-28 10:08:16 +02:00
|
|
|
void showLatestBacklog(void);
|
2019-08-11 16:07:53 +03:00
|
|
|
void rdbPipeReadHandler(struct aeEventLoop *eventLoop, int fd, void *clientData, int mask);
|
|
|
|
void rdbPipeWriteHandlerConnRemoved(struct connection *conn);
|
2024-06-07 20:32:27 -07:00
|
|
|
int rdbRegisterAuxField(char *auxfield, rdbAuxFieldEncoder encoder, rdbAuxFieldDecoder decoder);
|
2021-01-28 13:18:05 -08:00
|
|
|
void clearFailoverState(void);
|
|
|
|
void updateFailoverStatus(void);
|
|
|
|
void abortFailover(const char *err);
|
2023-05-02 17:31:32 -07:00
|
|
|
const char *getFailoverStateString(void);
|
Dual channel replication (#60)
In this PR we introduce the main benefit of dual channel replication by
continuously steaming the COB (client output buffers) in parallel to the
RDB and thus keeping the primary's side COB small AND accelerating the
overall sync process. By streaming the replication data to the replica
during the full sync, we reduce
1. Memory load from the primary's node.
2. CPU load from the primary's main process. [Latest performance
tests](#data)
## Motivation
* Reduce primary memory load. We do that by moving the COB tracking to
the replica side. This also decrease the chance for COB overruns. Note
that primary's input buffer limits at the replica side are less
restricted then primary's COB as the replica plays less critical part in
the replication group. While increasing the primary’s COB may end up
with primary reaching swap and clients suffering, at replica side we’re
more at ease with it. Larger COB means better chance to sync
successfully.
* Reduce primary main process CPU load. By opening a new, dedicated
connection for the RDB transfer, child processes can have direct access
to the new connection. Due to TLS connection restrictions, this was not
possible using one main connection. We eliminate the need for the child
process to use the primary's child-proc -> main-proc pipeline, thus
freeing up the main process to process clients queries.
## Dual Channel Replication high level interface design
- Dual channel replication begins when the replica sends a `REPLCONF
CAPA DUALCHANNEL` to the primary during initial
handshake. This is used to state that the replica is capable of dual
channel sync and that this is the replica's main channel, which is not
used for snapshot transfer.
- When replica lacks sufficient data for PSYNC, the primary will send
`-FULLSYNCNEEDED` response instead
of RDB data. As a next step, the replica creates a new connection
(rdb-channel) and configures it against
the primary with the appropriate capabilities and requirements. The
replica then requests a sync
using the RDB channel.
- Prior to forking, the primary sends the replica the snapshot's end
repl-offset, and attaches the replica
to the replication backlog to keep repl data until the replica requests
psync. The replica uses the main
channel to request a PSYNC starting at the snapshot end offset.
- The primary main threads sends incremental changes via the main
channel, while the bgsave process
sends the RDB directly to the replica via the rdb-channel. As for the
replica, the incremental
changes are stored on a local buffer, while the RDB is loaded into
memory.
- Once the replica completes loading the rdb, it drops the
rdb-connection and streams the accumulated incremental
changes into memory. Repl steady state continues normally.
## New replica state machine

## Data <a name="data"></a>



## Explanation
These graphs demonstrate performance improvements during full sync
sessions using rdb-channel + streaming rdb directly from the background
process to the replica.
First graph- with at most 50 clients and light weight commands, we saw
5%-7.5% improvement in write latency during sync session.
Two graphs below- full sync was tested during heavy read commands from
the primary (such as sdiff, sunion on large sets). In that case, the
child process writes to the replica without sharing CPU with the loaded
main process. As a result, this not only improves client response time,
but may also shorten sync time by about 50%. The shorter sync time
results in less memory being used to store replication diffs (>60% in
some of the tested cases).
## Test setup
Both primary and replica in the performance tests ran on the same
machine. RDB size in all tests is 3.7gb. I generated write load using
valkey-benchmark ` ./valkey-benchmark -r 100000 -n 6000000 lpush my_list
__rand_int__`.
---------
Signed-off-by: naglera <anagler123@gmail.com>
Signed-off-by: naglera <58042354+naglera@users.noreply.github.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Co-authored-by: Ping Xie <pingxie@outlook.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-17 23:59:33 +03:00
|
|
|
int sendCurrentOffsetToReplica(client *replica);
|
|
|
|
void addRdbReplicaToPsyncWait(client *replica);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2010-11-08 11:52:03 +01:00
|
|
|
/* Generic persistence functions */
|
2024-05-22 23:24:12 -07:00
|
|
|
void startLoadingFile(size_t size, char *filename, int rdbflags);
|
Replica keep serving data during repl-diskless-load=swapdb for better availability (#9323)
For diskless replication in swapdb mode, considering we already spend replica memory
having a backup of current db to restore in case of failure, we can have the following benefits
by instead swapping database only in case we succeeded in transferring db from master:
- Avoid `LOADING` response during failed and successful synchronization for cases where the
replica is already up and running with data.
- Faster total time of diskless replication, because now we're moving from Transfer + Flush + Load
time to Transfer + Load only. Flushing the tempDb is done asynchronously after swapping.
- This could be implemented also for disk replication with similar benefits if consumers are willing
to spend the extra memory usage.
General notes:
- The concept of `backupDb` becomes `tempDb` for clarity.
- Async loading mode will only kick in if the replica is syncing from a master that has the same
repl-id the one it had before. i.e. the data it's getting belongs to a different time of the same timeline.
- New property in INFO: `async_loading` to differentiate from the blocking loading
- Slot to Key mapping is now a field of `redisDb` as it's more natural to access it from both server.db
and the tempDb that is passed around.
- Because this is affecting replicas only, we assume that if they are not readonly and write commands
during replication, they are lost after SYNC same way as before, but we're still denying CONFIG SET
here anyways to avoid complications.
Considerations for review:
- We have many cases where server.loading flag is used and even though I tried my best, there may
be cases where async_loading should be checked as well and cases where it shouldn't (would require
very good understanding of whole code)
- Several places that had different behavior depending on the loading flag where actually meant to just
handle commands coming from the AOF client differently than ones coming from real clients, changed
to check CLIENT_ID_AOF instead.
**Additional for Release Notes**
- Bugfix - server.dirty was not incremented for any kind of diskless replication, as effect it wouldn't
contribute on triggering next database SAVE
- New flag for RM_GetContextFlags module API: REDISMODULE_CTX_FLAGS_ASYNC_LOADING
- Deprecated RedisModuleEvent_ReplBackup. Starting from Redis 7.0, we don't fire this event.
Instead, we have the new RedisModuleEvent_ReplAsyncLoad holding 3 sub-events: STARTED,
ABORTED and COMPLETED.
- New module flag REDISMODULE_OPTIONS_HANDLE_REPL_ASYNC_LOAD for RedisModule_SetModuleOptions
to allow modules to declare they support the diskless replication with async loading (when absent, we fall
back to disk-based loading).
Co-authored-by: Eduardo Semprebon <edus@saxobank.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-11-04 09:46:50 +01:00
|
|
|
void startLoading(size_t size, int rdbflags, int async);
|
Implement Multi Part AOF mechanism to avoid AOFRW overheads. (#9788)
Implement Multi-Part AOF mechanism to avoid overheads during AOFRW.
Introducing a folder with multiple AOF files tracked by a manifest file.
The main issues with the the original AOFRW mechanism are:
* buffering of commands that are processed during rewrite (consuming a lot of RAM)
* freezes of the main process when the AOFRW completes to drain the remaining part of the buffer and fsync it.
* double disk IO for the data that arrives during AOFRW (had to be written to both the old and new AOF files)
The main modifications of this PR:
1. Remove the AOF rewrite buffer and related code.
2. Divide the AOF into multiple files, they are classified as two types, one is the the `BASE` type,
it represents the full amount of data (Maybe AOF or RDB format) after each AOFRW, there is only
one `BASE` file at most. The second is `INCR` type, may have more than one. They represent the
incremental commands since the last AOFRW.
3. Use a AOF manifest file to record and manage these AOF files mentioned above.
4. The original configuration of `appendfilename` will be the base part of the new file name, for example:
`appendonly.aof.1.base.rdb` and `appendonly.aof.2.incr.aof`
5. Add manifest-related TCL tests, and modified some existing tests that depend on the `appendfilename`
6. Remove the `aof_rewrite_buffer_length` field in info.
7. Add `aof-disable-auto-gc` configuration. By default we're automatically deleting HISTORY type AOFs.
It also gives users the opportunity to preserve the history AOFs. just for testing use now.
8. Add AOFRW limiting measure. When the AOFRW failures reaches the threshold (3 times now),
we will delay the execution of the next AOFRW by 1 minute. If the next AOFRW also fails, it will be
delayed by 2 minutes. The next is 4, 8, 16, the maximum delay is 60 minutes (1 hour). During the limit
period, we can still use the 'bgrewriteaof' command to execute AOFRW immediately.
9. Support upgrade (load) data from old version redis.
10. Add `appenddirname` configuration, as the directory name of the append only files. All AOF files and
manifest file will be placed in this directory.
11. Only the last AOF file (BASE or INCR) can be truncated. Otherwise redis will exit even if
`aof-load-truncated` is enabled.
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-04 01:14:13 +08:00
|
|
|
void loadingAbsProgress(off_t pos);
|
|
|
|
void loadingIncrProgress(off_t size);
|
2019-10-29 17:59:09 +02:00
|
|
|
void stopLoading(int success);
|
2024-05-22 23:24:12 -07:00
|
|
|
void updateLoadingFileName(char *filename);
|
2019-10-29 17:59:09 +02:00
|
|
|
void startSaving(int rdbflags);
|
|
|
|
void stopSaving(int success);
|
2020-03-03 14:58:11 +01:00
|
|
|
int allPersistenceDisabled(void);
|
2010-11-08 11:52:03 +01:00
|
|
|
|
2024-05-22 23:24:12 -07:00
|
|
|
#define DISK_ERROR_TYPE_AOF 1 /* Don't accept writes: AOF errors. */
|
|
|
|
#define DISK_ERROR_TYPE_RDB 2 /* Don't accept writes: RDB errors. */
|
|
|
|
#define DISK_ERROR_TYPE_NONE 0 /* No problems, we can accept writes. */
|
2018-07-31 13:09:38 +02:00
|
|
|
int writeCommandsDeniedByDiskError(void);
|
Add new RM_Call flags for script mode, no writes, and error replies. (#10372)
The PR extends RM_Call with 3 new capabilities using new flags that
are given to RM_Call as part of the `fmt` argument.
It aims to assist modules that are getting a list of commands to be
executed from the user (not hard coded as part of the module logic),
think of a module that implements a new scripting language...
* `S` - Run the command in a script mode, this means that it will raise an
error if a command which are not allowed inside a script (flaged with the
`deny-script` flag) is invoked (like SHUTDOWN). In addition, on script mode,
write commands are not allowed if there is not enough good replicas (as
configured with `min-replicas-to-write`) and/or a disk error happened.
* `W` - no writes mode, Redis will reject any command that is marked with `write`
flag. Again can be useful to modules that implement a new scripting language
and wants to prevent any write commands.
* `E` - Return errors as RedisModuleCallReply. Today the errors that happened
before the command was invoked (like unknown commands or acl error) return
a NULL reply and set errno. This might be missing important information about
the failure and it is also impossible to just pass the error to the user using
RM_ReplyWithCallReply. This new flag allows you to get a RedisModuleCallReply
object with the relevant error message and treat it as if it was an error that was
raised by the command invocation.
Tests were added to verify the new code paths.
In addition small refactoring was done to share some code between modules,
scripts, and `processCommand` function:
1. `getAclErrorMessage` was added to `acl.c` to unified to log message extraction
from the acl result
2. `checkGoodReplicasStatus` was added to `replication.c` to check the status of
good replicas. It is used on `scriptVerifyWriteCommandAllow`, `RM_Call`, and
`processCommand`.
3. `writeCommandsGetDiskErrorMessage` was added to `server.c` to get the error
message on persistence failure. Again it is used on `scriptVerifyWriteCommandAllow`,
`RM_Call`, and `processCommand`.
2022-03-22 14:13:28 +02:00
|
|
|
sds writeCommandsGetDiskErrorMessage(int);
|
2018-07-31 13:09:38 +02:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* RDB persistence */
|
2011-05-13 17:31:00 +02:00
|
|
|
#include "rdb.h"
|
2019-01-21 11:28:44 +01:00
|
|
|
void killRDBChild(void);
|
2020-09-17 23:20:10 +08:00
|
|
|
int bg_unlink(const char *filename);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* AOF persistence */
|
2011-09-16 12:36:17 +02:00
|
|
|
void flushAppendOnlyFile(int force);
|
2021-05-29 23:20:32 -07:00
|
|
|
void feedAppendOnlyFile(int dictid, robj **argv, int argc);
|
2010-06-22 00:07:48 +02:00
|
|
|
void aofRemoveTempFile(pid_t childpid);
|
|
|
|
int rewriteAppendOnlyFileBackground(void);
|
Implement Multi Part AOF mechanism to avoid AOFRW overheads. (#9788)
Implement Multi-Part AOF mechanism to avoid overheads during AOFRW.
Introducing a folder with multiple AOF files tracked by a manifest file.
The main issues with the the original AOFRW mechanism are:
* buffering of commands that are processed during rewrite (consuming a lot of RAM)
* freezes of the main process when the AOFRW completes to drain the remaining part of the buffer and fsync it.
* double disk IO for the data that arrives during AOFRW (had to be written to both the old and new AOF files)
The main modifications of this PR:
1. Remove the AOF rewrite buffer and related code.
2. Divide the AOF into multiple files, they are classified as two types, one is the the `BASE` type,
it represents the full amount of data (Maybe AOF or RDB format) after each AOFRW, there is only
one `BASE` file at most. The second is `INCR` type, may have more than one. They represent the
incremental commands since the last AOFRW.
3. Use a AOF manifest file to record and manage these AOF files mentioned above.
4. The original configuration of `appendfilename` will be the base part of the new file name, for example:
`appendonly.aof.1.base.rdb` and `appendonly.aof.2.incr.aof`
5. Add manifest-related TCL tests, and modified some existing tests that depend on the `appendfilename`
6. Remove the `aof_rewrite_buffer_length` field in info.
7. Add `aof-disable-auto-gc` configuration. By default we're automatically deleting HISTORY type AOFs.
It also gives users the opportunity to preserve the history AOFs. just for testing use now.
8. Add AOFRW limiting measure. When the AOFRW failures reaches the threshold (3 times now),
we will delay the execution of the next AOFRW by 1 minute. If the next AOFRW also fails, it will be
delayed by 2 minutes. The next is 4, 8, 16, the maximum delay is 60 minutes (1 hour). During the limit
period, we can still use the 'bgrewriteaof' command to execute AOFRW immediately.
9. Support upgrade (load) data from old version redis.
10. Add `appenddirname` configuration, as the directory name of the append only files. All AOF files and
manifest file will be placed in this directory.
11. Only the last AOF file (BASE or INCR) can be truncated. Otherwise redis will exit even if
`aof-load-truncated` is enabled.
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-04 01:14:13 +08:00
|
|
|
int loadAppendOnlyFiles(aofManifest *am);
|
2010-06-22 00:07:48 +02:00
|
|
|
void stopAppendOnly(void);
|
|
|
|
int startAppendOnly(void);
|
2011-01-07 18:15:14 +01:00
|
|
|
void backgroundRewriteDoneHandler(int exitcode, int bysignal);
|
2019-01-21 11:28:44 +01:00
|
|
|
void killAppendOnlyChild(void);
|
2023-05-02 17:31:32 -07:00
|
|
|
void restartAOFAfterSYNC(void);
|
Implement Multi Part AOF mechanism to avoid AOFRW overheads. (#9788)
Implement Multi-Part AOF mechanism to avoid overheads during AOFRW.
Introducing a folder with multiple AOF files tracked by a manifest file.
The main issues with the the original AOFRW mechanism are:
* buffering of commands that are processed during rewrite (consuming a lot of RAM)
* freezes of the main process when the AOFRW completes to drain the remaining part of the buffer and fsync it.
* double disk IO for the data that arrives during AOFRW (had to be written to both the old and new AOF files)
The main modifications of this PR:
1. Remove the AOF rewrite buffer and related code.
2. Divide the AOF into multiple files, they are classified as two types, one is the the `BASE` type,
it represents the full amount of data (Maybe AOF or RDB format) after each AOFRW, there is only
one `BASE` file at most. The second is `INCR` type, may have more than one. They represent the
incremental commands since the last AOFRW.
3. Use a AOF manifest file to record and manage these AOF files mentioned above.
4. The original configuration of `appendfilename` will be the base part of the new file name, for example:
`appendonly.aof.1.base.rdb` and `appendonly.aof.2.incr.aof`
5. Add manifest-related TCL tests, and modified some existing tests that depend on the `appendfilename`
6. Remove the `aof_rewrite_buffer_length` field in info.
7. Add `aof-disable-auto-gc` configuration. By default we're automatically deleting HISTORY type AOFs.
It also gives users the opportunity to preserve the history AOFs. just for testing use now.
8. Add AOFRW limiting measure. When the AOFRW failures reaches the threshold (3 times now),
we will delay the execution of the next AOFRW by 1 minute. If the next AOFRW also fails, it will be
delayed by 2 minutes. The next is 4, 8, 16, the maximum delay is 60 minutes (1 hour). During the limit
period, we can still use the 'bgrewriteaof' command to execute AOFRW immediately.
9. Support upgrade (load) data from old version redis.
10. Add `appenddirname` configuration, as the directory name of the append only files. All AOF files and
manifest file will be placed in this directory.
11. Only the last AOF file (BASE or INCR) can be truncated. Otherwise redis will exit even if
`aof-load-truncated` is enabled.
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-04 01:14:13 +08:00
|
|
|
void aofLoadManifestFromDisk(void);
|
|
|
|
void aofOpenIfNeededOnServerStart(void);
|
|
|
|
void aofManifestFree(aofManifest *am);
|
|
|
|
int aofDelHistoryFiles(void);
|
|
|
|
int aofRewriteLimited(void);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2016-09-19 13:45:20 +02:00
|
|
|
/* Child info */
|
|
|
|
void openChildInfoPipe(void);
|
|
|
|
void closeChildInfoPipe(void);
|
2021-02-16 16:06:51 +02:00
|
|
|
void sendChildInfoGeneric(childInfoType info_type, size_t keys, double progress, char *pname);
|
|
|
|
void sendChildCowInfo(childInfoType info_type, char *pname);
|
|
|
|
void sendChildInfo(childInfoType info_type, size_t keys, char *pname);
|
2016-09-19 13:45:20 +02:00
|
|
|
void receiveChildInfo(void);
|
|
|
|
|
2019-07-17 08:51:02 +03:00
|
|
|
/* Fork helpers */
|
2024-04-04 01:26:33 +07:00
|
|
|
int serverFork(int purpose);
|
2023-05-02 17:31:32 -07:00
|
|
|
int hasActiveChildProcess(void);
|
|
|
|
void resetChildState(void);
|
Refactory fork child related infra, Unify child pid
This is a refactory commit, isn't suppose to have any actual impact.
it does the following:
- keep just one server struct fork child pid variable instead of 3
- have one server struct variable indicating the purpose of the current fork
child.
- redisFork is now responsible of updating the server struct with the pid,
which means it can be the one that calls updateDictResizePolicy
- move child info pipe handling into redisFork instead of having them
repeated outside
- there are two classes of fork purposes, mutually exclusive group (AOF, RDB,
Module), and one that can create several forks to coexist in parallel (LDB,
but maybe Modules some day too, Module API allows for that).
- minor fix to killRDBChild:
unlike killAppendOnlyChild and TerminateModuleForkChild, the killRDBChild
doesn't clear the pid variable or call wait4, so checkChildrenDone does
the cleanup for it.
This commit removes the explicit calls to rdbRemoveTempFile, closeChildInfoPipe,
updateDictResizePolicy, which didn't do any harm, but where unnecessary.
2020-12-16 15:14:04 +02:00
|
|
|
int isMutuallyExclusiveChildType(int type);
|
2019-07-17 08:51:02 +03:00
|
|
|
|
2018-12-21 17:16:22 +01:00
|
|
|
/* acl.c -- Authentication related prototypes. */
|
2019-02-05 10:48:17 +01:00
|
|
|
extern rax *Users;
|
2019-01-11 11:32:41 +01:00
|
|
|
extern user *DefaultUser;
|
2019-01-10 16:39:32 +01:00
|
|
|
void ACLInit(void);
|
2021-03-26 19:10:01 +08:00
|
|
|
/* Return values for ACLCheckAllPerm(). */
|
2019-01-16 18:31:05 +01:00
|
|
|
#define ACL_OK 0
|
|
|
|
#define ACL_DENIED_CMD 1
|
|
|
|
#define ACL_DENIED_KEY 2
|
2024-05-22 23:24:12 -07:00
|
|
|
#define ACL_DENIED_AUTH 3 /* Only used for ACL LOG entries. */
|
Adds pub/sub channel patterns to ACL (#7993)
Fixes #7923.
This PR appropriates the special `&` symbol (because `@` and `*` are taken),
followed by a literal value or pattern for describing the Pub/Sub patterns that
an ACL user can interact with. It is similar to the existing key patterns
mechanism in function (additive) and implementation (copy-pasta). It also adds
the allchannels and resetchannels ACL keywords, naturally.
The default user is given allchannels permissions, whereas new users get
whatever is defined by the acl-pubsub-default configuration directive. For
backward compatibility in 6.2, the default of this directive is allchannels but
this is likely to be changed to resetchannels in the next major version for
stronger default security settings.
Unless allchannels is set for the user, channel access permissions are checked
as follows :
* Calls to both PUBLISH and SUBSCRIBE will fail unless a pattern matching the
argumentative channel name(s) exists for the user.
* Calls to PSUBSCRIBE will fail unless the pattern(s) provided as an argument
literally exist(s) in the user's list.
Such failures are logged to the ACL log.
Runtime changes to channel permissions for a user with existing subscribing
clients cause said clients to disconnect unless the new permissions permit the
connections to continue. Note, however, that PSUBSCRIBErs' patterns are matched
literally, so given the change bar:* -> b*, pattern subscribers to bar:* will be
disconnected.
Notes/questions:
* UNSUBSCRIBE, PUNSUBSCRIBE and PUBSUB remain unprotected due to lack of reasons
for touching them.
2020-12-01 14:21:39 +02:00
|
|
|
#define ACL_DENIED_CHANNEL 4 /* Only used for pub/sub commands */
|
2021-09-23 08:52:56 +03:00
|
|
|
|
|
|
|
/* Context values for addACLLogEntry(). */
|
|
|
|
#define ACL_LOG_CTX_TOPLEVEL 0
|
|
|
|
#define ACL_LOG_CTX_LUA 1
|
|
|
|
#define ACL_LOG_CTX_MULTI 2
|
|
|
|
#define ACL_LOG_CTX_MODULE 3
|
|
|
|
|
2022-01-20 13:05:27 -08:00
|
|
|
/* ACL key permission types */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define ACL_READ_PERMISSION (1 << 0)
|
|
|
|
#define ACL_WRITE_PERMISSION (1 << 1)
|
|
|
|
#define ACL_ALL_PERMISSION (ACL_READ_PERMISSION | ACL_WRITE_PERMISSION)
|
2022-01-20 13:05:27 -08:00
|
|
|
|
Custom authentication for Modules (#11659)
This change adds new module callbacks that can override the default password based authentication associated with ACLs. With this, Modules can register auth callbacks through which they can implement their own Authentication logic. When `AUTH` and `HELLO AUTH ...` commands are used, Module based authentication is attempted and then normal password based authentication is attempted if needed.
The new Module APIs added in this PR are - `RM_RegisterCustomAuthCallback` and `RM_BlockClientOnAuth` and `RedisModule_ACLAddLogEntryByUserName `.
Module based authentication will be attempted for all Redis users (created through the ACL SETUSER cmd or through Module APIs) even if the Redis user does not exist at the time of the command. This gives a chance for the Module to create the RedisModule user and then authenticate via the RedisModule API - from the custom auth callback.
For the AUTH command, we will support both variations - `AUTH <username> <password>` and `AUTH <password>`. In case of the `AUTH <password>` variation, the custom auth callbacks are triggered with “default” as the username and password as what is provided.
### RedisModule_RegisterCustomAuthCallback
```
void RM_RegisterCustomAuthCallback(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback cb) {
```
This API registers a callback to execute to prior to normal password based authentication. Multiple callbacks can be registered across different modules. These callbacks are responsible for either handling the authentication, each authenticating the user or explicitly denying, or deferring it to other authentication mechanisms. Callbacks are triggered in the order they were registered. When a Module is unloaded, all the auth callbacks registered by it are unregistered. The callbacks are attempted, in the order of most recently registered callbacks, when the AUTH/HELLO (with AUTH field is provided) commands are called. The callbacks will be called with a module context along with a username and a password, and are expected to take one of the following actions:
(1) Authenticate - Use the RM_Authenticate* API successfully and return `REDISMODULE_AUTH_HANDLED`. This will immediately end the auth chain as successful and add the OK reply.
(2) Block a client on authentication - Use the `RM_BlockClientOnAuth` API and return `REDISMODULE_AUTH_HANDLED`. Here, the client will be blocked until the `RM_UnblockClient `API is used which will trigger the auth reply callback (provided earlier through the `RM_BlockClientOnAuth`). In this reply callback, the Module should authenticate, deny or skip handling authentication.
(3) Deny Authentication - Return `REDISMODULE_AUTH_HANDLED` without authenticating or blocking the client. Optionally, `err` can be set to a custom error message. This will immediately end the auth chain as unsuccessful and add the ERR reply.
(4) Skip handling Authentication - Return `REDISMODULE_AUTH_NOT_HANDLED` without blocking the client. This will allow the engine to attempt the next custom auth callback.
If none of the callbacks authenticate or deny auth, then password based auth is attempted and will authenticate or add failure logs and reply to the clients accordingly.
### RedisModule_BlockClientOnAuth
```
RedisModuleBlockedClient *RM_BlockClientOnAuth(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback reply_callback,
void (*free_privdata)(RedisModuleCtx*,void*))
```
This API can only be used from a Module from the custom auth callback. If a client is not in the middle of custom module based authentication, ERROR is returned. Otherwise, the client is blocked and the `RedisModule_BlockedClient` is returned similar to the `RedisModule_BlockClient` API.
### RedisModule_ACLAddLogEntryByUserName
```
int RM_ACLAddLogEntryByUserName(RedisModuleCtx *ctx, RedisModuleString *username, RedisModuleString *object, RedisModuleACLLogEntryReason reason)
```
Adds a new entry in the ACL log with the `username` RedisModuleString provided. This simplifies the Module usage because now, developers do not need to create a Module User just to add an error ACL Log entry. Aside from accepting username (RedisModuleString) instead of a RedisModuleUser, it is the same as the existing `RedisModule_ACLAddLogEntry` API.
### Breaking changes
- HELLO command - Clients can now only set the client name and RESP protocol from the `HELLO` command if they are authenticated. Also, we now finish command arg validation first and return early with a ERR reply if any arg is invalid. This is to avoid mutating the client name / RESP from a command that would have failed on invalid arguments.
### Notable behaviors
- Module unblocking - Now, we will not allow Modules to block the client from inside the context of a reply callback (triggered from the Module unblock flow `moduleHandleBlockedClients`).
---------
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
2023-03-15 15:18:42 -07:00
|
|
|
/* Return codes for Authentication functions to indicate the result. */
|
2024-05-22 23:24:12 -07:00
|
|
|
typedef enum { AUTH_OK = 0, AUTH_ERR, AUTH_NOT_HANDLED, AUTH_BLOCKED } AuthResult;
|
Custom authentication for Modules (#11659)
This change adds new module callbacks that can override the default password based authentication associated with ACLs. With this, Modules can register auth callbacks through which they can implement their own Authentication logic. When `AUTH` and `HELLO AUTH ...` commands are used, Module based authentication is attempted and then normal password based authentication is attempted if needed.
The new Module APIs added in this PR are - `RM_RegisterCustomAuthCallback` and `RM_BlockClientOnAuth` and `RedisModule_ACLAddLogEntryByUserName `.
Module based authentication will be attempted for all Redis users (created through the ACL SETUSER cmd or through Module APIs) even if the Redis user does not exist at the time of the command. This gives a chance for the Module to create the RedisModule user and then authenticate via the RedisModule API - from the custom auth callback.
For the AUTH command, we will support both variations - `AUTH <username> <password>` and `AUTH <password>`. In case of the `AUTH <password>` variation, the custom auth callbacks are triggered with “default” as the username and password as what is provided.
### RedisModule_RegisterCustomAuthCallback
```
void RM_RegisterCustomAuthCallback(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback cb) {
```
This API registers a callback to execute to prior to normal password based authentication. Multiple callbacks can be registered across different modules. These callbacks are responsible for either handling the authentication, each authenticating the user or explicitly denying, or deferring it to other authentication mechanisms. Callbacks are triggered in the order they were registered. When a Module is unloaded, all the auth callbacks registered by it are unregistered. The callbacks are attempted, in the order of most recently registered callbacks, when the AUTH/HELLO (with AUTH field is provided) commands are called. The callbacks will be called with a module context along with a username and a password, and are expected to take one of the following actions:
(1) Authenticate - Use the RM_Authenticate* API successfully and return `REDISMODULE_AUTH_HANDLED`. This will immediately end the auth chain as successful and add the OK reply.
(2) Block a client on authentication - Use the `RM_BlockClientOnAuth` API and return `REDISMODULE_AUTH_HANDLED`. Here, the client will be blocked until the `RM_UnblockClient `API is used which will trigger the auth reply callback (provided earlier through the `RM_BlockClientOnAuth`). In this reply callback, the Module should authenticate, deny or skip handling authentication.
(3) Deny Authentication - Return `REDISMODULE_AUTH_HANDLED` without authenticating or blocking the client. Optionally, `err` can be set to a custom error message. This will immediately end the auth chain as unsuccessful and add the ERR reply.
(4) Skip handling Authentication - Return `REDISMODULE_AUTH_NOT_HANDLED` without blocking the client. This will allow the engine to attempt the next custom auth callback.
If none of the callbacks authenticate or deny auth, then password based auth is attempted and will authenticate or add failure logs and reply to the clients accordingly.
### RedisModule_BlockClientOnAuth
```
RedisModuleBlockedClient *RM_BlockClientOnAuth(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback reply_callback,
void (*free_privdata)(RedisModuleCtx*,void*))
```
This API can only be used from a Module from the custom auth callback. If a client is not in the middle of custom module based authentication, ERROR is returned. Otherwise, the client is blocked and the `RedisModule_BlockedClient` is returned similar to the `RedisModule_BlockClient` API.
### RedisModule_ACLAddLogEntryByUserName
```
int RM_ACLAddLogEntryByUserName(RedisModuleCtx *ctx, RedisModuleString *username, RedisModuleString *object, RedisModuleACLLogEntryReason reason)
```
Adds a new entry in the ACL log with the `username` RedisModuleString provided. This simplifies the Module usage because now, developers do not need to create a Module User just to add an error ACL Log entry. Aside from accepting username (RedisModuleString) instead of a RedisModuleUser, it is the same as the existing `RedisModule_ACLAddLogEntry` API.
### Breaking changes
- HELLO command - Clients can now only set the client name and RESP protocol from the `HELLO` command if they are authenticated. Also, we now finish command arg validation first and return early with a ERR reply if any arg is invalid. This is to avoid mutating the client name / RESP from a command that would have failed on invalid arguments.
### Notable behaviors
- Module unblocking - Now, we will not allow Modules to block the client from inside the context of a reply callback (triggered from the Module unblock flow `moduleHandleBlockedClients`).
---------
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
2023-03-15 15:18:42 -07:00
|
|
|
|
2018-12-21 17:16:22 +01:00
|
|
|
int ACLCheckUserCredentials(robj *username, robj *password);
|
Custom authentication for Modules (#11659)
This change adds new module callbacks that can override the default password based authentication associated with ACLs. With this, Modules can register auth callbacks through which they can implement their own Authentication logic. When `AUTH` and `HELLO AUTH ...` commands are used, Module based authentication is attempted and then normal password based authentication is attempted if needed.
The new Module APIs added in this PR are - `RM_RegisterCustomAuthCallback` and `RM_BlockClientOnAuth` and `RedisModule_ACLAddLogEntryByUserName `.
Module based authentication will be attempted for all Redis users (created through the ACL SETUSER cmd or through Module APIs) even if the Redis user does not exist at the time of the command. This gives a chance for the Module to create the RedisModule user and then authenticate via the RedisModule API - from the custom auth callback.
For the AUTH command, we will support both variations - `AUTH <username> <password>` and `AUTH <password>`. In case of the `AUTH <password>` variation, the custom auth callbacks are triggered with “default” as the username and password as what is provided.
### RedisModule_RegisterCustomAuthCallback
```
void RM_RegisterCustomAuthCallback(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback cb) {
```
This API registers a callback to execute to prior to normal password based authentication. Multiple callbacks can be registered across different modules. These callbacks are responsible for either handling the authentication, each authenticating the user or explicitly denying, or deferring it to other authentication mechanisms. Callbacks are triggered in the order they were registered. When a Module is unloaded, all the auth callbacks registered by it are unregistered. The callbacks are attempted, in the order of most recently registered callbacks, when the AUTH/HELLO (with AUTH field is provided) commands are called. The callbacks will be called with a module context along with a username and a password, and are expected to take one of the following actions:
(1) Authenticate - Use the RM_Authenticate* API successfully and return `REDISMODULE_AUTH_HANDLED`. This will immediately end the auth chain as successful and add the OK reply.
(2) Block a client on authentication - Use the `RM_BlockClientOnAuth` API and return `REDISMODULE_AUTH_HANDLED`. Here, the client will be blocked until the `RM_UnblockClient `API is used which will trigger the auth reply callback (provided earlier through the `RM_BlockClientOnAuth`). In this reply callback, the Module should authenticate, deny or skip handling authentication.
(3) Deny Authentication - Return `REDISMODULE_AUTH_HANDLED` without authenticating or blocking the client. Optionally, `err` can be set to a custom error message. This will immediately end the auth chain as unsuccessful and add the ERR reply.
(4) Skip handling Authentication - Return `REDISMODULE_AUTH_NOT_HANDLED` without blocking the client. This will allow the engine to attempt the next custom auth callback.
If none of the callbacks authenticate or deny auth, then password based auth is attempted and will authenticate or add failure logs and reply to the clients accordingly.
### RedisModule_BlockClientOnAuth
```
RedisModuleBlockedClient *RM_BlockClientOnAuth(RedisModuleCtx *ctx, RedisModuleCustomAuthCallback reply_callback,
void (*free_privdata)(RedisModuleCtx*,void*))
```
This API can only be used from a Module from the custom auth callback. If a client is not in the middle of custom module based authentication, ERROR is returned. Otherwise, the client is blocked and the `RedisModule_BlockedClient` is returned similar to the `RedisModule_BlockClient` API.
### RedisModule_ACLAddLogEntryByUserName
```
int RM_ACLAddLogEntryByUserName(RedisModuleCtx *ctx, RedisModuleString *username, RedisModuleString *object, RedisModuleACLLogEntryReason reason)
```
Adds a new entry in the ACL log with the `username` RedisModuleString provided. This simplifies the Module usage because now, developers do not need to create a Module User just to add an error ACL Log entry. Aside from accepting username (RedisModuleString) instead of a RedisModuleUser, it is the same as the existing `RedisModule_ACLAddLogEntry` API.
### Breaking changes
- HELLO command - Clients can now only set the client name and RESP protocol from the `HELLO` command if they are authenticated. Also, we now finish command arg validation first and return early with a ERR reply if any arg is invalid. This is to avoid mutating the client name / RESP from a command that would have failed on invalid arguments.
### Notable behaviors
- Module unblocking - Now, we will not allow Modules to block the client from inside the context of a reply callback (triggered from the Module unblock flow `moduleHandleBlockedClients`).
---------
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
2023-03-15 15:18:42 -07:00
|
|
|
int ACLAuthenticateUser(client *c, robj *username, robj *password, robj **err);
|
|
|
|
int checkModuleAuthentication(client *c, robj *username, robj *password, robj **err);
|
|
|
|
void addAuthErrReply(client *c, robj *err);
|
2022-01-23 16:05:06 +08:00
|
|
|
unsigned long ACLGetCommandID(sds cmdname);
|
2020-10-19 00:33:55 -04:00
|
|
|
void ACLClearCommandID(void);
|
2019-01-10 16:33:48 +01:00
|
|
|
user *ACLGetUserByName(const char *name, size_t namelen);
|
2022-01-20 13:05:27 -08:00
|
|
|
int ACLUserCheckKeyPerm(user *u, const char *key, int keylen, int flags);
|
|
|
|
int ACLUserCheckChannelPerm(user *u, sds channel, int literal);
|
2024-04-03 23:54:33 +07:00
|
|
|
int ACLCheckAllUserCommandPerm(user *u, struct serverCommand *cmd, robj **argv, int argc, int *idxptr);
|
|
|
|
int ACLUserCheckCmdWithUnrestrictedKeyAccess(user *u, struct serverCommand *cmd, robj **argv, int argc, int flags);
|
2021-03-26 19:10:01 +08:00
|
|
|
int ACLCheckAllPerm(client *c, int *idxptr);
|
2019-01-17 18:05:43 +01:00
|
|
|
int ACLSetUser(user *u, const char *op, ssize_t oplen);
|
2022-09-22 16:29:00 +03:00
|
|
|
sds ACLStringSetUser(user *u, sds username, sds *argv, int argc);
|
2019-01-23 16:59:09 +01:00
|
|
|
uint64_t ACLGetCommandCategoryFlagByName(const char *name);
|
2023-08-30 13:01:24 -07:00
|
|
|
int ACLAddCommandCategory(const char *name, uint64_t flag);
|
|
|
|
void ACLCleanupCategoriesOnFailure(size_t num_acl_categories_added);
|
2019-02-04 13:00:38 +01:00
|
|
|
int ACLAppendUserForLoading(sds *argv, int argc, int *argc_err);
|
2021-01-21 11:56:08 +02:00
|
|
|
const char *ACLSetUserStringError(void);
|
2019-02-04 16:35:15 +01:00
|
|
|
int ACLLoadConfiguredUsers(void);
|
2022-09-22 16:29:00 +03:00
|
|
|
robj *ACLDescribeUser(user *u);
|
2019-02-07 17:20:03 +01:00
|
|
|
void ACLLoadUsersAtStartup(void);
|
2024-04-03 23:54:33 +07:00
|
|
|
void addReplyCommandCategories(client *c, struct serverCommand *cmd);
|
2023-05-02 17:31:32 -07:00
|
|
|
user *ACLCreateUnlinkedUser(void);
|
2019-02-26 01:23:11 +00:00
|
|
|
void ACLFreeUserAndKillClients(user *u);
|
2021-09-23 08:52:56 +03:00
|
|
|
void addACLLogEntry(client *c, int reason, int context, int argpos, sds username, sds object);
|
2024-04-03 23:54:33 +07:00
|
|
|
sds getAclErrorMessage(int acl_res, user *user, struct serverCommand *cmd, sds errored_val, int verbose);
|
2021-02-25 21:00:27 -08:00
|
|
|
void ACLUpdateDefaultUserPassword(sds password);
|
2024-05-08 09:44:05 -04:00
|
|
|
sds genValkeyInfoStringACLStats(sds info);
|
2023-05-02 17:31:32 -07:00
|
|
|
void ACLRecomputeCommandBitsFromCommandRulesAllUsers(void);
|
2018-12-21 17:16:22 +01:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Sorted sets data type */
|
2011-04-28 19:00:33 +02:00
|
|
|
|
2016-04-14 15:58:49 +02:00
|
|
|
/* Input flags. */
|
2021-03-10 15:09:43 +01:00
|
|
|
#define ZADD_IN_NONE 0
|
2024-05-22 23:24:12 -07:00
|
|
|
#define ZADD_IN_INCR (1 << 0) /* Increment the score instead of setting it. */
|
|
|
|
#define ZADD_IN_NX (1 << 1) /* Don't touch elements not already existing. */
|
|
|
|
#define ZADD_IN_XX (1 << 2) /* Only touch elements already existing. */
|
|
|
|
#define ZADD_IN_GT (1 << 3) /* Only update existing when new scores are higher. */
|
|
|
|
#define ZADD_IN_LT (1 << 4) /* Only update existing when new scores are lower. */
|
2016-04-14 15:58:49 +02:00
|
|
|
|
|
|
|
/* Output flags. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define ZADD_OUT_NOP (1 << 0) /* Operation not performed because of conditionals.*/
|
|
|
|
#define ZADD_OUT_NAN (1 << 1) /* Only touch elements already existing. */
|
|
|
|
#define ZADD_OUT_ADDED (1 << 2) /* The element was new and was added. */
|
|
|
|
#define ZADD_OUT_UPDATED (1 << 3) /* The element already existed, score updated. */
|
2016-04-14 15:58:49 +02:00
|
|
|
|
Squash merging 125 typo/grammar/comment/doc PRs (#7773)
List of squashed commits or PRs
===============================
commit 66801ea
Author: hwware <wen.hui.ware@gmail.com>
Date: Mon Jan 13 00:54:31 2020 -0500
typo fix in acl.c
commit 46f55db
Author: Itamar Haber <itamar@redislabs.com>
Date: Sun Sep 6 18:24:11 2020 +0300
Updates a couple of comments
Specifically:
* RM_AutoMemory completed instead of pointing to docs
* Updated link to custom type doc
commit 61a2aa0
Author: xindoo <xindoo@qq.com>
Date: Tue Sep 1 19:24:59 2020 +0800
Correct errors in code comments
commit a5871d1
Author: yz1509 <pro-756@qq.com>
Date: Tue Sep 1 18:36:06 2020 +0800
fix typos in module.c
commit 41eede7
Author: bookug <bookug@qq.com>
Date: Sat Aug 15 01:11:33 2020 +0800
docs: fix typos in comments
commit c303c84
Author: lazy-snail <ws.niu@outlook.com>
Date: Fri Aug 7 11:15:44 2020 +0800
fix spelling in redis.conf
commit 1eb76bf
Author: zhujian <zhujianxyz@gmail.com>
Date: Thu Aug 6 15:22:10 2020 +0800
add a missing 'n' in comment
commit 1530ec2
Author: Daniel Dai <764122422@qq.com>
Date: Mon Jul 27 00:46:35 2020 -0400
fix spelling in tracking.c
commit e517b31
Author: Hunter-Chen <huntcool001@gmail.com>
Date: Fri Jul 17 22:33:32 2020 +0800
Update redis.conf
Co-authored-by: Itamar Haber <itamar@redislabs.com>
commit c300eff
Author: Hunter-Chen <huntcool001@gmail.com>
Date: Fri Jul 17 22:33:23 2020 +0800
Update redis.conf
Co-authored-by: Itamar Haber <itamar@redislabs.com>
commit 4c058a8
Author: 陈浩鹏 <chenhaopeng@heytea.com>
Date: Thu Jun 25 19:00:56 2020 +0800
Grammar fix and clarification
commit 5fcaa81
Author: bodong.ybd <bodong.ybd@alibaba-inc.com>
Date: Fri Jun 19 10:09:00 2020 +0800
Fix typos
commit 4caca9a
Author: Pruthvi P <pruthvi@ixigo.com>
Date: Fri May 22 00:33:22 2020 +0530
Fix typo eviciton => eviction
commit b2a25f6
Author: Brad Dunbar <dunbarb2@gmail.com>
Date: Sun May 17 12:39:59 2020 -0400
Fix a typo.
commit 12842ae
Author: hwware <wen.hui.ware@gmail.com>
Date: Sun May 3 17:16:59 2020 -0400
fix spelling in redis conf
commit ddba07c
Author: Chris Lamb <chris@chris-lamb.co.uk>
Date: Sat May 2 23:25:34 2020 +0100
Correct a "conflicts" spelling error.
commit 8fc7bf2
Author: Nao YONASHIRO <yonashiro@r.recruit.co.jp>
Date: Thu Apr 30 10:25:27 2020 +0900
docs: fix EXPIRE_FAST_CYCLE_DURATION to ACTIVE_EXPIRE_CYCLE_FAST_DURATION
commit 9b2b67a
Author: Brad Dunbar <dunbarb2@gmail.com>
Date: Fri Apr 24 11:46:22 2020 -0400
Fix a typo.
commit 0746f10
Author: devilinrust <63737265+devilinrust@users.noreply.github.com>
Date: Thu Apr 16 00:17:53 2020 +0200
Fix typos in server.c
commit 92b588d
Author: benjessop12 <56115861+benjessop12@users.noreply.github.com>
Date: Mon Apr 13 13:43:55 2020 +0100
Fix spelling mistake in lazyfree.c
commit 1da37aa
Merge: 2d4ba28 af347a8
Author: hwware <wen.hui.ware@gmail.com>
Date: Thu Mar 5 22:41:31 2020 -0500
Merge remote-tracking branch 'upstream/unstable' into expiretypofix
commit 2d4ba28
Author: hwware <wen.hui.ware@gmail.com>
Date: Mon Mar 2 00:09:40 2020 -0500
fix typo in expire.c
commit 1a746f7
Author: SennoYuki <minakami1yuki@gmail.com>
Date: Thu Feb 27 16:54:32 2020 +0800
fix typo
commit 8599b1a
Author: dongheejeong <donghee950403@gmail.com>
Date: Sun Feb 16 20:31:43 2020 +0000
Fix typo in server.c
commit f38d4e8
Author: hwware <wen.hui.ware@gmail.com>
Date: Sun Feb 2 22:58:38 2020 -0500
fix typo in evict.c
commit fe143fc
Author: Leo Murillo <leonardo.murillo@gmail.com>
Date: Sun Feb 2 01:57:22 2020 -0600
Fix a few typos in redis.conf
commit 1ab4d21
Author: viraja1 <anchan.viraj@gmail.com>
Date: Fri Dec 27 17:15:58 2019 +0530
Fix typo in Latency API docstring
commit ca1f70e
Author: gosth <danxuedexing@qq.com>
Date: Wed Dec 18 15:18:02 2019 +0800
fix typo in sort.c
commit a57c06b
Author: ZYunH <zyunhjob@163.com>
Date: Mon Dec 16 22:28:46 2019 +0800
fix-zset-typo
commit b8c92b5
Author: git-hulk <hulk.website@gmail.com>
Date: Mon Dec 16 15:51:42 2019 +0800
FIX: typo in cluster.c, onformation->information
commit 9dd981c
Author: wujm2007 <jim.wujm@gmail.com>
Date: Mon Dec 16 09:37:52 2019 +0800
Fix typo
commit e132d7a
Author: Sebastien Williams-Wynn <s.williamswynn.mail@gmail.com>
Date: Fri Nov 15 00:14:07 2019 +0000
Minor typo change
commit 47f44d5
Author: happynote3966 <01ssrmikururudevice01@gmail.com>
Date: Mon Nov 11 22:08:48 2019 +0900
fix comment typo in redis-cli.c
commit b8bdb0d
Author: fulei <fulei@kuaishou.com>
Date: Wed Oct 16 18:00:17 2019 +0800
Fix a spelling mistake of comments in defragDictBucketCallback
commit 0def46a
Author: fulei <fulei@kuaishou.com>
Date: Wed Oct 16 13:09:27 2019 +0800
fix some spelling mistakes of comments in defrag.c
commit f3596fd
Author: Phil Rajchgot <tophil@outlook.com>
Date: Sun Oct 13 02:02:32 2019 -0400
Typo and grammar fixes
Redis and its documentation are great -- just wanted to submit a few corrections in the spirit of Hacktoberfest. Thanks for all your work on this project. I use it all the time and it works beautifully.
commit 2b928cd
Author: KangZhiDong <worldkzd@gmail.com>
Date: Sun Sep 1 07:03:11 2019 +0800
fix typos
commit 33aea14
Author: Axlgrep <axlgrep@gmail.com>
Date: Tue Aug 27 11:02:18 2019 +0800
Fixed eviction spelling issues
commit e282a80
Author: Simen Flatby <simen@oms.no>
Date: Tue Aug 20 15:25:51 2019 +0200
Update comments to reflect prop name
In the comments the prop is referenced as replica-validity-factor,
but it is really named cluster-replica-validity-factor.
commit 74d1f9a
Author: Jim Green <jimgreen2013@qq.com>
Date: Tue Aug 20 20:00:31 2019 +0800
fix comment error, the code is ok
commit eea1407
Author: Liao Tonglang <liaotonglang@gmail.com>
Date: Fri May 31 10:16:18 2019 +0800
typo fix
fix cna't to can't
commit 0da553c
Author: KAWACHI Takashi <tkawachi@gmail.com>
Date: Wed Jul 17 00:38:16 2019 +0900
Fix typo
commit 7fc8fb6
Author: Michael Prokop <mika@grml.org>
Date: Tue May 28 17:58:42 2019 +0200
Typo fixes
s/familar/familiar/
s/compatiblity/compatibility/
s/ ot / to /
s/itsef/itself/
commit 5f46c9d
Author: zhumoing <34539422+zhumoing@users.noreply.github.com>
Date: Tue May 21 21:16:50 2019 +0800
typo-fixes
typo-fixes
commit 321dfe1
Author: wxisme <850885154@qq.com>
Date: Sat Mar 16 15:10:55 2019 +0800
typo fix
commit b4fb131
Merge: 267e0e6 3df1eb8
Author: Nikitas Bastas <nikitasbst@gmail.com>
Date: Fri Feb 8 22:55:45 2019 +0200
Merge branch 'unstable' of antirez/redis into unstable
commit 267e0e6
Author: Nikitas Bastas <nikitasbst@gmail.com>
Date: Wed Jan 30 21:26:04 2019 +0200
Minor typo fix
commit 30544e7
Author: inshal96 <39904558+inshal96@users.noreply.github.com>
Date: Fri Jan 4 16:54:50 2019 +0500
remove an extra 'a' in the comments
commit 337969d
Author: BrotherGao <yangdongheng11@gmail.com>
Date: Sat Dec 29 12:37:29 2018 +0800
fix typo in redis.conf
commit 9f4b121
Merge: 423a030 e504583
Author: BrotherGao <yangdongheng@xiaomi.com>
Date: Sat Dec 29 11:41:12 2018 +0800
Merge branch 'unstable' of antirez/redis into unstable
commit 423a030
Merge: 42b02b7 46a51cd
Author: 杨东衡 <yangdongheng@xiaomi.com>
Date: Tue Dec 4 23:56:11 2018 +0800
Merge branch 'unstable' of antirez/redis into unstable
commit 42b02b7
Merge: 68c0e6e b8febe6
Author: Dongheng Yang <yangdongheng11@gmail.com>
Date: Sun Oct 28 15:54:23 2018 +0800
Merge pull request #1 from antirez/unstable
update local data
commit 714b589
Author: Christian <crifei93@gmail.com>
Date: Fri Dec 28 01:17:26 2018 +0100
fix typo "resulution"
commit e23259d
Author: garenchan <1412950785@qq.com>
Date: Wed Dec 26 09:58:35 2018 +0800
fix typo: segfauls -> segfault
commit a9359f8
Author: xjp <jianping_xie@aliyun.com>
Date: Tue Dec 18 17:31:44 2018 +0800
Fixed REDISMODULE_H spell bug
commit a12c3e4
Author: jdiaz <jrd.palacios@gmail.com>
Date: Sat Dec 15 23:39:52 2018 -0600
Fixes hyperloglog hash function comment block description
commit 770eb11
Author: 林上耀 <1210tom@163.com>
Date: Sun Nov 25 17:16:10 2018 +0800
fix typo
commit fd97fbb
Author: Chris Lamb <chris@chris-lamb.co.uk>
Date: Fri Nov 23 17:14:01 2018 +0100
Correct "unsupported" typo.
commit a85522d
Author: Jungnam Lee <jungnam.lee@oracle.com>
Date: Thu Nov 8 23:01:29 2018 +0900
fix typo in test comments
commit ade8007
Author: Arun Kumar <palerdot@users.noreply.github.com>
Date: Tue Oct 23 16:56:35 2018 +0530
Fixed grammatical typo
Fixed typo for word 'dictionary'
commit 869ee39
Author: Hamid Alaei <hamid.a85@gmail.com>
Date: Sun Aug 12 16:40:02 2018 +0430
fix documentations: (ThreadSafeContextStart/Stop -> ThreadSafeContextLock/Unlock), minor typo
commit f89d158
Author: Mayank Jain <mayankjain255@gmail.com>
Date: Tue Jul 31 23:01:21 2018 +0530
Updated README.md with some spelling corrections.
Made correction in spelling of some misspelled words.
commit 892198e
Author: dsomeshwar <someshwar.dhayalan@gmail.com>
Date: Sat Jul 21 23:23:04 2018 +0530
typo fix
commit 8a4d780
Author: Itamar Haber <itamar@redislabs.com>
Date: Mon Apr 30 02:06:52 2018 +0300
Fixes some typos
commit e3acef6
Author: Noah Rosamilia <ivoahivoah@gmail.com>
Date: Sat Mar 3 23:41:21 2018 -0500
Fix typo in /deps/README.md
commit 04442fb
Author: WuYunlong <xzsyeb@126.com>
Date: Sat Mar 3 10:32:42 2018 +0800
Fix typo in readSyncBulkPayload() comment.
commit 9f36880
Author: WuYunlong <xzsyeb@126.com>
Date: Sat Mar 3 10:20:37 2018 +0800
replication.c comment: run_id -> replid.
commit f866b4a
Author: Francesco 'makevoid' Canessa <makevoid@gmail.com>
Date: Thu Feb 22 22:01:56 2018 +0000
fix comment typo in server.c
commit 0ebc69b
Author: 줍 <jubee0124@gmail.com>
Date: Mon Feb 12 16:38:48 2018 +0900
Fix typo in redis.conf
Fix `five behaviors` to `eight behaviors` in [this sentence ](antirez/redis@unstable/redis.conf#L564)
commit b50a620
Author: martinbroadhurst <martinbroadhurst@users.noreply.github.com>
Date: Thu Dec 28 12:07:30 2017 +0000
Fix typo in valgrind.sup
commit 7d8f349
Author: Peter Boughton <peter@sorcerersisle.com>
Date: Mon Nov 27 19:52:19 2017 +0000
Update CONTRIBUTING; refer doc updates to redis-doc repo.
commit 02dec7e
Author: Klauswk <klauswk1@hotmail.com>
Date: Tue Oct 24 16:18:38 2017 -0200
Fix typo in comment
commit e1efbc8
Author: chenshi <baiwfg2@gmail.com>
Date: Tue Oct 3 18:26:30 2017 +0800
Correct two spelling errors of comments
commit 93327d8
Author: spacewander <spacewanderlzx@gmail.com>
Date: Wed Sep 13 16:47:24 2017 +0800
Update the comment for OBJ_ENCODING_EMBSTR_SIZE_LIMIT's value
The value of OBJ_ENCODING_EMBSTR_SIZE_LIMIT is 44 now instead of 39.
commit 63d361f
Author: spacewander <spacewanderlzx@gmail.com>
Date: Tue Sep 12 15:06:42 2017 +0800
Fix <prevlen> related doc in ziplist.c
According to the definition of ZIP_BIG_PREVLEN and other related code,
the guard of single byte <prevlen> should be 254 instead of 255.
commit ebe228d
Author: hanael80 <hanael80@gmail.com>
Date: Tue Aug 15 09:09:40 2017 +0900
Fix typo
commit 6b696e6
Author: Matt Robenolt <matt@ydekproductions.com>
Date: Mon Aug 14 14:50:47 2017 -0700
Fix typo in LATENCY DOCTOR output
commit a2ec6ae
Author: caosiyang <caosiyang@qiyi.com>
Date: Tue Aug 15 14:15:16 2017 +0800
Fix a typo: form => from
commit 3ab7699
Author: caosiyang <caosiyang@qiyi.com>
Date: Thu Aug 10 18:40:33 2017 +0800
Fix a typo: replicationFeedSlavesFromMaster() => replicationFeedSlavesFromMasterStream()
commit 72d43ef
Author: caosiyang <caosiyang@qiyi.com>
Date: Tue Aug 8 15:57:25 2017 +0800
fix a typo: servewr => server
commit 707c958
Author: Bo Cai <charpty@gmail.com>
Date: Wed Jul 26 21:49:42 2017 +0800
redis-cli.c typo: conut -> count.
Signed-off-by: Bo Cai <charpty@gmail.com>
commit b9385b2
Author: JackDrogon <jack.xsuperman@gmail.com>
Date: Fri Jun 30 14:22:31 2017 +0800
Fix some spell problems
commit 20d9230
Author: akosel <aaronjkosel@gmail.com>
Date: Sun Jun 4 19:35:13 2017 -0500
Fix typo
commit b167bfc
Author: Krzysiek Witkowicz <krzysiekwitkowicz@gmail.com>
Date: Mon May 22 21:32:27 2017 +0100
Fix #4008 small typo in comment
commit 2b78ac8
Author: Jake Clarkson <jacobwclarkson@gmail.com>
Date: Wed Apr 26 15:49:50 2017 +0100
Correct typo in tests/unit/hyperloglog.tcl
commit b0f1cdb
Author: Qi Luo <qiluo-msft@users.noreply.github.com>
Date: Wed Apr 19 14:25:18 2017 -0700
Fix typo
commit a90b0f9
Author: charsyam <charsyam@naver.com>
Date: Thu Mar 16 18:19:53 2017 +0900
fix typos
fix typos
fix typos
commit 8430a79
Author: Richard Hart <richardhart92@gmail.com>
Date: Mon Mar 13 22:17:41 2017 -0400
Fixed log message typo in listenToPort.
commit 481a1c2
Author: Vinod Kumar <kumar003vinod@gmail.com>
Date: Sun Jan 15 23:04:51 2017 +0530
src/db.c: Correct "save" -> "safe" typo
commit 586b4d3
Author: wangshaonan <wshn13@gmail.com>
Date: Wed Dec 21 20:28:27 2016 +0800
Fix typo they->the in helloworld.c
commit c1c4b5e
Author: Jenner <hypxm@qq.com>
Date: Mon Dec 19 16:39:46 2016 +0800
typo error
commit 1ee1a3f
Author: tielei <43289893@qq.com>
Date: Mon Jul 18 13:52:25 2016 +0800
fix some comments
commit 11a41fb
Author: Otto Kekäläinen <otto@seravo.fi>
Date: Sun Jul 3 10:23:55 2016 +0100
Fix spelling in documentation and comments
commit 5fb5d82
Author: francischan <f1ancis621@gmail.com>
Date: Tue Jun 28 00:19:33 2016 +0800
Fix outdated comments about redis.c file.
It should now refer to server.c file.
commit 6b254bc
Author: lmatt-bit <lmatt123n@gmail.com>
Date: Thu Apr 21 21:45:58 2016 +0800
Refine the comment of dictRehashMilliseconds func
SLAVECONF->REPLCONF in comment - by andyli029
commit ee9869f
Author: clark.kang <charsyam@naver.com>
Date: Tue Mar 22 11:09:51 2016 +0900
fix typos
commit f7b3b11
Author: Harisankar H <harisankarh@gmail.com>
Date: Wed Mar 9 11:49:42 2016 +0530
Typo correction: "faield" --> "failed"
Typo correction: "faield" --> "failed"
commit 3fd40fc
Author: Itamar Haber <itamar@redislabs.com>
Date: Thu Feb 25 10:31:51 2016 +0200
Fixes a typo in comments
commit 621c160
Author: Prayag Verma <prayag.verma@gmail.com>
Date: Mon Feb 1 12:36:20 2016 +0530
Fix typo in Readme.md
Spelling mistakes -
`eviciton` > `eviction`
`familar` > `familiar`
commit d7d07d6
Author: WonCheol Lee <toctoc21c@gmail.com>
Date: Wed Dec 30 15:11:34 2015 +0900
Typo fixed
commit a4dade7
Author: Felix Bünemann <buenemann@louis.info>
Date: Mon Dec 28 11:02:55 2015 +0100
[ci skip] Improve supervised upstart config docs
This mentions that "expect stop" is required for supervised upstart
to work correctly. See http://upstart.ubuntu.com/cookbook/#expect-stop
for an explanation.
commit d9caba9
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:30:03 2015 +1100
README: Remove trailing whitespace
commit 72d42e5
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:29:32 2015 +1100
README: Fix typo. th => the
commit dd6e957
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:29:20 2015 +1100
README: Fix typo. familar => familiar
commit 3a12b23
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:28:54 2015 +1100
README: Fix typo. eviciton => eviction
commit 2d1d03b
Author: daurnimator <quae@daurnimator.com>
Date: Mon Dec 21 18:21:45 2015 +1100
README: Fix typo. sever => server
commit 3973b06
Author: Itamar Haber <itamar@garantiadata.com>
Date: Sat Dec 19 17:01:20 2015 +0200
Typo fix
commit 4f2e460
Author: Steve Gao <fu@2token.com>
Date: Fri Dec 4 10:22:05 2015 +0800
Update README - fix typos
commit b21667c
Author: binyan <binbin.yan@nokia.com>
Date: Wed Dec 2 22:48:37 2015 +0800
delete redundancy color judge in sdscatcolor
commit 88894c7
Author: binyan <binbin.yan@nokia.com>
Date: Wed Dec 2 22:14:42 2015 +0800
the example output shoule be HelloWorld
commit 2763470
Author: binyan <binbin.yan@nokia.com>
Date: Wed Dec 2 17:41:39 2015 +0800
modify error word keyevente
Signed-off-by: binyan <binbin.yan@nokia.com>
commit 0847b3d
Author: Bruno Martins <bscmartins@gmail.com>
Date: Wed Nov 4 11:37:01 2015 +0000
typo
commit bbb9e9e
Author: dawedawe <dawedawe@gmx.de>
Date: Fri Mar 27 00:46:41 2015 +0100
typo: zimap -> zipmap
commit 5ed297e
Author: Axel Advento <badwolf.bloodseeker.rev@gmail.com>
Date: Tue Mar 3 15:58:29 2015 +0800
Fix 'salve' typos to 'slave'
commit edec9d6
Author: LudwikJaniuk <ludvig.janiuk@gmail.com>
Date: Wed Jun 12 14:12:47 2019 +0200
Update README.md
Co-Authored-By: Qix <Qix-@users.noreply.github.com>
commit 692a7af
Author: LudwikJaniuk <ludvig.janiuk@gmail.com>
Date: Tue May 28 14:32:04 2019 +0200
grammar
commit d962b0a
Author: Nick Frost <nickfrostatx@gmail.com>
Date: Wed Jul 20 15:17:12 2016 -0700
Minor grammar fix
commit 24fff01aaccaf5956973ada8c50ceb1462e211c6 (typos)
Author: Chad Miller <chadm@squareup.com>
Date: Tue Sep 8 13:46:11 2020 -0400
Fix faulty comment about operation of unlink()
commit 3cd5c1f3326c52aa552ada7ec797c6bb16452355
Author: Kevin <kevin.xgr@gmail.com>
Date: Wed Nov 20 00:13:50 2019 +0800
Fix typo in server.c.
From a83af59 Mon Sep 17 00:00:00 2001
From: wuwo <wuwo@wacai.com>
Date: Fri, 17 Mar 2017 20:37:45 +0800
Subject: [PATCH] falure to failure
From c961896 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=B7=A6=E6=87=B6?= <veficos@gmail.com>
Date: Sat, 27 May 2017 15:33:04 +0800
Subject: [PATCH] fix typo
From e600ef2 Mon Sep 17 00:00:00 2001
From: "rui.zou" <rui.zou@yunify.com>
Date: Sat, 30 Sep 2017 12:38:15 +0800
Subject: [PATCH] fix a typo
From c7d07fa Mon Sep 17 00:00:00 2001
From: Alexandre Perrin <alex@kaworu.ch>
Date: Thu, 16 Aug 2018 10:35:31 +0200
Subject: [PATCH] deps README.md typo
From b25cb67 Mon Sep 17 00:00:00 2001
From: Guy Korland <gkorland@gmail.com>
Date: Wed, 26 Sep 2018 10:55:37 +0300
Subject: [PATCH 1/2] fix typos in header
From ad28ca6 Mon Sep 17 00:00:00 2001
From: Guy Korland <gkorland@gmail.com>
Date: Wed, 26 Sep 2018 11:02:36 +0300
Subject: [PATCH 2/2] fix typos
commit 34924cdedd8552466fc22c1168d49236cb7ee915
Author: Adrian Lynch <adi_ady_ade@hotmail.com>
Date: Sat Apr 4 21:59:15 2015 +0100
Typos fixed
commit fd2a1e7
Author: Jan <jsteemann@users.noreply.github.com>
Date: Sat Oct 27 19:13:01 2018 +0200
Fix typos
Fix typos
commit e14e47c1a234b53b0e103c5f6a1c61481cbcbb02
Author: Andy Lester <andy@petdance.com>
Date: Fri Aug 2 22:30:07 2019 -0500
Fix multiple misspellings of "following"
commit 79b948ce2dac6b453fe80995abbcaac04c213d5a
Author: Andy Lester <andy@petdance.com>
Date: Fri Aug 2 22:24:28 2019 -0500
Fix misspelling of create-cluster
commit 1fffde52666dc99ab35efbd31071a4c008cb5a71
Author: Andy Lester <andy@petdance.com>
Date: Wed Jul 31 17:57:56 2019 -0500
Fix typos
commit 204c9ba9651e9e05fd73936b452b9a30be456cfe
Author: Xiaobo Zhu <xiaobo.zhu@shopee.com>
Date: Tue Aug 13 22:19:25 2019 +0800
fix typos
Squashed commit of the following:
commit 1d9aaf8
Author: danmedani <danmedani@gmail.com>
Date: Sun Aug 2 11:40:26 2015 -0700
README typo fix.
Squashed commit of the following:
commit 32bfa7c
Author: Erik Dubbelboer <erik@dubbelboer.com>
Date: Mon Jul 6 21:15:08 2015 +0200
Fixed grammer
Squashed commit of the following:
commit b24f69c
Author: Sisir Koppaka <sisir.koppaka@gmail.com>
Date: Mon Mar 2 22:38:45 2015 -0500
utils/hashtable/rehashing.c: Fix typos
Squashed commit of the following:
commit 4e04082
Author: Erik Dubbelboer <erik@dubbelboer.com>
Date: Mon Mar 23 08:22:21 2015 +0000
Small config file documentation improvements
Squashed commit of the following:
commit acb8773
Author: ctd1500 <ctd1500@gmail.com>
Date: Fri May 8 01:52:48 2015 -0700
Typo and grammar fixes in readme
commit 2eb75b6
Author: ctd1500 <ctd1500@gmail.com>
Date: Fri May 8 01:36:18 2015 -0700
fixed redis.conf comment
Squashed commit of the following:
commit a8249a2
Author: Masahiko Sawada <sawada.mshk@gmail.com>
Date: Fri Dec 11 11:39:52 2015 +0530
Revise correction of typos.
Squashed commit of the following:
commit 3c02028
Author: zhaojun11 <zhaojun11@jd.com>
Date: Wed Jan 17 19:05:28 2018 +0800
Fix typos include two code typos in cluster.c and latency.c
Squashed commit of the following:
commit 9dba47c
Author: q191201771 <191201771@qq.com>
Date: Sat Jan 4 11:31:04 2020 +0800
fix function listCreate comment in adlist.c
Update src/server.c
commit 2c7c2cb536e78dd211b1ac6f7bda00f0f54faaeb
Author: charpty <charpty@gmail.com>
Date: Tue May 1 23:16:59 2018 +0800
server.c typo: modules system dictionary type comment
Signed-off-by: charpty <charpty@gmail.com>
commit a8395323fb63cb59cb3591cb0f0c8edb7c29a680
Author: Itamar Haber <itamar@redislabs.com>
Date: Sun May 6 00:25:18 2018 +0300
Updates test_helper.tcl's help with undocumented options
Specifically:
* Host
* Port
* Client
commit bde6f9ced15755cd6407b4af7d601b030f36d60b
Author: wxisme <850885154@qq.com>
Date: Wed Aug 8 15:19:19 2018 +0800
fix comments in deps files
commit 3172474ba991532ab799ee1873439f3402412331
Author: wxisme <850885154@qq.com>
Date: Wed Aug 8 14:33:49 2018 +0800
fix some comments
commit 01b6f2b6858b5cf2ce4ad5092d2c746e755f53f0
Author: Thor Juhasz <thor@juhasz.pro>
Date: Sun Nov 18 14:37:41 2018 +0100
Minor fixes to comments
Found some parts a little unclear on a first read, which prompted me to have a better look at the file and fix some minor things I noticed.
Fixing minor typos and grammar. There are no changes to configuration options.
These changes are only meant to help the user better understand the explanations to the various configuration options
2020-09-10 13:43:38 +03:00
|
|
|
/* Struct to hold an inclusive/exclusive range spec by score comparison. */
|
2011-04-28 19:00:33 +02:00
|
|
|
typedef struct {
|
|
|
|
double min, max;
|
|
|
|
int minex, maxex; /* are min or max exclusive? */
|
|
|
|
} zrangespec;
|
|
|
|
|
2012-02-04 08:11:31 +01:00
|
|
|
/* Struct to hold an inclusive/exclusive range spec by lexicographic comparison. */
|
|
|
|
typedef struct {
|
2015-08-04 09:20:55 +02:00
|
|
|
sds min, max; /* May be set to shared.(minstring|maxstring) */
|
2012-02-04 08:11:31 +01:00
|
|
|
int minex, maxex; /* are min or max exclusive? */
|
|
|
|
} zlexrangespec;
|
|
|
|
|
Sort out the mess around Lua error messages and error stats (#10329)
This PR fix 2 issues on Lua scripting:
* Server error reply statistics (some errors were counted twice).
* Error code and error strings returning from scripts (error code was missing / misplaced).
## Statistics
a Lua script user is considered part of the user application, a sophisticated transaction,
so we want to count an error even if handled silently by the script, but when it is
propagated outwards from the script we don't wanna count it twice. on the other hand,
if the script decides to throw an error on its own (using `redis.error_reply`), we wanna
count that too.
Besides, we do count the `calls` in command statistics for the commands the script calls,
we we should certainly also count `failed_calls`.
So when a simple `eval "return redis.call('set','x','y')" 0` fails, it should count the failed call
to both SET and EVAL, but the `errorstats` and `total_error_replies` should be counted only once.
The PR changes the error object that is raised on errors. Instead of raising a simple Lua
string, Redis will raise a Lua table in the following format:
```
{
err='<error message (including error code)>',
source='<User source file name>',
line='<line where the error happned>',
ignore_error_stats_update=true/false,
}
```
The `luaPushError` function was modified to construct the new error table as describe above.
The `luaRaiseError` was renamed to `luaError` and is now simply called `lua_error` to raise
the table on the top of the Lua stack as the error object.
The reason is that since its functionality is changed, in case some Redis branch / fork uses it,
it's better to have a compilation error than a bug.
The `source` and `line` fields are enriched by the error handler (if possible) and the
`ignore_error_stats_update` is optional and if its not present then the default value is `false`.
If `ignore_error_stats_update` is true, the error will not be counted on the error stats.
When parsing Redis call reply, each error is translated to a Lua table on the format describe
above and the `ignore_error_stats_update` field is set to `true` so we will not count errors
twice (we counted this error when we invoke the command).
The changes in this PR might have been considered as a breaking change for users that used
Lua `pcall` function. Before, the error was a string and now its a table. To keep backward
comparability the PR override the `pcall` implementation and extract the error message from
the error table and return it.
Example of the error stats update:
```
127.0.0.1:6379> lpush l 1
(integer) 2
127.0.0.1:6379> eval "return redis.call('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value. script: e471b73f1ef44774987ab00bdf51f21fd9f7974a, on @user_script:1.
127.0.0.1:6379> info Errorstats
# Errorstats
errorstat_WRONGTYPE:count=1
127.0.0.1:6379> info commandstats
# Commandstats
cmdstat_eval:calls=1,usec=341,usec_per_call=341.00,rejected_calls=0,failed_calls=1
cmdstat_info:calls=1,usec=35,usec_per_call=35.00,rejected_calls=0,failed_calls=0
cmdstat_lpush:calls=1,usec=14,usec_per_call=14.00,rejected_calls=0,failed_calls=0
cmdstat_get:calls=1,usec=10,usec_per_call=10.00,rejected_calls=0,failed_calls=1
```
## error message
We can now construct the error message (sent as a reply to the user) from the error table,
so this solves issues where the error message was malformed and the error code appeared
in the middle of the error message:
```diff
127.0.0.1:6379> eval "return redis.call('set','x','y')" 0
-(error) ERR Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479): @user_script:1: OOM command not allowed when used memory > 'maxmemory'.
+(error) OOM command not allowed when used memory > 'maxmemory' @user_script:1. Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479)
```
```diff
127.0.0.1:6379> eval "redis.call('get', 'l')" 0
-(error) ERR Error running script (call to f_8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1): @user_script:1: WRONGTYPE Operation against a key holding the wrong kind of value
+(error) WRONGTYPE Operation against a key holding the wrong kind of value script: 8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1, on @user_script:1.
```
Notica that `redis.pcall` was not change:
```
127.0.0.1:6379> eval "return redis.pcall('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value
```
## other notes
Notice that Some commands (like GEOADD) changes the cmd variable on the client stats so we
can not count on it to update the command stats. In order to be able to update those stats correctly
we needed to promote `realcmd` variable to be located on the client struct.
Tests was added and modified to verify the changes.
Related PR's: #10279, #10218, #10278, #10309
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-27 13:40:57 +02:00
|
|
|
/* flags for incrCommandFailedCalls */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define ERROR_COMMAND_REJECTED (1 << 0) /* Indicate to update the command rejected stats */
|
|
|
|
#define ERROR_COMMAND_FAILED (1 << 1) /* Indicate to update the command failed stats */
|
Sort out the mess around Lua error messages and error stats (#10329)
This PR fix 2 issues on Lua scripting:
* Server error reply statistics (some errors were counted twice).
* Error code and error strings returning from scripts (error code was missing / misplaced).
## Statistics
a Lua script user is considered part of the user application, a sophisticated transaction,
so we want to count an error even if handled silently by the script, but when it is
propagated outwards from the script we don't wanna count it twice. on the other hand,
if the script decides to throw an error on its own (using `redis.error_reply`), we wanna
count that too.
Besides, we do count the `calls` in command statistics for the commands the script calls,
we we should certainly also count `failed_calls`.
So when a simple `eval "return redis.call('set','x','y')" 0` fails, it should count the failed call
to both SET and EVAL, but the `errorstats` and `total_error_replies` should be counted only once.
The PR changes the error object that is raised on errors. Instead of raising a simple Lua
string, Redis will raise a Lua table in the following format:
```
{
err='<error message (including error code)>',
source='<User source file name>',
line='<line where the error happned>',
ignore_error_stats_update=true/false,
}
```
The `luaPushError` function was modified to construct the new error table as describe above.
The `luaRaiseError` was renamed to `luaError` and is now simply called `lua_error` to raise
the table on the top of the Lua stack as the error object.
The reason is that since its functionality is changed, in case some Redis branch / fork uses it,
it's better to have a compilation error than a bug.
The `source` and `line` fields are enriched by the error handler (if possible) and the
`ignore_error_stats_update` is optional and if its not present then the default value is `false`.
If `ignore_error_stats_update` is true, the error will not be counted on the error stats.
When parsing Redis call reply, each error is translated to a Lua table on the format describe
above and the `ignore_error_stats_update` field is set to `true` so we will not count errors
twice (we counted this error when we invoke the command).
The changes in this PR might have been considered as a breaking change for users that used
Lua `pcall` function. Before, the error was a string and now its a table. To keep backward
comparability the PR override the `pcall` implementation and extract the error message from
the error table and return it.
Example of the error stats update:
```
127.0.0.1:6379> lpush l 1
(integer) 2
127.0.0.1:6379> eval "return redis.call('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value. script: e471b73f1ef44774987ab00bdf51f21fd9f7974a, on @user_script:1.
127.0.0.1:6379> info Errorstats
# Errorstats
errorstat_WRONGTYPE:count=1
127.0.0.1:6379> info commandstats
# Commandstats
cmdstat_eval:calls=1,usec=341,usec_per_call=341.00,rejected_calls=0,failed_calls=1
cmdstat_info:calls=1,usec=35,usec_per_call=35.00,rejected_calls=0,failed_calls=0
cmdstat_lpush:calls=1,usec=14,usec_per_call=14.00,rejected_calls=0,failed_calls=0
cmdstat_get:calls=1,usec=10,usec_per_call=10.00,rejected_calls=0,failed_calls=1
```
## error message
We can now construct the error message (sent as a reply to the user) from the error table,
so this solves issues where the error message was malformed and the error code appeared
in the middle of the error message:
```diff
127.0.0.1:6379> eval "return redis.call('set','x','y')" 0
-(error) ERR Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479): @user_script:1: OOM command not allowed when used memory > 'maxmemory'.
+(error) OOM command not allowed when used memory > 'maxmemory' @user_script:1. Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479)
```
```diff
127.0.0.1:6379> eval "redis.call('get', 'l')" 0
-(error) ERR Error running script (call to f_8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1): @user_script:1: WRONGTYPE Operation against a key holding the wrong kind of value
+(error) WRONGTYPE Operation against a key holding the wrong kind of value script: 8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1, on @user_script:1.
```
Notica that `redis.pcall` was not change:
```
127.0.0.1:6379> eval "return redis.pcall('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value
```
## other notes
Notice that Some commands (like GEOADD) changes the cmd variable on the client stats so we
can not count on it to update the command stats. In order to be able to update those stats correctly
we needed to promote `realcmd` variable to be located on the client struct.
Tests was added and modified to verify the changes.
Related PR's: #10279, #10218, #10278, #10309
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-27 13:40:57 +02:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
zskiplist *zslCreate(void);
|
|
|
|
void zslFree(zskiplist *zsl);
|
2015-08-04 09:20:55 +02:00
|
|
|
zskiplistNode *zslInsert(zskiplist *zsl, double score, sds ele);
|
|
|
|
unsigned char *zzlInsert(unsigned char *zl, sds ele, double score);
|
|
|
|
int zslDelete(zskiplist *zsl, double score, sds ele, zskiplistNode **node);
|
Optimize ZRANGE offset location from linear search to skiplist jump. (#12450)
ZRANGE BYSCORE/BYLEX with [LIMIT offset count] option was
using every level in skiplist to jump to the first/last node in range,
but only use level[0] in skiplist to locate the node at offset, resulting
in sub-optimal performance using LIMIT:
```
while (ln && offset--) {
if (reverse) {
ln = ln->backward;
} else {
ln = ln->level[0].forward;
}
}
```
It could be slow when offset is very big. We can get the total rank of
the offset location and use skiplist to jump to it. It is an improvement
from O(offset) to O(log rank).
Below shows how this is implemented (if the offset is positve):
Use the skiplist to seach for the first element in the range, record its
rank `rank_0`, so we can have the rank of the target node `rank_t`.
Meanwhile we record the last node we visited which has zsl->level-1
levels and its rank `rank_1`. Then we start from the zsl->level-1 node,
use skiplist to go forward `rank_t-rank_1` nodes to reach the target node.
It is very similiar when the offset is reversed.
Note that if `rank_t` is very close to `rank_0`, we just start from the first
element in range and go node by node, this for the case when zsl->level-1
node is to far away and it is quicker to reach the target node by node.
Here is a test using a random generated zset including 10000 elements
(with different positive scores), doing a bench mark which compares how
fast the `ZRANGE` command is exucuted before and after the optimization.
The start score is set to 0 and the count is set to 1 to make sure that
most of the time is spent on locating the offset.
```
memtier_benchmark -h 127.0.0.1 -p 6379 --command="zrange test 0 +inf byscore limit <offset> 1"
```
| offset | QPS(unstable) | QPS(optimized) |
|--------|--------|--------|
| 10 | 73386.02 | 74819.82 |
| 1000 | 48084.96 | 73177.73 |
| 2000 | 31156.79 | 72805.83 |
| 5000 | 10954.83 | 71218.21 |
With the result above, we can see that the original code is greatly
slowed down when offset gets bigger, and with the optimization the
speed is almost not affected.
Similiar results are generated when testing reversed offset:
```
memtier_benchmark -h 127.0.0.1 -p 6379 --command="zrange test +inf 0 byscore rev limit <offset> 1"
```
| offset | QPS(unstable) | QPS(optimized) |
|--------|--------|--------|
| 10 | 74505.14 | 71653.67 |
| 1000 | 46829.25 | 72842.75 |
| 2000 | 28985.48 | 73669.01 |
| 5000 | 11066.22 | 73963.45 |
And the same conclusion is drawn from the tests of ZRANGE BYLEX.
2023-08-31 19:42:08 +08:00
|
|
|
zskiplistNode *zslNthInRange(zskiplist *zsl, zrangespec *range, long n);
|
2011-03-14 13:30:06 +01:00
|
|
|
double zzlGetScore(unsigned char *sptr);
|
|
|
|
void zzlNext(unsigned char *zl, unsigned char **eptr, unsigned char **sptr);
|
|
|
|
void zzlPrev(unsigned char *zl, unsigned char **eptr, unsigned char **sptr);
|
2016-04-19 15:22:33 +02:00
|
|
|
unsigned char *zzlFirstInRange(unsigned char *zl, zrangespec *range);
|
2016-04-20 12:38:14 +02:00
|
|
|
unsigned char *zzlLastInRange(unsigned char *zl, zrangespec *range);
|
2017-12-08 15:37:08 +08:00
|
|
|
unsigned long zsetLength(const robj *zobj);
|
2011-03-10 17:50:13 +01:00
|
|
|
void zsetConvert(robj *zobj, int encoding);
|
2021-10-04 12:11:02 +03:00
|
|
|
void zsetConvertToListpackIfNeeded(robj *zobj, size_t maxelelen, size_t totelelen);
|
2015-08-04 09:20:55 +02:00
|
|
|
int zsetScore(robj *zobj, sds member, double *score);
|
|
|
|
unsigned long zslGetRank(zskiplist *zsl, double score, sds o);
|
2021-03-10 15:09:43 +01:00
|
|
|
int zsetAdd(robj *zobj, double score, sds ele, int in_flags, int *out_flags, double *newscore);
|
2022-11-28 17:57:11 +08:00
|
|
|
long zsetRank(robj *zobj, sds ele, int reverse, double *score);
|
2016-04-15 15:20:25 +02:00
|
|
|
int zsetDel(robj *zobj, sds ele);
|
2020-11-17 19:03:05 +09:00
|
|
|
robj *zsetDup(robj *o);
|
2024-05-22 23:24:12 -07:00
|
|
|
void genericZpopCommand(client *c,
|
|
|
|
robj **keyv,
|
|
|
|
int keyc,
|
|
|
|
int where,
|
|
|
|
int emitkey,
|
|
|
|
long count,
|
|
|
|
int use_nested_array,
|
|
|
|
int reply_nil_when_empty,
|
|
|
|
int *deleted);
|
2021-09-09 23:18:53 +08:00
|
|
|
sds lpGetObject(unsigned char *sptr);
|
2016-04-19 17:02:24 +02:00
|
|
|
int zslValueGteMin(double value, zrangespec *spec);
|
|
|
|
int zslValueLteMax(double value, zrangespec *spec);
|
2016-04-20 23:01:40 +02:00
|
|
|
void zslFreeLexRange(zlexrangespec *spec);
|
2016-04-21 09:27:13 +02:00
|
|
|
int zslParseLexRange(robj *min, robj *max, zlexrangespec *spec);
|
|
|
|
unsigned char *zzlFirstInLexRange(unsigned char *zl, zlexrangespec *range);
|
|
|
|
unsigned char *zzlLastInLexRange(unsigned char *zl, zlexrangespec *range);
|
Optimize ZRANGE offset location from linear search to skiplist jump. (#12450)
ZRANGE BYSCORE/BYLEX with [LIMIT offset count] option was
using every level in skiplist to jump to the first/last node in range,
but only use level[0] in skiplist to locate the node at offset, resulting
in sub-optimal performance using LIMIT:
```
while (ln && offset--) {
if (reverse) {
ln = ln->backward;
} else {
ln = ln->level[0].forward;
}
}
```
It could be slow when offset is very big. We can get the total rank of
the offset location and use skiplist to jump to it. It is an improvement
from O(offset) to O(log rank).
Below shows how this is implemented (if the offset is positve):
Use the skiplist to seach for the first element in the range, record its
rank `rank_0`, so we can have the rank of the target node `rank_t`.
Meanwhile we record the last node we visited which has zsl->level-1
levels and its rank `rank_1`. Then we start from the zsl->level-1 node,
use skiplist to go forward `rank_t-rank_1` nodes to reach the target node.
It is very similiar when the offset is reversed.
Note that if `rank_t` is very close to `rank_0`, we just start from the first
element in range and go node by node, this for the case when zsl->level-1
node is to far away and it is quicker to reach the target node by node.
Here is a test using a random generated zset including 10000 elements
(with different positive scores), doing a bench mark which compares how
fast the `ZRANGE` command is exucuted before and after the optimization.
The start score is set to 0 and the count is set to 1 to make sure that
most of the time is spent on locating the offset.
```
memtier_benchmark -h 127.0.0.1 -p 6379 --command="zrange test 0 +inf byscore limit <offset> 1"
```
| offset | QPS(unstable) | QPS(optimized) |
|--------|--------|--------|
| 10 | 73386.02 | 74819.82 |
| 1000 | 48084.96 | 73177.73 |
| 2000 | 31156.79 | 72805.83 |
| 5000 | 10954.83 | 71218.21 |
With the result above, we can see that the original code is greatly
slowed down when offset gets bigger, and with the optimization the
speed is almost not affected.
Similiar results are generated when testing reversed offset:
```
memtier_benchmark -h 127.0.0.1 -p 6379 --command="zrange test +inf 0 byscore rev limit <offset> 1"
```
| offset | QPS(unstable) | QPS(optimized) |
|--------|--------|--------|
| 10 | 74505.14 | 71653.67 |
| 1000 | 46829.25 | 72842.75 |
| 2000 | 28985.48 | 73669.01 |
| 5000 | 11066.22 | 73963.45 |
And the same conclusion is drawn from the tests of ZRANGE BYLEX.
2023-08-31 19:42:08 +08:00
|
|
|
zskiplistNode *zslNthInLexRange(zskiplist *zsl, zlexrangespec *range, long n);
|
2016-04-21 11:17:00 +02:00
|
|
|
int zzlLexValueGteMin(unsigned char *p, zlexrangespec *spec);
|
|
|
|
int zzlLexValueLteMax(unsigned char *p, zlexrangespec *spec);
|
|
|
|
int zslLexValueGteMin(sds value, zlexrangespec *spec);
|
|
|
|
int zslLexValueLteMax(sds value, zlexrangespec *spec);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* Core functions */
|
2018-04-11 12:48:26 +02:00
|
|
|
int getMaxmemoryState(size_t *total, size_t *logical, size_t *tofree, float *level);
|
2023-05-02 17:31:32 -07:00
|
|
|
size_t freeMemoryGetNotCountedMemory(void);
|
Limit the main db and expires dictionaries to expand (#7954)
As we know, redis may reject user's requests or evict some keys if
used memory is over maxmemory. Dictionaries expanding may make
things worse, some big dictionaries, such as main db and expires dict,
may eat huge memory at once for allocating a new big hash table and be
far more than maxmemory after expanding.
There are related issues: #4213 #4583
More details, when expand dict in redis, we will allocate a new big
ht[1] that generally is double of ht[0], The size of ht[1] will be
very big if ht[0] already is big. For db dict, if we have more than
64 million keys, we need to cost 1GB for ht[1] when dict expands.
If the sum of used memory and new hash table of dict needed exceeds
maxmemory, we shouldn't allow the dict to expand. Because, if we
enable keys eviction, we still couldn't add much more keys after
eviction and rehashing, what's worse, redis will keep less keys when
redis only remains a little memory for storing new hash table instead
of users' data. Moreover users can't write data in redis if disable
keys eviction.
What this commit changed ?
Add a new member function expandAllowed for dict type, it provide a way
for caller to allow expand or not. We expose two parameters for this
function: more memory needed for expanding and dict current load factor,
users can implement a function to make a decision by them.
For main db dict and expires dict type, these dictionaries may be very
big and cost huge memory for expanding, so we implement a judgement
function: we can stop dict to expand provisionally if used memory will
be over maxmemory after dict expands, but to guarantee the performance
of redis, we still allow dict to expand if dict load factor exceeds the
safe load factor.
Add test cases to verify we don't allow main db to expand when left
memory is not enough, so that avoid keys eviction.
Other changes:
For new hash table size when expand. Before this commit, the size is
that double used of dict and later _dictNextPower. Actually we aim to
control a dict load factor between 0.5 and 1.0. Now we replace *2 with
+1, since the first check is that used >= size, the outcome of before
will usually be the same as _dictNextPower(used+1). The only case where
it'll differ is when dict_can_resize is false during fork, so that later
the _dictNextPower(used*2) will cause the dict to jump to *4 (i.e.
_dictNextPower(1025*2) will return 4096).
Fix rehash test cases due to changing algorithm of new hash table size
when expand.
2020-12-06 17:53:04 +08:00
|
|
|
int overMaxmemoryAfterAlloc(size_t moremem);
|
2022-08-28 13:10:10 +03:00
|
|
|
uint64_t getCommandFlags(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
int processCommand(client *c);
|
optimize(remove) usage of client's pending_querybuf (#10413)
To remove `pending_querybuf`, the key point is reusing `querybuf`, it means master client's `querybuf` is not only used to parse command, but also proxy to sub-replicas.
1. add a new variable `repl_applied` for master client to record how many data applied (propagated via `replicationFeedStreamFromMasterStream()`) but not trimmed in `querybuf`.
2. don't sdsrange `querybuf` in `commandProcessed()`, we trim it to `repl_applied` after the whole replication pipeline processed to avoid fragmented `sdsrange`. And here are some scenarios we cannot trim to `qb_pos`:
* we don't receive complete command from master
* master client blocked because of client pause
* IO threads operate read, master client flagged with CLIENT_PENDING_COMMAND
In these scenarios, `qb_pos` points to the part of the current command or the beginning of next command, and the current command is not applied yet, so the `repl_applied` is not equal to `qb_pos`.
Some other notes:
* Do not do big arg optimization on master client, since we can only sdsrange `querybuf` after data sent to replicas.
* Set `qb_pos` and `repl_applied` to 0 when `freeClient` in `replicationCacheMaster`.
* Rewrite `processPendingCommandsAndResetClient` to `processPendingCommandAndInputBuffer`, let `processInputBuffer` to be called successively after `processCommandAndResetClient`.
2022-03-25 10:45:40 +08:00
|
|
|
int processPendingCommandAndInputBuffer(client *c);
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
int processCommandAndResetClient(client *c);
|
2011-03-06 17:49:22 +01:00
|
|
|
void setupSignalHandlers(void);
|
Introduce .listen into connection type
Introduce listen method into connection type, this allows no hard code
of listen logic. Originally, we initialize server during startup like
this:
if (server.port)
listenToPort(server.port,&server.ipfd);
if (server.tls_port)
listenToPort(server.port,&server.tlsfd);
if (server.unixsocket)
anetUnixServer(...server.unixsocket...);
...
if (createSocketAcceptHandler(&server.ipfd, acceptTcpHandler) != C_OK)
if (createSocketAcceptHandler(&server.tlsfd, acceptTcpHandler) != C_OK)
if (createSocketAcceptHandler(&server.sofd, acceptTcpHandler) != C_OK)
...
If a new connection type gets supported, we have to add more hard code
to setup listener.
Introduce .listen and refactor listener, and Unix socket supports this.
this allows to setup listener arguments and create listener in a loop.
What's more, '.listen' is defined in connection.h, so we should include
server.h to import 'struct socketFds', but server.h has already include
'connection.h'. To avoid including loop(also to make code reasonable),
define 'struct connListener' in connection.h instead of 'struct socketFds'
in server.h. This leads this commit to get more changes.
There are more fields in 'struct connListener', hence it's possible to
simplify changeBindAddr & applyTLSPort() & updatePort() into a single
logic: update the listener config from the server.xxx, and re-create
the listener.
Because of the new field 'priv' in struct connListener, we expect to pass
this to the accept handler(even it's not used currently), this may be used
in the future.
Signed-off-by: zhenwei pi <pizhenwei@bytedance.com>
2022-07-27 12:18:28 +08:00
|
|
|
int createSocketAcceptHandler(connListener *sfd, aeFileProc *accept_handler);
|
|
|
|
connListener *listenerByType(const char *typename);
|
|
|
|
int changeListener(connListener *listener);
|
|
|
|
void closeListener(connListener *listener);
|
2024-04-03 23:54:33 +07:00
|
|
|
struct serverCommand *lookupSubcommand(struct serverCommand *container, sds sub_name);
|
|
|
|
struct serverCommand *lookupCommand(robj **argv, int argc);
|
|
|
|
struct serverCommand *lookupCommandBySdsLogic(dict *commands, sds s);
|
|
|
|
struct serverCommand *lookupCommandBySds(sds s);
|
|
|
|
struct serverCommand *lookupCommandByCStringLogic(dict *commands, const char *s);
|
|
|
|
struct serverCommand *lookupCommandByCString(const char *s);
|
|
|
|
struct serverCommand *lookupCommandOrOriginal(robj **argv, int argc);
|
2022-04-25 12:08:13 +02:00
|
|
|
int commandCheckExistence(client *c, sds *err);
|
Io thread work offload (#763)
### IO-Threads Work Offloading
This PR is the 2nd of 3 PRs intended to achieve the goal of 1M requests
per second.
(1st PR: https://github.com/valkey-io/valkey/pull/758)
This PR offloads additional work to the I/O threads, beyond the current
read-parse/write operations, to better utilize the I/O threads and
reduce the load on the main thread.
It contains the following 3 commits:
### Poll Offload
Currently, the main thread is responsible for executing the poll-wait
system call, while the IO threads wait for tasks from the main thread.
The poll-wait operation is expensive and can consume up to 30% of the
main thread's time. We could have let the IO threads do the poll-wait by
themselves, with each thread listening to some of the clients and
notifying the main thread when a client's command is ready to execute.
However, the current approach, where the main thread listens for events
from the network, has several benefits. The main thread remains in
charge, allowing it to know the state of each client
(idle/read/write/close) at any given time. Additionally, it makes the
threads flexible, enabling us to drain an IO thread's job queue and stop
a thread when the load is light without modifying the event loop and
moving its clients to a different IO thread. Furthermore, with this
approach, the IO threads don't need to wait for both messages from the
network and from the main thread instead, the threads wait only for
tasks from the main thread.
To enjoy the benefits of both the main thread remaining in charge and
the poll being offloaded, we propose offloading the poll-wait as a
single-time, non-blocking job to one of the IO threads. The IO thread
will perform a poll-wait non-blocking call while the main thread
processes the client commands. Later, in `aeProcessEvents`, instead of
sleeping on the poll, we check for the IO thread's poll-wait results.
The poll-wait will be offloaded in `beforeSleep` only when there are
ready events for the main thread to process. If no events are pending,
the main thread will revert to the current behavior and sleep on the
poll by itself.
**Implementation Details**
A new call back `custompoll` was added to the `aeEventLoop` when not set
to `NULL` the ae will call the `custompoll` callback instead of the
`aeApiPoll`.
When the poll is offloaded we will set the `custompoll` to
`getIOThreadPollResults` and send a poll-job to the thread. the thread
will take a mutex, call a non-blocking (with timeout 0) to `aePoll`
which will populate the fired events array. the IO thread will set the
`server.io_fired_events` to the number of the returning `numevents`,
later the main-thread in `custompoll` will return the
`server.io_fired_events` and will set the `customPoll` back to `NULL`.
To ensure thread safety when accessing server.el, all functions that
modify the eventloop events were wrapped with a mutex to ensure mutual
exclusion when modifying the events.
### Command Lookup Offload
As the IO thread parses the command from the client's Querybuf, it can
perform a command lookup in the commands dictionary, which can consume
up to ~5% of the main-thread runtime.
**Implementation details**
The IO thread will store the looked-up command in the client's new field
`io_parsed_cmd` field. We can't use `c->cmd` for that since we use
`c->cmd `to check if a command was reprocessed or not.
To ensure thread safety when accessing the command dictionary, we make
sure the main thread isn't changing the dictionary while IO threads are
accessing it. This is accomplished by introducing a new flag called
`no_incremental_rehash` for the `dictType` commands. When performing
`dictResize`, we will rehash the entire dictionary in place rather than
deferring the process.
### Free Offload
Since the command arguments are allocated by the I/O thread, it would be
beneficial if they were also freed by the same thread. If the main
thread frees objects allocated by the I/O thread, two issues arise:
1. During the freeing process, the main thread needs to access the SDS
pointed to by the object to get its length.
2. With Jemalloc, each thread manages thread local pool (`tcache`) of
buffers for quick reallocation without accessing the arena. If the main
thread constantly frees objects allocated by other threads, those
threads will have to frequently access the shared arena to obtain new
memory allocations
**Implementation Details**
When freeing the client's argv, we will send the argv array to the
thread that allocated it. The thread will be identified by the client
ID. When freeing an object during `dbOverwrite`, we will offload the
object free as well. We will extend this to offload the free during
`dbDelete` in a future PR, as its effects on defrag/memory evictions
need to be studied.
---------
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-19 05:21:45 +03:00
|
|
|
int commandCheckArity(struct serverCommand *cmd, int argc, sds *err);
|
2023-05-02 17:31:32 -07:00
|
|
|
void startCommandExecution(void);
|
2024-04-03 23:54:33 +07:00
|
|
|
int incrCommandStatsOnError(struct serverCommand *cmd, int flags);
|
2015-07-26 15:20:46 +02:00
|
|
|
void call(client *c, int flags);
|
2021-09-15 11:53:42 +02:00
|
|
|
void alsoPropagate(int dbid, robj **argv, int argc, int target);
|
2023-05-02 17:31:32 -07:00
|
|
|
void postExecutionUnitOperations(void);
|
2024-04-03 11:26:20 +07:00
|
|
|
void serverOpArrayFree(serverOpArray *oa);
|
2015-07-26 15:20:46 +02:00
|
|
|
void forceCommandPropagation(client *c, int flags);
|
|
|
|
void preventCommandPropagation(client *c);
|
2015-10-29 11:05:27 +01:00
|
|
|
void preventCommandAOF(client *c);
|
|
|
|
void preventCommandReplication(client *c);
|
2024-04-03 23:54:33 +07:00
|
|
|
void slowlogPushCurrentCommand(client *c, struct serverCommand *cmd, ustime_t duration);
|
2024-05-22 23:24:12 -07:00
|
|
|
void updateCommandLatencyHistogram(struct hdr_histogram **latency_histogram, int64_t duration_hist);
|
2020-07-26 08:27:30 +03:00
|
|
|
int prepareForShutdown(int flags);
|
2022-01-02 08:50:15 +01:00
|
|
|
void replyToClientsBlockedOnShutdown(void);
|
|
|
|
int abortShutdown(void);
|
2021-10-07 20:13:42 +08:00
|
|
|
void afterCommand(client *c);
|
Fixes around clients that must be obeyed. Replica report disk errors in PING. (#10603)
This PR unifies all the places that test if the current client is the
master client or AOF client, and uses a method to test that on all of
these.
Other than some refactoring, these are the actual implications:
- Replicas **don't** ignore disk error when processing commands not
coming from their master.
**This is important for PING to be used for health check of replicas**
- SETRANGE, APPEND, SETBIT, BITFIELD don't do proto_max_bulk_len check for AOF
- RM_Call in SCRIPT_MODE ignores disk error when coming from master /
AOF
- RM_Call in cluster mode ignores slot check when processing AOF
- Scripts ignore disk error when processing AOF
- Scripts **don't** ignore disk error on a replica, if the command comes
from clients other than the master
- SCRIPT KILL won't kill script coming from AOF
- Scripts **don't** skip OOM check on replica if the command comes from
clients other than the master
Note that Script, AOF, and module clients don't reach processCommand,
which is why some of the changes don't actually have any implications.
Note, reverting the change done to processCommand in 2f4240b9d9
should be dead code due to the above mentioned fact.
2022-04-20 11:11:21 +03:00
|
|
|
int mustObeyClient(client *c);
|
2013-02-27 12:27:15 +01:00
|
|
|
#ifdef __GNUC__
|
2024-05-22 23:24:12 -07:00
|
|
|
void _serverLog(int level, const char *fmt, ...) __attribute__((format(printf, 2, 3)));
|
|
|
|
void serverLogFromHandler(int level, const char *fmt, ...) __attribute__((format(printf, 2, 3)));
|
2013-02-27 12:27:15 +01:00
|
|
|
#else
|
Fix async safety in signal handlers (#12658)
see discussion from after https://github.com/redis/redis/pull/12453 was
merged
----
This PR replaces signals that are not considered async-signal-safe
(AS-safe) with safe calls.
#### **1. serverLog() and serverLogFromHandler()**
`serverLog` uses unsafe calls. It was decided that we will **avoid**
`serverLog` calls by the signal handlers when:
* The signal is not fatal, such as SIGALRM. In these cases, we prefer
using `serverLogFromHandler` which is the safe version of `serverLog`.
Note they have different prompts:
`serverLog`: `62220:M 26 Oct 2023 14:39:04.526 # <msg>`
`serverLogFromHandler`: `62220:signal-handler (1698331136) <msg>`
* The code was added recently. Calls to `serverLog` by the signal
handler have been there ever since Redis exists and it hasn't caused
problems so far. To avoid regression, from now we should use
`serverLogFromHandler`
#### **2. `snprintf` `fgets` and `strtoul`(base = 16) -------->
`_safe_snprintf`, `fgets_async_signal_safe`, `string_to_hex`**
The safe version of `snprintf` was taken from
[here](https://github.com/twitter/twemcache/blob/8cfc4ca5e76ed936bd3786c8cc43ed47e7778c08/src/mc_util.c#L754)
#### **3. fopen(), fgets(), fclose() --------> open(), read(), close()**
#### **4. opendir(), readdir(), closedir() --------> open(),
syscall(SYS_getdents64), close()**
#### **5. Threads_mngr sync mechanisms**
* waiting for the thread to generate stack trace: semaphore -------->
busy-wait
* `globals_rw_lock` was removed: as we are not using malloc and the
semaphore anymore we don't need to protect `ThreadsManager_cleanups`.
#### **6. Stacktraces buffer**
The initial problem was that we were not able to safely call malloc
within the signal handler.
To solve that we created a buffer on the stack of `writeStacktraces` and
saved it in a global pointer, assuming that under normal circumstances,
the function `writeStacktraces` would complete before any thread
attempted to write to it. However, **if threads lag behind, they might
access this global pointer after it no longer belongs to the
`writeStacktraces` stack, potentially corrupting memory.**
To address this, various solutions were discussed
[here](https://github.com/redis/redis/pull/12658#discussion_r1390442896)
Eventually, we decided to **create a pipe** at server startup that will
remain valid as long as the process is alive.
We chose this solution due to its minimal memory usage, and since
`write()` and `read()` are atomic operations. It ensures that stack
traces from different threads won't mix.
**The stacktraces collection process is now as follows:**
* Cleaning the pipe to eliminate writes of late threads from previous
runs.
* Each thread writes to the pipe its stacktrace
* Waiting for all the threads to mark completion or until a timeout (2
sec) is reached
* Reading from the pipe to print the stacktraces.
#### **7. Changes that were considered and eventually were dropped**
* replace watchdog timer with a POSIX timer:
according to [settimer man](https://linux.die.net/man/2/setitimer)
> POSIX.1-2008 marks getitimer() and setitimer() obsolete, recommending
the use of the POSIX timers API
([timer_gettime](https://linux.die.net/man/2/timer_gettime)(2),
[timer_settime](https://linux.die.net/man/2/timer_settime)(2), etc.)
instead.
However, although it is supposed to conform to POSIX std, POSIX timers
API is not supported on Mac.
You can take a look here at the Linux implementation:
[here](https://github.com/redis/redis/commit/c7562ee13546e504977372fdf40d33c3f86775a5)
To avoid messing up the code, and uncertainty regarding compatibility,
it was decided to drop it for now.
* avoid using sds (uses malloc) in logConfigDebugInfo
It was considered to print config info instead of using sds, however
apparently, `logConfigDebugInfo` does more than just print the sds, so
it was decided this fix is out of this issue scope.
#### **8. fix Signal mask check**
The check `signum & sig_mask` intended to indicate whether the signal is
blocked by the thread was incorrect. Actually, the bit position in the
signal mask corresponds to the signal number. We fixed this by changing
the condition to: `sig_mask & (1L << (sig_num - 1))`
#### **9. Unrelated changes**
both `fork.tcl `and `util.tcl` implemented a function called
`count_log_message` expecting different parameters. This caused
confusion when trying to run daily tests with additional test parameters
to run a specific test.
The `count_log_message` in `fork.tcl` was removed and the calls were
replaced with calls to `count_log_message` located in `util.tcl`
---------
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
2023-11-23 13:22:20 +02:00
|
|
|
void serverLogFromHandler(int level, const char *fmt, ...);
|
2021-03-24 08:22:12 +02:00
|
|
|
void _serverLog(int level, const char *fmt, ...);
|
2013-02-27 12:27:15 +01:00
|
|
|
#endif
|
2015-07-26 15:17:43 +02:00
|
|
|
void serverLogRaw(int level, const char *msg);
|
Fix async safety in signal handlers (#12658)
see discussion from after https://github.com/redis/redis/pull/12453 was
merged
----
This PR replaces signals that are not considered async-signal-safe
(AS-safe) with safe calls.
#### **1. serverLog() and serverLogFromHandler()**
`serverLog` uses unsafe calls. It was decided that we will **avoid**
`serverLog` calls by the signal handlers when:
* The signal is not fatal, such as SIGALRM. In these cases, we prefer
using `serverLogFromHandler` which is the safe version of `serverLog`.
Note they have different prompts:
`serverLog`: `62220:M 26 Oct 2023 14:39:04.526 # <msg>`
`serverLogFromHandler`: `62220:signal-handler (1698331136) <msg>`
* The code was added recently. Calls to `serverLog` by the signal
handler have been there ever since Redis exists and it hasn't caused
problems so far. To avoid regression, from now we should use
`serverLogFromHandler`
#### **2. `snprintf` `fgets` and `strtoul`(base = 16) -------->
`_safe_snprintf`, `fgets_async_signal_safe`, `string_to_hex`**
The safe version of `snprintf` was taken from
[here](https://github.com/twitter/twemcache/blob/8cfc4ca5e76ed936bd3786c8cc43ed47e7778c08/src/mc_util.c#L754)
#### **3. fopen(), fgets(), fclose() --------> open(), read(), close()**
#### **4. opendir(), readdir(), closedir() --------> open(),
syscall(SYS_getdents64), close()**
#### **5. Threads_mngr sync mechanisms**
* waiting for the thread to generate stack trace: semaphore -------->
busy-wait
* `globals_rw_lock` was removed: as we are not using malloc and the
semaphore anymore we don't need to protect `ThreadsManager_cleanups`.
#### **6. Stacktraces buffer**
The initial problem was that we were not able to safely call malloc
within the signal handler.
To solve that we created a buffer on the stack of `writeStacktraces` and
saved it in a global pointer, assuming that under normal circumstances,
the function `writeStacktraces` would complete before any thread
attempted to write to it. However, **if threads lag behind, they might
access this global pointer after it no longer belongs to the
`writeStacktraces` stack, potentially corrupting memory.**
To address this, various solutions were discussed
[here](https://github.com/redis/redis/pull/12658#discussion_r1390442896)
Eventually, we decided to **create a pipe** at server startup that will
remain valid as long as the process is alive.
We chose this solution due to its minimal memory usage, and since
`write()` and `read()` are atomic operations. It ensures that stack
traces from different threads won't mix.
**The stacktraces collection process is now as follows:**
* Cleaning the pipe to eliminate writes of late threads from previous
runs.
* Each thread writes to the pipe its stacktrace
* Waiting for all the threads to mark completion or until a timeout (2
sec) is reached
* Reading from the pipe to print the stacktraces.
#### **7. Changes that were considered and eventually were dropped**
* replace watchdog timer with a POSIX timer:
according to [settimer man](https://linux.die.net/man/2/setitimer)
> POSIX.1-2008 marks getitimer() and setitimer() obsolete, recommending
the use of the POSIX timers API
([timer_gettime](https://linux.die.net/man/2/timer_gettime)(2),
[timer_settime](https://linux.die.net/man/2/timer_settime)(2), etc.)
instead.
However, although it is supposed to conform to POSIX std, POSIX timers
API is not supported on Mac.
You can take a look here at the Linux implementation:
[here](https://github.com/redis/redis/commit/c7562ee13546e504977372fdf40d33c3f86775a5)
To avoid messing up the code, and uncertainty regarding compatibility,
it was decided to drop it for now.
* avoid using sds (uses malloc) in logConfigDebugInfo
It was considered to print config info instead of using sds, however
apparently, `logConfigDebugInfo` does more than just print the sds, so
it was decided this fix is out of this issue scope.
#### **8. fix Signal mask check**
The check `signum & sig_mask` intended to indicate whether the signal is
blocked by the thread was incorrect. Actually, the bit position in the
signal mask corresponds to the signal number. We fixed this by changing
the condition to: `sig_mask & (1L << (sig_num - 1))`
#### **9. Unrelated changes**
both `fork.tcl `and `util.tcl` implemented a function called
`count_log_message` expecting different parameters. This caused
confusion when trying to run daily tests with additional test parameters
to run a specific test.
The `count_log_message` in `fork.tcl` was removed and the calls were
replaced with calls to `count_log_message` located in `util.tcl`
---------
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
2023-11-23 13:22:20 +02:00
|
|
|
void serverLogRawFromHandler(int level, const char *msg);
|
2014-03-16 08:51:54 +09:00
|
|
|
void usage(void);
|
2010-06-22 00:07:48 +02:00
|
|
|
void updateDictResizePolicy(void);
|
2010-11-03 11:23:59 +01:00
|
|
|
void populateCommandTable(void);
|
2024-05-22 23:24:12 -07:00
|
|
|
void resetCommandTableStats(dict *commands);
|
2020-12-31 14:53:43 +00:00
|
|
|
void resetErrorTableStats(void);
|
2013-06-28 17:08:03 +02:00
|
|
|
void adjustOpenFilesLimit(void);
|
2020-12-31 14:53:43 +00:00
|
|
|
void incrementErrorCount(const char *fullerr, size_t namelen);
|
2013-07-05 11:47:20 +02:00
|
|
|
void closeListeningSockets(int unlink_unix_socket);
|
2019-11-05 10:14:34 +01:00
|
|
|
void updateCachedTime(int update_daylight_info);
|
Support for RM_Call on blocking commands (#11568)
Allow running blocking commands from within a module using `RM_Call`.
Today, when `RM_Call` is used, the fake client that is used to run command
is marked with `CLIENT_DENY_BLOCKING` flag. This flag tells the command
that it is not allowed to block the client and in case it needs to block, it must
fallback to some alternative (either return error or perform some default behavior).
For example, `BLPOP` fallback to simple `LPOP` if it is not allowed to block.
All the commands must respect the `CLIENT_DENY_BLOCKING` flag (including
module commands). When the command invocation finished, Redis asserts that
the client was not blocked.
This PR introduces the ability to call blocking command using `RM_Call` by
passing a callback that will be called when the client will get unblocked.
In order to do that, the user must explicitly say that he allow to perform blocking
command by passing a new format specifier argument, `K`, to the `RM_Call`
function. This new flag will tell Redis that it is allow to run blocking command
and block the client. In case the command got blocked, Redis will return a new
type of call reply (`REDISMODULE_REPLY_PROMISE`). This call reply indicates
that the command got blocked and the user can set the on_unblocked handler using
`RM_CallReplyPromiseSetUnblockHandler`.
When clients gets unblocked, it eventually reaches `processUnblockedClients` function.
This is where we check if the client is a fake module client and if it is, we call the unblock
callback instead of performing the usual unblock operations.
**Notice**: `RM_CallReplyPromiseSetUnblockHandler` must be called atomically
along side the command invocation (without releasing the Redis lock in between).
In addition, unlike other CallReply types, the promise call reply must be released
by the module when the Redis GIL is acquired.
The module can abort the execution on the blocking command (if it was not yet
executed) using `RM_CallReplyPromiseAbort`. the API will return `REDISMODULE_OK`
on success and `REDISMODULE_ERR` if the operation is already executed.
**Notice** that in case of misbehave module, Abort might finished successfully but the
operation will not really be aborted. This can only happened if the module do not respect
the disconnect callback of the blocked client.
For pure Redis commands this can not happened.
### Atomicity Guarantees
The API promise that the unblock handler will run atomically as an execution unit.
This means that all the operation performed on the unblock handler will be wrapped
with a multi exec transaction when replicated to the replica and AOF.
The API **do not** grantee any other atomicity properties such as when the unblock
handler will be called. This gives us the flexibility to strengthen the grantees (or not)
in the future if we will decide that we need a better guarantees.
That said, the implementation **does** provide a better guarantees when performing
pure Redis blocking command like `BLPOP`. In this case the unblock handler will run
atomically with the operation that got unblocked (for example, in case of `BLPOP`, the
unblock handler will run atomically with the `LPOP` operation that run when the command
got unblocked). This is an implementation detail that might be change in the future and the
module writer should not count on that.
### Calling blocking commands while running on script mode (`S`)
`RM_Call` script mode (`S`) was introduced on #0372. It is used for usecases where the
command that was invoked on `RM_Call` comes from a user input and we want to make
sure the user will not run dangerous commands like `shutdown`. Some command, such
as `BLPOP`, are marked with `NO_SCRIPT` flag, which means they will not be allowed on
script mode. Those commands are marked with `NO_SCRIPT` just because they are
blocking commands and not because they are dangerous. Now that we can run blocking
commands on RM_Call, there is no real reason not to allow such commands on script mode.
The underline problem is that the `NO_SCRIPT` flag is abused to also mark some of the
blocking commands (notice that those commands know not to block the client if it is not
allowed to do so, and have a fallback logic to such cases. So even if those commands
were not marked with `NO_SCRIPT` flag, it would not harm Redis, and today we can
already run those commands within multi exec).
In addition, not all blocking commands are marked with `NO_SCRIPT` flag, for example
`blmpop` are not marked and can run from within a script.
Those facts shows that there are some ambiguity about the meaning of the `NO_SCRIPT`
flag, and its not fully clear where it should be use.
The PR suggest that blocking commands should not be marked with `NO_SCRIPT` flag,
those commands should handle `CLIENT_DENY_BLOCKING` flag and only block when
it's safe (like they already does today). To achieve that, the PR removes the `NO_SCRIPT`
flag from the following commands:
* `blmove`
* `blpop`
* `brpop`
* `brpoplpush`
* `bzpopmax`
* `bzpopmin`
* `wait`
This might be considered a breaking change as now, on scripts, instead of getting
`command is not allowed from script` error, the user will get some fallback behavior
base on the command implementation. That said, the change matches the behavior
of scripts and multi exec with respect to those commands and allow running them on
`RM_Call` even when script mode is used.
### Additional RedisModule API and changes
* `RM_BlockClientSetPrivateData` - Set private data on the blocked client without the
need to unblock the client. This allows up to set the promise CallReply as the private
data of the blocked client and abort it if the client gets disconnected.
* `RM_BlockClientGetPrivateData` - Return the current private data set on a blocked client.
We need it so we will have access to this private data on the disconnect callback.
* On RM_Call, the returned reply will be added to the auto memory context only if auto
memory is enabled, this allows us to keep the call reply for longer time then the context
lifetime and does not force an unneeded borrow relationship between the CallReply and
the RedisModuleContext.
2023-03-16 14:04:31 +02:00
|
|
|
void enterExecutionUnit(int update_cached_time, long long us);
|
2023-05-02 17:31:32 -07:00
|
|
|
void exitExecutionUnit(void);
|
2014-03-19 12:55:49 +01:00
|
|
|
void resetServerStats(void);
|
2016-12-30 03:37:52 +02:00
|
|
|
void activeDefragCycle(void);
|
2014-03-20 11:47:12 +01:00
|
|
|
unsigned int getLRUClock(void);
|
2017-05-09 11:57:09 +02:00
|
|
|
unsigned int LRU_CLOCK(void);
|
2016-04-22 10:43:48 -05:00
|
|
|
const char *evictPolicyToString(void);
|
2024-04-03 09:29:42 +07:00
|
|
|
struct serverMemOverhead *getMemoryOverheadData(void);
|
|
|
|
void freeMemoryOverheadData(struct serverMemOverhead *mh);
|
2019-08-11 16:07:53 +03:00
|
|
|
void checkChildrenDone(void);
|
2020-08-12 17:58:56 +03:00
|
|
|
int setOOMScoreAdj(int process_class);
|
Adds pub/sub channel patterns to ACL (#7993)
Fixes #7923.
This PR appropriates the special `&` symbol (because `@` and `*` are taken),
followed by a literal value or pattern for describing the Pub/Sub patterns that
an ACL user can interact with. It is similar to the existing key patterns
mechanism in function (additive) and implementation (copy-pasta). It also adds
the allchannels and resetchannels ACL keywords, naturally.
The default user is given allchannels permissions, whereas new users get
whatever is defined by the acl-pubsub-default configuration directive. For
backward compatibility in 6.2, the default of this directive is allchannels but
this is likely to be changed to resetchannels in the next major version for
stronger default security settings.
Unless allchannels is set for the user, channel access permissions are checked
as follows :
* Calls to both PUBLISH and SUBSCRIBE will fail unless a pattern matching the
argumentative channel name(s) exists for the user.
* Calls to PSUBSCRIBE will fail unless the pattern(s) provided as an argument
literally exist(s) in the user's list.
Such failures are logged to the ACL log.
Runtime changes to channel permissions for a user with existing subscribing
clients cause said clients to disconnect unless the new permissions permit the
connections to continue. Note, however, that PSUBSCRIBErs' patterns are matched
literally, so given the change bar:* -> b*, pattern subscribers to bar:* will be
disconnected.
Notes/questions:
* UNSUBSCRIBE, PUNSUBSCRIBE and PUBSUB remain unprotected due to lack of reasons
for touching them.
2020-12-01 14:21:39 +02:00
|
|
|
void rejectCommandFormat(client *c, const char *fmt, ...);
|
2020-12-13 09:56:01 +02:00
|
|
|
void *activeDefragAlloc(void *ptr);
|
2024-05-22 23:24:12 -07:00
|
|
|
robj *activeDefragStringOb(robj *ob);
|
Use madvise(MADV_DONTNEED) to release memory to reduce COW (#8974)
## Backgroud
As we know, after `fork`, one process will copy pages when writing data to these
pages(CoW), and another process still keep old pages, they totally cost more memory.
For redis, we suffered that redis consumed much memory when the fork child is serializing
key/values, even that maybe cause OOM.
But actually we find, in redis fork child process, the child process don't need to keep some
memory and parent process may write or update that, for example, child process will never
access the key-value that is serialized but users may update it in parent process.
So we think it may reduce COW if the child process release memory that it is not needed.
## Implementation
For releasing key value in child process, we may think we call `decrRefCount` to free memory,
but i find the fork child process still use much memory when we don't write any data to redis,
and it costs much more time that slows down bgsave. Maybe because memory allocator doesn't
really release memory to OS, and it may modify some inner data for this free operation, especially
when we free small objects.
Moreover, CoW is based on pages, so it is a easy way that we only free the memory bulk that is
not less than kernel page size. madvise(MADV_DONTNEED) can quickly release specified region
pages to OS bypassing memory allocator, and allocator still consider that this memory still is used
and don't change its inner data.
There are some buffers we can release in the fork child process:
- **Serialized key-values**
the fork child process never access serialized key-values, so we try to free them.
Because we only can release big bulk memory, and it is time consumed to iterate all
items/members/fields/entries of complex data type. So we decide to iterate them and
try to release them only when their average size of item/member/field/entry is more
than page size of OS.
- **Replication backlog**
Because replication backlog is a cycle buffer, it will be changed quickly if redis has heavy
write traffic, but in fork child process, we don't need to access that.
- **Client buffers**
If clients have requests during having the fork child process, clients' buffer also be changed
frequently. The memory includes client query buffer, output buffer, and client struct used memory.
To get child process peak private dirty memory, we need to count peak memory instead
of last used memory, because the child process may continue to release memory (since
COW used to only grow till now, the last was equivalent to the peak).
Also we're adding a new `current_cow_peak` info variable (to complement the existing
`current_cow_size`)
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-08-05 04:01:46 +08:00
|
|
|
void dismissSds(sds s);
|
2024-05-22 23:24:12 -07:00
|
|
|
void dismissMemory(void *ptr, size_t size_hint);
|
Use madvise(MADV_DONTNEED) to release memory to reduce COW (#8974)
## Backgroud
As we know, after `fork`, one process will copy pages when writing data to these
pages(CoW), and another process still keep old pages, they totally cost more memory.
For redis, we suffered that redis consumed much memory when the fork child is serializing
key/values, even that maybe cause OOM.
But actually we find, in redis fork child process, the child process don't need to keep some
memory and parent process may write or update that, for example, child process will never
access the key-value that is serialized but users may update it in parent process.
So we think it may reduce COW if the child process release memory that it is not needed.
## Implementation
For releasing key value in child process, we may think we call `decrRefCount` to free memory,
but i find the fork child process still use much memory when we don't write any data to redis,
and it costs much more time that slows down bgsave. Maybe because memory allocator doesn't
really release memory to OS, and it may modify some inner data for this free operation, especially
when we free small objects.
Moreover, CoW is based on pages, so it is a easy way that we only free the memory bulk that is
not less than kernel page size. madvise(MADV_DONTNEED) can quickly release specified region
pages to OS bypassing memory allocator, and allocator still consider that this memory still is used
and don't change its inner data.
There are some buffers we can release in the fork child process:
- **Serialized key-values**
the fork child process never access serialized key-values, so we try to free them.
Because we only can release big bulk memory, and it is time consumed to iterate all
items/members/fields/entries of complex data type. So we decide to iterate them and
try to release them only when their average size of item/member/field/entry is more
than page size of OS.
- **Replication backlog**
Because replication backlog is a cycle buffer, it will be changed quickly if redis has heavy
write traffic, but in fork child process, we don't need to access that.
- **Client buffers**
If clients have requests during having the fork child process, clients' buffer also be changed
frequently. The memory includes client query buffer, output buffer, and client struct used memory.
To get child process peak private dirty memory, we need to count peak memory instead
of last used memory, because the child process may continue to release memory (since
COW used to only grow till now, the last was equivalent to the peak).
Also we're adding a new `current_cow_peak` info variable (to complement the existing
`current_cow_size`)
Co-authored-by: Oran Agra <oran@redislabs.com>
2021-08-05 04:01:46 +08:00
|
|
|
void dismissMemoryInChild(void);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2015-10-13 10:58:08 +02:00
|
|
|
#define RESTART_SERVER_NONE 0
|
2024-05-22 23:24:12 -07:00
|
|
|
#define RESTART_SERVER_GRACEFULLY (1 << 0) /* Do proper shutdown. */
|
|
|
|
#define RESTART_SERVER_CONFIG_REWRITE (1 << 1) /* CONFIG REWRITE before restart.*/
|
2015-10-13 10:58:08 +02:00
|
|
|
int restartServer(int flags, mstime_t delay);
|
Replace cluster metadata with slot specific dictionaries (#11695)
This is an implementation of https://github.com/redis/redis/issues/10589 that eliminates 16 bytes per entry in cluster mode, that are currently used to create a linked list between entries in the same slot. Main idea is splitting main dictionary into 16k smaller dictionaries (one per slot), so we can perform all slot specific operations, such as iteration, without any additional info in the `dictEntry`. For Redis cluster, the expectation is that there will be a larger number of keys, so the fixed overhead of 16k dictionaries will be The expire dictionary is also split up so that each slot is logically decoupled, so that in subsequent revisions we will be able to atomically flush a slot of data.
## Important changes
* Incremental rehashing - one big change here is that it's not one, but rather up to 16k dictionaries that can be rehashing at the same time, in order to keep track of them, we introduce a separate queue for dictionaries that are rehashing. Also instead of rehashing a single dictionary, cron job will now try to rehash as many as it can in 1ms.
* getRandomKey - now needs to not only select a random key, from the random bucket, but also needs to select a random dictionary. Fairness is a major concern here, as it's possible that keys can be unevenly distributed across the slots. In order to address this search we introduced binary index tree). With that data structure we are able to efficiently find a random slot using binary search in O(log^2(slot count)) time.
* Iteration efficiency - when iterating dictionary with a lot of empty slots, we want to skip them efficiently. We can do this using same binary index that is used for random key selection, this index allows us to find a slot for a specific key index. For example if there are 10 keys in the slot 0, then we can quickly find a slot that contains 11th key using binary search on top of the binary index tree.
* scan API - in order to perform a scan across the entire DB, the cursor now needs to not only save position within the dictionary but also the slot id. In this change we append slot id into LSB of the cursor so it can be passed around between client and the server. This has interesting side effect, now you'll be able to start scanning specific slot by simply providing slot id as a cursor value. The plan is to not document this as defined behavior, however. It's also worth nothing the SCAN API is now technically incompatible with previous versions, although practically we don't believe it's an issue.
* Checksum calculation optimizations - During command execution, we know that all of the keys are from the same slot (outside of a few notable exceptions such as cross slot scripts and modules). We don't want to compute the checksum multiple multiple times, hence we are relying on cached slot id in the client during the command executions. All operations that access random keys, either should pass in the known slot or recompute the slot.
* Slot info in RDB - in order to resize individual dictionaries correctly, while loading RDB, it's not enough to know total number of keys (of course we could approximate number of keys per slot, but it won't be precise). To address this issue, we've added additional metadata into RDB that contains number of keys in each slot, which can be used as a hint during loading.
* DB size - besides `DBSIZE` API, we need to know size of the DB in many places want, in order to avoid scanning all dictionaries and summing up their sizes in a loop, we've introduced a new field into `redisDb` that keeps track of `key_count`. This way we can keep DBSIZE operation O(1). This is also kept for O(1) expires computation as well.
## Performance
This change improves SET performance in cluster mode by ~5%, most of the gains come from us not having to maintain linked lists for keys in slot, non-cluster mode has same performance. For workloads that rely on evictions, the performance is similar because of the extra overhead for finding keys to evict.
RDB loading performance is slightly reduced, as the slot of each key needs to be computed during the load.
## Interface changes
* Removed `overhead.hashtable.slot-to-keys` to `MEMORY STATS`
* Scan API will now require 64 bits to store the cursor, even on 32 bit systems, as the slot information will be stored.
* New RDB version to support the new op code for SLOT information.
---------
Co-authored-by: Vitaly Arbuzov <arvit@amazon.com>
Co-authored-by: Harkrishn Patro <harkrisp@amazon.com>
Co-authored-by: Roshan Khatri <rvkhatri@amazon.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
2023-10-14 23:58:26 -07:00
|
|
|
int getKeySlot(sds key);
|
|
|
|
int calculateKeySlot(sds key);
|
Refactor the per-slot dict-array db.c into a new kvstore data structure (#12822)
# Description
Gather most of the scattered `redisDb`-related code from the per-slot
dict PR (#11695) and turn it to a new data structure, `kvstore`. i.e.
it's a class that represents an array of dictionaries.
# Motivation
The main motivation is code cleanliness, the idea of using an array of
dictionaries is very well-suited to becoming a self-contained data
structure.
This allowed cleaning some ugly code, among others: loops that run twice
on the main dict and expires dict, and duplicate code for allocating and
releasing this data structure.
# Notes
1. This PR reverts the part of https://github.com/redis/redis/pull/12848
where the `rehashing` list is global (handling rehashing `dict`s is
under the responsibility of `kvstore`, and should not be managed by the
server)
2. This PR also replaces the type of `server.pubsubshard_channels` from
`dict**` to `kvstore` (original PR:
https://github.com/redis/redis/pull/12804). After that was done,
server.pubsub_channels was also chosen to be a `kvstore` (with only one
`dict`, which seems odd) just to make the code cleaner by making it the
same type as `server.pubsubshard_channels`, see
`pubsubtype.serverPubSubChannels`
3. the keys and expires kvstores are currenlty configured to allocate
the individual dicts only when the first key is added (unlike before, in
which they allocated them in advance), but they won't release them when
the last key is deleted.
Worth mentioning that due to the recent change the reply of DEBUG
HTSTATS changed, in case no keys were ever added to the db.
before:
```
127.0.0.1:6379> DEBUG htstats 9
[Dictionary HT]
Hash table 0 stats (main hash table):
No stats available for empty dictionaries
[Expires HT]
Hash table 0 stats (main hash table):
No stats available for empty dictionaries
```
after:
```
127.0.0.1:6379> DEBUG htstats 9
[Dictionary HT]
[Expires HT]
```
2024-02-05 22:21:35 +07:00
|
|
|
|
|
|
|
/* kvstore wrappers */
|
2024-04-03 10:02:43 +07:00
|
|
|
int dbExpand(serverDb *db, uint64_t db_size, int try_expand);
|
|
|
|
int dbExpandExpires(serverDb *db, uint64_t db_size, int try_expand);
|
|
|
|
dictEntry *dbFind(serverDb *db, void *key);
|
|
|
|
dictEntry *dbFindExpires(serverDb *db, void *key);
|
|
|
|
unsigned long long dbSize(serverDb *db);
|
|
|
|
unsigned long long dbScan(serverDb *db, unsigned long long cursor, dictScanFunction *scan_cb, void *privdata);
|
2015-10-13 10:58:08 +02:00
|
|
|
|
2010-07-02 19:57:12 +02:00
|
|
|
/* Set data type */
|
2023-05-08 16:11:20 -07:00
|
|
|
robj *setTypeCreate(sds value, size_t size_hint);
|
2015-07-31 18:01:23 +02:00
|
|
|
int setTypeAdd(robj *subject, sds value);
|
Fix zuiFind crash / RM_ScanKey hang on SET object listpack encoding (#11581)
In #11290, we added listpack encoding for SET object.
But forgot to support it in zuiFind, causes ZINTER, ZINTERSTORE,
ZINTERCARD, ZIDFF, ZDIFFSTORE to crash.
And forgot to support it in RM_ScanKey, causes it hang.
This PR add support SET listpack in zuiFind, and in RM_ScanKey.
And add tests for related commands to cover this case.
Other changes:
- There is no reason for zuiFind to go into the internals of the SET.
It can simply use setTypeIsMember and don't care about encoding.
- Remove the `#include "intset.h"` from server.h reduce the chance of
accidental intset API use.
- Move setTypeAddAux, setTypeRemoveAux and setTypeIsMemberAux
interfaces to the header.
- In scanGenericCommand, use setTypeInitIterator and setTypeNext
to handle OBJ_SET scan.
- In RM_ScanKey, improve hash scan mode, use lpGetValue like zset,
they can share code and better performance.
The zuiFind part fixes #11578
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2022-12-09 23:08:01 +08:00
|
|
|
int setTypeAddAux(robj *set, char *str, size_t len, int64_t llval, int str_is_sds);
|
2015-07-31 18:01:23 +02:00
|
|
|
int setTypeRemove(robj *subject, sds value);
|
Fix zuiFind crash / RM_ScanKey hang on SET object listpack encoding (#11581)
In #11290, we added listpack encoding for SET object.
But forgot to support it in zuiFind, causes ZINTER, ZINTERSTORE,
ZINTERCARD, ZIDFF, ZDIFFSTORE to crash.
And forgot to support it in RM_ScanKey, causes it hang.
This PR add support SET listpack in zuiFind, and in RM_ScanKey.
And add tests for related commands to cover this case.
Other changes:
- There is no reason for zuiFind to go into the internals of the SET.
It can simply use setTypeIsMember and don't care about encoding.
- Remove the `#include "intset.h"` from server.h reduce the chance of
accidental intset API use.
- Move setTypeAddAux, setTypeRemoveAux and setTypeIsMemberAux
interfaces to the header.
- In scanGenericCommand, use setTypeInitIterator and setTypeNext
to handle OBJ_SET scan.
- In RM_ScanKey, improve hash scan mode, use lpGetValue like zset,
they can share code and better performance.
The zuiFind part fixes #11578
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2022-12-09 23:08:01 +08:00
|
|
|
int setTypeRemoveAux(robj *set, char *str, size_t len, int64_t llval, int str_is_sds);
|
2015-07-31 18:01:23 +02:00
|
|
|
int setTypeIsMember(robj *subject, sds value);
|
Fix zuiFind crash / RM_ScanKey hang on SET object listpack encoding (#11581)
In #11290, we added listpack encoding for SET object.
But forgot to support it in zuiFind, causes ZINTER, ZINTERSTORE,
ZINTERCARD, ZIDFF, ZDIFFSTORE to crash.
And forgot to support it in RM_ScanKey, causes it hang.
This PR add support SET listpack in zuiFind, and in RM_ScanKey.
And add tests for related commands to cover this case.
Other changes:
- There is no reason for zuiFind to go into the internals of the SET.
It can simply use setTypeIsMember and don't care about encoding.
- Remove the `#include "intset.h"` from server.h reduce the chance of
accidental intset API use.
- Move setTypeAddAux, setTypeRemoveAux and setTypeIsMemberAux
interfaces to the header.
- In scanGenericCommand, use setTypeInitIterator and setTypeNext
to handle OBJ_SET scan.
- In RM_ScanKey, improve hash scan mode, use lpGetValue like zset,
they can share code and better performance.
The zuiFind part fixes #11578
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2022-12-09 23:08:01 +08:00
|
|
|
int setTypeIsMemberAux(robj *set, char *str, size_t len, int64_t llval, int str_is_sds);
|
2010-08-21 11:25:13 +02:00
|
|
|
setTypeIterator *setTypeInitIterator(robj *subject);
|
|
|
|
void setTypeReleaseIterator(setTypeIterator *si);
|
2022-11-09 18:50:07 +01:00
|
|
|
int setTypeNext(setTypeIterator *si, char **str, size_t *len, int64_t *llele);
|
2015-07-31 18:01:23 +02:00
|
|
|
sds setTypeNextObject(setTypeIterator *si);
|
2022-11-09 18:50:07 +01:00
|
|
|
int setTypeRandomElement(robj *setobj, char **str, size_t *len, int64_t *llele);
|
2016-06-20 23:08:06 +03:00
|
|
|
unsigned long setTypeSize(const robj *subject);
|
2010-07-02 19:57:12 +02:00
|
|
|
void setTypeConvert(robj *subject, int enc);
|
2022-12-06 10:25:51 +01:00
|
|
|
int setTypeConvertAndExpand(robj *setobj, int enc, unsigned long cap, int panic);
|
2020-11-17 19:03:05 +09:00
|
|
|
robj *setTypeDup(robj *o);
|
2010-07-02 19:57:12 +02:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Hash data type */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define HASH_SET_TAKE_FIELD (1 << 0)
|
|
|
|
#define HASH_SET_TAKE_VALUE (1 << 1)
|
2016-04-25 15:39:33 +02:00
|
|
|
#define HASH_SET_COPY 0
|
|
|
|
|
2012-01-02 22:14:10 -08:00
|
|
|
void hashTypeConvert(robj *o, int enc);
|
2010-06-22 00:07:48 +02:00
|
|
|
void hashTypeTryConversion(robj *subject, robj **argv, int start, int end);
|
2015-09-23 09:33:23 +02:00
|
|
|
int hashTypeExists(robj *o, sds key);
|
|
|
|
int hashTypeDelete(robj *o, sds key);
|
2016-06-20 23:08:06 +03:00
|
|
|
unsigned long hashTypeLength(const robj *o);
|
2010-06-22 00:07:48 +02:00
|
|
|
hashTypeIterator *hashTypeInitIterator(robj *subject);
|
|
|
|
void hashTypeReleaseIterator(hashTypeIterator *hi);
|
|
|
|
int hashTypeNext(hashTypeIterator *hi);
|
2024-05-22 23:24:12 -07:00
|
|
|
void hashTypeCurrentFromListpack(hashTypeIterator *hi,
|
|
|
|
int what,
|
2021-08-10 14:18:49 +08:00
|
|
|
unsigned char **vstr,
|
|
|
|
unsigned int *vlen,
|
|
|
|
long long *vll);
|
2015-09-23 09:33:23 +02:00
|
|
|
sds hashTypeCurrentFromHashTable(hashTypeIterator *hi, int what);
|
|
|
|
void hashTypeCurrentObject(hashTypeIterator *hi, int what, unsigned char **vstr, unsigned int *vlen, long long *vll);
|
|
|
|
sds hashTypeCurrentObjectNewSds(hashTypeIterator *hi, int what);
|
2015-07-26 15:20:46 +02:00
|
|
|
robj *hashTypeLookupWriteOrCreate(client *c, robj *key);
|
2015-09-23 09:33:23 +02:00
|
|
|
robj *hashTypeGetValueObject(robj *o, sds field);
|
2016-04-25 15:39:33 +02:00
|
|
|
int hashTypeSet(robj *o, sds field, sds value, int flags);
|
2020-11-17 19:03:05 +09:00
|
|
|
robj *hashTypeDup(robj *o);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* Pub / Sub */
|
2015-07-26 15:20:46 +02:00
|
|
|
int pubsubUnsubscribeAllChannels(client *c, int notify);
|
2022-01-03 01:54:47 +01:00
|
|
|
int pubsubUnsubscribeShardAllChannels(client *c, int notify);
|
Replace slots_to_channels radix tree with slot specific dictionaries for shard channels. (#12804)
We have achieved replacing `slots_to_keys` radix tree with key->slot
linked list (#9356), and then replacing the list with slot specific
dictionaries for keys (#11695).
Shard channels behave just like keys in many ways, and we also need a
slots->channels mapping. Currently this is still done by using a radix
tree. So we should split `server.pubsubshard_channels` into 16384 dicts
and drop the radix tree, just like what we did to DBs.
Some benefits (basically the benefits of what we've done to DBs):
1. Optimize counting channels in a slot. This is currently used only in
removing channels in a slot. But this is potentially more useful:
sometimes we need to know how many channels there are in a specific slot
when doing slot migration. Counting is now implemented by traversing the
radix tree, and with this PR it will be as simple as calling `dictSize`,
from O(n) to O(1).
2. The radix tree in the cluster has been removed. The shard channel
names no longer require additional storage, which can save memory.
3. Potentially useful in slot migration, as shard channels are logically
split by slots, thus making it easier to migrate, remove or add as a
whole.
4. Avoid rehashing a big dict when there is a large number of channels.
Drawbacks:
1. Takes more memory than using radix tree when there are relatively few
shard channels.
What this PR does:
1. in cluster mode, split `server.pubsubshard_channels` into 16384
dicts, in standalone mode, still use only one dict.
2. drop the `slots_to_channels` radix tree.
3. to save memory (to solve the drawback above), all 16384 dicts are
created lazily, which means only when a channel is about to be inserted
to the dict will the dict be initialized, and when all channels are
deleted, the dict would delete itself.
5. use `server.shard_channel_count` to keep track of the number of all
shard channels.
---------
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
2023-12-27 17:40:45 +08:00
|
|
|
void pubsubShardUnsubscribeAllChannelsInSlot(unsigned int slot);
|
2015-07-26 15:20:46 +02:00
|
|
|
int pubsubUnsubscribeAllPatterns(client *c, int notify);
|
2022-04-17 14:43:22 +02:00
|
|
|
int pubsubPublishMessage(robj *channel, robj *message, int sharded);
|
|
|
|
int pubsubPublishMessageAndPropagateToCluster(robj *channel, robj *message, int sharded);
|
2022-05-30 22:03:59 -07:00
|
|
|
void addReplyPubsubMessage(client *c, robj *channel, robj *msg, robj *message_bulk);
|
2023-05-02 17:31:32 -07:00
|
|
|
int serverPubsubSubscriptionCount(void);
|
|
|
|
int serverPubsubShardSubscriptionCount(void);
|
2022-07-03 23:18:57 -07:00
|
|
|
size_t pubsubMemOverhead(client *c);
|
2023-12-13 13:44:13 +08:00
|
|
|
void unmarkClientAsPubSub(client *c);
|
Refactor the per-slot dict-array db.c into a new kvstore data structure (#12822)
# Description
Gather most of the scattered `redisDb`-related code from the per-slot
dict PR (#11695) and turn it to a new data structure, `kvstore`. i.e.
it's a class that represents an array of dictionaries.
# Motivation
The main motivation is code cleanliness, the idea of using an array of
dictionaries is very well-suited to becoming a self-contained data
structure.
This allowed cleaning some ugly code, among others: loops that run twice
on the main dict and expires dict, and duplicate code for allocating and
releasing this data structure.
# Notes
1. This PR reverts the part of https://github.com/redis/redis/pull/12848
where the `rehashing` list is global (handling rehashing `dict`s is
under the responsibility of `kvstore`, and should not be managed by the
server)
2. This PR also replaces the type of `server.pubsubshard_channels` from
`dict**` to `kvstore` (original PR:
https://github.com/redis/redis/pull/12804). After that was done,
server.pubsub_channels was also chosen to be a `kvstore` (with only one
`dict`, which seems odd) just to make the code cleaner by making it the
same type as `server.pubsubshard_channels`, see
`pubsubtype.serverPubSubChannels`
3. the keys and expires kvstores are currenlty configured to allocate
the individual dicts only when the first key is added (unlike before, in
which they allocated them in advance), but they won't release them when
the last key is deleted.
Worth mentioning that due to the recent change the reply of DEBUG
HTSTATS changed, in case no keys were ever added to the db.
before:
```
127.0.0.1:6379> DEBUG htstats 9
[Dictionary HT]
Hash table 0 stats (main hash table):
No stats available for empty dictionaries
[Expires HT]
Hash table 0 stats (main hash table):
No stats available for empty dictionaries
```
after:
```
127.0.0.1:6379> DEBUG htstats 9
[Dictionary HT]
[Expires HT]
```
2024-02-05 22:21:35 +07:00
|
|
|
int pubsubTotalSubscriptions(void);
|
2024-03-04 22:56:50 +08:00
|
|
|
dict *getClientPubSubChannels(client *c);
|
|
|
|
dict *getClientPubSubShardChannels(client *c);
|
2013-01-25 13:19:08 +01:00
|
|
|
|
|
|
|
/* Keyspace events notification */
|
|
|
|
void notifyKeyspaceEvent(int type, char *event, robj *key, int dbid);
|
|
|
|
int keyspaceEventsStringToFlags(char *classes);
|
|
|
|
sds keyspaceEventsFlagsToString(int flags);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* Configuration */
|
Module Configurations (#10285)
This feature adds the ability to add four different types (Bool, Numeric,
String, Enum) of configurations to a module to be accessed via the redis
config file, and the CONFIG command.
**Configuration Names**:
We impose a restriction that a module configuration always starts with the
module name and contains a '.' followed by the config name. If a module passes
"config1" as the name to a register function, it will be registered as MODULENAME.config1.
**Configuration Persistence**:
Module Configurations exist only as long as a module is loaded. If a module is
unloaded, the configurations are removed.
There is now also a minimal core API for removal of standardConfig objects
from configs by name.
**Get and Set Callbacks**:
Storage of config values is owned by the module that registers them, and provides
callbacks for Redis to access and manipulate the values.
This is exposed through a GET and SET callback.
The get callback returns a typed value of the config to redis. The callback takes
the name of the configuration, and also a privdata pointer. Note that these only
take the CONFIGNAME portion of the config, not the entire MODULENAME.CONFIGNAME.
```
typedef RedisModuleString * (*RedisModuleConfigGetStringFunc)(const char *name, void *privdata);
typedef long long (*RedisModuleConfigGetNumericFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetBoolFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetEnumFunc)(const char *name, void *privdata);
```
Configs must also must specify a set callback, i.e. what to do on a CONFIG SET XYZ 123
or when loading configurations from cli/.conf file matching these typedefs. *name* is
again just the CONFIGNAME portion, *val* is the parsed value from the core,
*privdata* is the registration time privdata pointer, and *err* is for providing errors to a client.
```
typedef int (*RedisModuleConfigSetStringFunc)(const char *name, RedisModuleString *val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetNumericFunc)(const char *name, long long val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetBoolFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetEnumFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
```
Modules can also specify an optional apply callback that will be called after
value(s) have been set via CONFIG SET:
```
typedef int (*RedisModuleConfigApplyFunc)(RedisModuleCtx *ctx, void *privdata, RedisModuleString **err);
```
**Flags:**
We expose 7 new flags to the module, which are used as part of the config registration.
```
#define REDISMODULE_CONFIG_MODIFIABLE 0 /* This is the default for a module config. */
#define REDISMODULE_CONFIG_IMMUTABLE (1ULL<<0) /* Can this value only be set at startup? */
#define REDISMODULE_CONFIG_SENSITIVE (1ULL<<1) /* Does this value contain sensitive information */
#define REDISMODULE_CONFIG_HIDDEN (1ULL<<4) /* This config is hidden in `config get <pattern>` (used for tests/debugging) */
#define REDISMODULE_CONFIG_PROTECTED (1ULL<<5) /* Becomes immutable if enable-protected-configs is enabled. */
#define REDISMODULE_CONFIG_DENY_LOADING (1ULL<<6) /* This config is forbidden during loading. */
/* Numeric Specific Configs */
#define REDISMODULE_CONFIG_MEMORY (1ULL<<7) /* Indicates if this value can be set as a memory value */
```
**Module Registration APIs**:
```
int (*RedisModule_RegisterBoolConfig)(RedisModuleCtx *ctx, char *name, int default_val, unsigned int flags, RedisModuleConfigGetBoolFunc getfn, RedisModuleConfigSetBoolFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterNumericConfig)(RedisModuleCtx *ctx, const char *name, long long default_val, unsigned int flags, long long min, long long max, RedisModuleConfigGetNumericFunc getfn, RedisModuleConfigSetNumericFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterStringConfig)(RedisModuleCtx *ctx, const char *name, const char *default_val, unsigned int flags, RedisModuleConfigGetStringFunc getfn, RedisModuleConfigSetStringFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterEnumConfig)(RedisModuleCtx *ctx, const char *name, int default_val, unsigned int flags, const char **enum_values, const int *int_values, int num_enum_vals, RedisModuleConfigGetEnumFunc getfn, RedisModuleConfigSetEnumFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_LoadConfigs)(RedisModuleCtx *ctx);
```
The module name will be auto appended along with a "." to the front of the name of the config.
**What RM_Register[...]Config does**:
A RedisModule struct now keeps a list of ModuleConfig objects which look like:
```
typedef struct ModuleConfig {
sds name; /* Name of config without the module name appended to the front */
void *privdata; /* Optional data passed into the module config callbacks */
union get_fn { /* The get callback specificed by the module */
RedisModuleConfigGetStringFunc get_string;
RedisModuleConfigGetNumericFunc get_numeric;
RedisModuleConfigGetBoolFunc get_bool;
RedisModuleConfigGetEnumFunc get_enum;
} get_fn;
union set_fn { /* The set callback specified by the module */
RedisModuleConfigSetStringFunc set_string;
RedisModuleConfigSetNumericFunc set_numeric;
RedisModuleConfigSetBoolFunc set_bool;
RedisModuleConfigSetEnumFunc set_enum;
} set_fn;
RedisModuleConfigApplyFunc apply_fn;
RedisModule *module;
} ModuleConfig;
```
It also registers a standardConfig in the configs array, with a pointer to the
ModuleConfig object associated with it.
**What happens on a CONFIG GET/SET MODULENAME.MODULECONFIG:**
For CONFIG SET, we do the same parsing as is done in config.c and pass that
as the argument to the module set callback. For CONFIG GET, we call the
module get callback and return that value to config.c to return to a client.
**CONFIG REWRITE**:
Starting up a server with module configurations in a .conf file but no module load
directive will fail. The flip side is also true, specifying a module load and a bunch
of module configurations will load those configurations in using the module defined
set callbacks on a RM_LoadConfigs call. Configs being rewritten works the same
way as it does for standard configs, as the module has the ability to specify a
default value. If a module is unloaded with configurations specified in the .conf file
those configurations will be commented out from the .conf file on the next config rewrite.
**RM_LoadConfigs:**
`RedisModule_LoadConfigs(RedisModuleCtx *ctx);`
This last API is used to make configs available within the onLoad() after they have
been registered. The expected usage is that a module will register all of its configs,
then call LoadConfigs to trigger all of the set callbacks, and then can error out if any
of them were malformed. LoadConfigs will attempt to set all configs registered to
either a .conf file argument/loadex argument or their default value if an argument is
not specified. **LoadConfigs is a required function if configs are registered.
** Also note that LoadConfigs **does not** call the apply callbacks, but a module
can do that directly after the LoadConfigs call.
**New Command: MODULE LOADEX [CONFIG NAME VALUE] [ARGS ...]:**
This command provides the ability to provide startup context information to a module.
LOADEX stands for "load extended" similar to GETEX. Note that provided config
names need the full MODULENAME.MODULECONFIG name. Any additional
arguments a module might want are intended to be specified after ARGS.
Everything after ARGS is passed to onLoad as RedisModuleString **argv.
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: Madelyn Olson <matolson@amazon.com>
Co-authored-by: sundb <sundbcn@gmail.com>
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Yossi Gottlieb <yossigo@gmail.com>
2022-03-30 05:47:06 -07:00
|
|
|
/* Configuration Flags */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define MODIFIABLE_CONFIG \
|
|
|
|
0 /* This is the implied default for a standard \
|
|
|
|
* config, which is mutable. */
|
|
|
|
#define IMMUTABLE_CONFIG (1ULL << 0) /* Can this value only be set at startup? */
|
|
|
|
#define SENSITIVE_CONFIG (1ULL << 1) /* Does this value contain sensitive information */
|
|
|
|
#define DEBUG_CONFIG (1ULL << 2) /* Values that are useful for debugging. */
|
|
|
|
#define MULTI_ARG_CONFIG (1ULL << 3) /* This config receives multiple arguments. */
|
|
|
|
#define HIDDEN_CONFIG (1ULL << 4) /* This config is hidden in `config get <pattern>` (used for tests/debugging) */
|
|
|
|
#define PROTECTED_CONFIG (1ULL << 5) /* Becomes immutable if enable-protected-configs is enabled. */
|
|
|
|
#define DENY_LOADING_CONFIG (1ULL << 6) /* This config is forbidden during loading. */
|
|
|
|
#define ALIAS_CONFIG (1ULL << 7) /* For configs with multiple names, this flag is set on the alias. */
|
|
|
|
#define MODULE_CONFIG (1ULL << 8) /* This config is a module config */
|
|
|
|
#define VOLATILE_CONFIG \
|
|
|
|
(1ULL << 9) /* The config is a reference to the config data and not the config data itself (ex. \
|
|
|
|
* a file name containing more configuration like a tls key). In this case we want \
|
|
|
|
* to apply the configuration change even if the new config value is the same as \
|
|
|
|
* the old. */
|
|
|
|
|
|
|
|
#define INTEGER_CONFIG 0 /* No flags means a simple integer configuration */
|
|
|
|
#define MEMORY_CONFIG (1 << 0) /* Indicates if this value can be loaded as a memory value */
|
|
|
|
#define PERCENT_CONFIG (1 << 1) /* Indicates if this value can be loaded as a percent (and stored as a negative int) */
|
|
|
|
#define OCTAL_CONFIG (1 << 2) /* This value uses octal representation */
|
Module Configurations (#10285)
This feature adds the ability to add four different types (Bool, Numeric,
String, Enum) of configurations to a module to be accessed via the redis
config file, and the CONFIG command.
**Configuration Names**:
We impose a restriction that a module configuration always starts with the
module name and contains a '.' followed by the config name. If a module passes
"config1" as the name to a register function, it will be registered as MODULENAME.config1.
**Configuration Persistence**:
Module Configurations exist only as long as a module is loaded. If a module is
unloaded, the configurations are removed.
There is now also a minimal core API for removal of standardConfig objects
from configs by name.
**Get and Set Callbacks**:
Storage of config values is owned by the module that registers them, and provides
callbacks for Redis to access and manipulate the values.
This is exposed through a GET and SET callback.
The get callback returns a typed value of the config to redis. The callback takes
the name of the configuration, and also a privdata pointer. Note that these only
take the CONFIGNAME portion of the config, not the entire MODULENAME.CONFIGNAME.
```
typedef RedisModuleString * (*RedisModuleConfigGetStringFunc)(const char *name, void *privdata);
typedef long long (*RedisModuleConfigGetNumericFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetBoolFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetEnumFunc)(const char *name, void *privdata);
```
Configs must also must specify a set callback, i.e. what to do on a CONFIG SET XYZ 123
or when loading configurations from cli/.conf file matching these typedefs. *name* is
again just the CONFIGNAME portion, *val* is the parsed value from the core,
*privdata* is the registration time privdata pointer, and *err* is for providing errors to a client.
```
typedef int (*RedisModuleConfigSetStringFunc)(const char *name, RedisModuleString *val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetNumericFunc)(const char *name, long long val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetBoolFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetEnumFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
```
Modules can also specify an optional apply callback that will be called after
value(s) have been set via CONFIG SET:
```
typedef int (*RedisModuleConfigApplyFunc)(RedisModuleCtx *ctx, void *privdata, RedisModuleString **err);
```
**Flags:**
We expose 7 new flags to the module, which are used as part of the config registration.
```
#define REDISMODULE_CONFIG_MODIFIABLE 0 /* This is the default for a module config. */
#define REDISMODULE_CONFIG_IMMUTABLE (1ULL<<0) /* Can this value only be set at startup? */
#define REDISMODULE_CONFIG_SENSITIVE (1ULL<<1) /* Does this value contain sensitive information */
#define REDISMODULE_CONFIG_HIDDEN (1ULL<<4) /* This config is hidden in `config get <pattern>` (used for tests/debugging) */
#define REDISMODULE_CONFIG_PROTECTED (1ULL<<5) /* Becomes immutable if enable-protected-configs is enabled. */
#define REDISMODULE_CONFIG_DENY_LOADING (1ULL<<6) /* This config is forbidden during loading. */
/* Numeric Specific Configs */
#define REDISMODULE_CONFIG_MEMORY (1ULL<<7) /* Indicates if this value can be set as a memory value */
```
**Module Registration APIs**:
```
int (*RedisModule_RegisterBoolConfig)(RedisModuleCtx *ctx, char *name, int default_val, unsigned int flags, RedisModuleConfigGetBoolFunc getfn, RedisModuleConfigSetBoolFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterNumericConfig)(RedisModuleCtx *ctx, const char *name, long long default_val, unsigned int flags, long long min, long long max, RedisModuleConfigGetNumericFunc getfn, RedisModuleConfigSetNumericFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterStringConfig)(RedisModuleCtx *ctx, const char *name, const char *default_val, unsigned int flags, RedisModuleConfigGetStringFunc getfn, RedisModuleConfigSetStringFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterEnumConfig)(RedisModuleCtx *ctx, const char *name, int default_val, unsigned int flags, const char **enum_values, const int *int_values, int num_enum_vals, RedisModuleConfigGetEnumFunc getfn, RedisModuleConfigSetEnumFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_LoadConfigs)(RedisModuleCtx *ctx);
```
The module name will be auto appended along with a "." to the front of the name of the config.
**What RM_Register[...]Config does**:
A RedisModule struct now keeps a list of ModuleConfig objects which look like:
```
typedef struct ModuleConfig {
sds name; /* Name of config without the module name appended to the front */
void *privdata; /* Optional data passed into the module config callbacks */
union get_fn { /* The get callback specificed by the module */
RedisModuleConfigGetStringFunc get_string;
RedisModuleConfigGetNumericFunc get_numeric;
RedisModuleConfigGetBoolFunc get_bool;
RedisModuleConfigGetEnumFunc get_enum;
} get_fn;
union set_fn { /* The set callback specified by the module */
RedisModuleConfigSetStringFunc set_string;
RedisModuleConfigSetNumericFunc set_numeric;
RedisModuleConfigSetBoolFunc set_bool;
RedisModuleConfigSetEnumFunc set_enum;
} set_fn;
RedisModuleConfigApplyFunc apply_fn;
RedisModule *module;
} ModuleConfig;
```
It also registers a standardConfig in the configs array, with a pointer to the
ModuleConfig object associated with it.
**What happens on a CONFIG GET/SET MODULENAME.MODULECONFIG:**
For CONFIG SET, we do the same parsing as is done in config.c and pass that
as the argument to the module set callback. For CONFIG GET, we call the
module get callback and return that value to config.c to return to a client.
**CONFIG REWRITE**:
Starting up a server with module configurations in a .conf file but no module load
directive will fail. The flip side is also true, specifying a module load and a bunch
of module configurations will load those configurations in using the module defined
set callbacks on a RM_LoadConfigs call. Configs being rewritten works the same
way as it does for standard configs, as the module has the ability to specify a
default value. If a module is unloaded with configurations specified in the .conf file
those configurations will be commented out from the .conf file on the next config rewrite.
**RM_LoadConfigs:**
`RedisModule_LoadConfigs(RedisModuleCtx *ctx);`
This last API is used to make configs available within the onLoad() after they have
been registered. The expected usage is that a module will register all of its configs,
then call LoadConfigs to trigger all of the set callbacks, and then can error out if any
of them were malformed. LoadConfigs will attempt to set all configs registered to
either a .conf file argument/loadex argument or their default value if an argument is
not specified. **LoadConfigs is a required function if configs are registered.
** Also note that LoadConfigs **does not** call the apply callbacks, but a module
can do that directly after the LoadConfigs call.
**New Command: MODULE LOADEX [CONFIG NAME VALUE] [ARGS ...]:**
This command provides the ability to provide startup context information to a module.
LOADEX stands for "load extended" similar to GETEX. Note that provided config
names need the full MODULENAME.MODULECONFIG name. Any additional
arguments a module might want are intended to be specified after ARGS.
Everything after ARGS is passed to onLoad as RedisModuleString **argv.
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: Madelyn Olson <matolson@amazon.com>
Co-authored-by: sundb <sundbcn@gmail.com>
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Yossi Gottlieb <yossigo@gmail.com>
2022-03-30 05:47:06 -07:00
|
|
|
|
|
|
|
/* Enum Configs contain an array of configEnum objects that match a string with an integer. */
|
|
|
|
typedef struct configEnum {
|
|
|
|
char *name;
|
|
|
|
int val;
|
|
|
|
} configEnum;
|
|
|
|
|
|
|
|
/* Type of configuration. */
|
|
|
|
typedef enum {
|
|
|
|
BOOL_CONFIG,
|
|
|
|
NUMERIC_CONFIG,
|
|
|
|
STRING_CONFIG,
|
|
|
|
SDS_CONFIG,
|
|
|
|
ENUM_CONFIG,
|
|
|
|
SPECIAL_CONFIG,
|
|
|
|
} configType;
|
|
|
|
|
2020-10-11 13:43:23 +03:00
|
|
|
void loadServerConfig(char *filename, char config_from_stdin, char *options);
|
2010-06-22 00:07:48 +02:00
|
|
|
void appendServerSaveParams(time_t seconds, int changes);
|
2014-03-16 08:51:54 +09:00
|
|
|
void resetServerSaveParams(void);
|
2013-11-19 09:48:12 +01:00
|
|
|
struct rewriteConfigState; /* Forward declaration to export API. */
|
2023-04-04 10:53:57 +02:00
|
|
|
int rewriteConfigRewriteLine(struct rewriteConfigState *state, const char *option, sds line, int force);
|
Sentinel: Fix Config Dependency and Rewrite Sequence (#8271)
This commit fixes a well known and an annoying issue in Sentinel mode.
Cause of this issue:
Currently, Redis rewrite process works well in server mode, however in sentinel mode,
the sentinel config has variant semantics for different configurations, in example configuration
https://github.com/redis/redis/blob/unstable/sentinel.conf, we put comments on these.
However the rewrite process only treat the sentinel config as a single option. During rewrite
process, it will mess up with the lines and comments.
Approaches:
In order to solve this issue, we need to differentiate different subconfig options in sentinel separately,
for example, sentinel monitor <master-name> <ip> <redis-port> <quorum>
we can treat it as sentinel monitor option, instead of the sentinel option.
This commit also fixes the dependency issue when putting configurations in sentinel.conf.
For example before this commit,we must put
`sentinel monitor <master-name> <ip> <redis-port> <quorum>` before
`sentinel auth-pass <master-name> <password>` for a single master,
otherwise the server cannot start and will return error. This commit fixes this issue, as long as
the monitoring master was configured, no matter the sequence is, the sentinel can start and run properly.
2021-01-26 02:31:54 -05:00
|
|
|
void rewriteConfigMarkAsProcessed(struct rewriteConfigState *state, const char *option);
|
2022-02-04 17:39:51 +08:00
|
|
|
int rewriteConfig(char *path, int force_write);
|
2023-05-02 17:31:32 -07:00
|
|
|
void initConfigValues(void);
|
Module Configurations (#10285)
This feature adds the ability to add four different types (Bool, Numeric,
String, Enum) of configurations to a module to be accessed via the redis
config file, and the CONFIG command.
**Configuration Names**:
We impose a restriction that a module configuration always starts with the
module name and contains a '.' followed by the config name. If a module passes
"config1" as the name to a register function, it will be registered as MODULENAME.config1.
**Configuration Persistence**:
Module Configurations exist only as long as a module is loaded. If a module is
unloaded, the configurations are removed.
There is now also a minimal core API for removal of standardConfig objects
from configs by name.
**Get and Set Callbacks**:
Storage of config values is owned by the module that registers them, and provides
callbacks for Redis to access and manipulate the values.
This is exposed through a GET and SET callback.
The get callback returns a typed value of the config to redis. The callback takes
the name of the configuration, and also a privdata pointer. Note that these only
take the CONFIGNAME portion of the config, not the entire MODULENAME.CONFIGNAME.
```
typedef RedisModuleString * (*RedisModuleConfigGetStringFunc)(const char *name, void *privdata);
typedef long long (*RedisModuleConfigGetNumericFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetBoolFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetEnumFunc)(const char *name, void *privdata);
```
Configs must also must specify a set callback, i.e. what to do on a CONFIG SET XYZ 123
or when loading configurations from cli/.conf file matching these typedefs. *name* is
again just the CONFIGNAME portion, *val* is the parsed value from the core,
*privdata* is the registration time privdata pointer, and *err* is for providing errors to a client.
```
typedef int (*RedisModuleConfigSetStringFunc)(const char *name, RedisModuleString *val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetNumericFunc)(const char *name, long long val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetBoolFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetEnumFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
```
Modules can also specify an optional apply callback that will be called after
value(s) have been set via CONFIG SET:
```
typedef int (*RedisModuleConfigApplyFunc)(RedisModuleCtx *ctx, void *privdata, RedisModuleString **err);
```
**Flags:**
We expose 7 new flags to the module, which are used as part of the config registration.
```
#define REDISMODULE_CONFIG_MODIFIABLE 0 /* This is the default for a module config. */
#define REDISMODULE_CONFIG_IMMUTABLE (1ULL<<0) /* Can this value only be set at startup? */
#define REDISMODULE_CONFIG_SENSITIVE (1ULL<<1) /* Does this value contain sensitive information */
#define REDISMODULE_CONFIG_HIDDEN (1ULL<<4) /* This config is hidden in `config get <pattern>` (used for tests/debugging) */
#define REDISMODULE_CONFIG_PROTECTED (1ULL<<5) /* Becomes immutable if enable-protected-configs is enabled. */
#define REDISMODULE_CONFIG_DENY_LOADING (1ULL<<6) /* This config is forbidden during loading. */
/* Numeric Specific Configs */
#define REDISMODULE_CONFIG_MEMORY (1ULL<<7) /* Indicates if this value can be set as a memory value */
```
**Module Registration APIs**:
```
int (*RedisModule_RegisterBoolConfig)(RedisModuleCtx *ctx, char *name, int default_val, unsigned int flags, RedisModuleConfigGetBoolFunc getfn, RedisModuleConfigSetBoolFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterNumericConfig)(RedisModuleCtx *ctx, const char *name, long long default_val, unsigned int flags, long long min, long long max, RedisModuleConfigGetNumericFunc getfn, RedisModuleConfigSetNumericFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterStringConfig)(RedisModuleCtx *ctx, const char *name, const char *default_val, unsigned int flags, RedisModuleConfigGetStringFunc getfn, RedisModuleConfigSetStringFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterEnumConfig)(RedisModuleCtx *ctx, const char *name, int default_val, unsigned int flags, const char **enum_values, const int *int_values, int num_enum_vals, RedisModuleConfigGetEnumFunc getfn, RedisModuleConfigSetEnumFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_LoadConfigs)(RedisModuleCtx *ctx);
```
The module name will be auto appended along with a "." to the front of the name of the config.
**What RM_Register[...]Config does**:
A RedisModule struct now keeps a list of ModuleConfig objects which look like:
```
typedef struct ModuleConfig {
sds name; /* Name of config without the module name appended to the front */
void *privdata; /* Optional data passed into the module config callbacks */
union get_fn { /* The get callback specificed by the module */
RedisModuleConfigGetStringFunc get_string;
RedisModuleConfigGetNumericFunc get_numeric;
RedisModuleConfigGetBoolFunc get_bool;
RedisModuleConfigGetEnumFunc get_enum;
} get_fn;
union set_fn { /* The set callback specified by the module */
RedisModuleConfigSetStringFunc set_string;
RedisModuleConfigSetNumericFunc set_numeric;
RedisModuleConfigSetBoolFunc set_bool;
RedisModuleConfigSetEnumFunc set_enum;
} set_fn;
RedisModuleConfigApplyFunc apply_fn;
RedisModule *module;
} ModuleConfig;
```
It also registers a standardConfig in the configs array, with a pointer to the
ModuleConfig object associated with it.
**What happens on a CONFIG GET/SET MODULENAME.MODULECONFIG:**
For CONFIG SET, we do the same parsing as is done in config.c and pass that
as the argument to the module set callback. For CONFIG GET, we call the
module get callback and return that value to config.c to return to a client.
**CONFIG REWRITE**:
Starting up a server with module configurations in a .conf file but no module load
directive will fail. The flip side is also true, specifying a module load and a bunch
of module configurations will load those configurations in using the module defined
set callbacks on a RM_LoadConfigs call. Configs being rewritten works the same
way as it does for standard configs, as the module has the ability to specify a
default value. If a module is unloaded with configurations specified in the .conf file
those configurations will be commented out from the .conf file on the next config rewrite.
**RM_LoadConfigs:**
`RedisModule_LoadConfigs(RedisModuleCtx *ctx);`
This last API is used to make configs available within the onLoad() after they have
been registered. The expected usage is that a module will register all of its configs,
then call LoadConfigs to trigger all of the set callbacks, and then can error out if any
of them were malformed. LoadConfigs will attempt to set all configs registered to
either a .conf file argument/loadex argument or their default value if an argument is
not specified. **LoadConfigs is a required function if configs are registered.
** Also note that LoadConfigs **does not** call the apply callbacks, but a module
can do that directly after the LoadConfigs call.
**New Command: MODULE LOADEX [CONFIG NAME VALUE] [ARGS ...]:**
This command provides the ability to provide startup context information to a module.
LOADEX stands for "load extended" similar to GETEX. Note that provided config
names need the full MODULENAME.MODULECONFIG name. Any additional
arguments a module might want are intended to be specified after ARGS.
Everything after ARGS is passed to onLoad as RedisModuleString **argv.
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: Madelyn Olson <matolson@amazon.com>
Co-authored-by: sundb <sundbcn@gmail.com>
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Yossi Gottlieb <yossigo@gmail.com>
2022-03-30 05:47:06 -07:00
|
|
|
void removeConfig(sds name);
|
2023-05-02 17:31:32 -07:00
|
|
|
sds getConfigDebugInfo(void);
|
Protected configs and sensitive commands (#9920)
Block sensitive configs and commands by default.
* `enable-protected-configs` - block modification of configs with the new `PROTECTED_CONFIG` flag.
Currently we add this flag to `dbfilename`, and `dir` configs,
all of which are non-mutable configs that can set a file redis will write to.
* `enable-debug-command` - block the `DEBUG` command
* `enable-module-command` - block the `MODULE` command
These have a default value set to `no`, so that these features are not
exposed by default to client connections, and can only be set by modifying the config file.
Users can change each of these to either `yes` (allow all access), or `local` (allow access from
local TCP connections and unix domain connections)
Note that this is a **breaking change** (specifically the part about MODULE command being disabled by default).
I.e. we don't consider DEBUG command being blocked as an issue (people shouldn't have been using it),
and the few configs we protected are unlikely to have been set at runtime anyway.
On the other hand, it's likely to assume some users who use modules, load them from the config file anyway.
Note that's the whole point of this PR, for redis to be more secure by default and reduce the attack surface on
innocent users, so secure defaults will necessarily mean a breaking change.
2021-12-19 10:46:16 +02:00
|
|
|
int allowProtectedAction(int config, client *c);
|
2024-04-18 14:10:24 +02:00
|
|
|
void createSharedObjectsWithCompat(void);
|
2023-05-02 17:31:32 -07:00
|
|
|
void initServerClientMemUsageBuckets(void);
|
|
|
|
void freeServerClientMemUsageBuckets(void);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
Module Configurations (#10285)
This feature adds the ability to add four different types (Bool, Numeric,
String, Enum) of configurations to a module to be accessed via the redis
config file, and the CONFIG command.
**Configuration Names**:
We impose a restriction that a module configuration always starts with the
module name and contains a '.' followed by the config name. If a module passes
"config1" as the name to a register function, it will be registered as MODULENAME.config1.
**Configuration Persistence**:
Module Configurations exist only as long as a module is loaded. If a module is
unloaded, the configurations are removed.
There is now also a minimal core API for removal of standardConfig objects
from configs by name.
**Get and Set Callbacks**:
Storage of config values is owned by the module that registers them, and provides
callbacks for Redis to access and manipulate the values.
This is exposed through a GET and SET callback.
The get callback returns a typed value of the config to redis. The callback takes
the name of the configuration, and also a privdata pointer. Note that these only
take the CONFIGNAME portion of the config, not the entire MODULENAME.CONFIGNAME.
```
typedef RedisModuleString * (*RedisModuleConfigGetStringFunc)(const char *name, void *privdata);
typedef long long (*RedisModuleConfigGetNumericFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetBoolFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetEnumFunc)(const char *name, void *privdata);
```
Configs must also must specify a set callback, i.e. what to do on a CONFIG SET XYZ 123
or when loading configurations from cli/.conf file matching these typedefs. *name* is
again just the CONFIGNAME portion, *val* is the parsed value from the core,
*privdata* is the registration time privdata pointer, and *err* is for providing errors to a client.
```
typedef int (*RedisModuleConfigSetStringFunc)(const char *name, RedisModuleString *val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetNumericFunc)(const char *name, long long val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetBoolFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetEnumFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
```
Modules can also specify an optional apply callback that will be called after
value(s) have been set via CONFIG SET:
```
typedef int (*RedisModuleConfigApplyFunc)(RedisModuleCtx *ctx, void *privdata, RedisModuleString **err);
```
**Flags:**
We expose 7 new flags to the module, which are used as part of the config registration.
```
#define REDISMODULE_CONFIG_MODIFIABLE 0 /* This is the default for a module config. */
#define REDISMODULE_CONFIG_IMMUTABLE (1ULL<<0) /* Can this value only be set at startup? */
#define REDISMODULE_CONFIG_SENSITIVE (1ULL<<1) /* Does this value contain sensitive information */
#define REDISMODULE_CONFIG_HIDDEN (1ULL<<4) /* This config is hidden in `config get <pattern>` (used for tests/debugging) */
#define REDISMODULE_CONFIG_PROTECTED (1ULL<<5) /* Becomes immutable if enable-protected-configs is enabled. */
#define REDISMODULE_CONFIG_DENY_LOADING (1ULL<<6) /* This config is forbidden during loading. */
/* Numeric Specific Configs */
#define REDISMODULE_CONFIG_MEMORY (1ULL<<7) /* Indicates if this value can be set as a memory value */
```
**Module Registration APIs**:
```
int (*RedisModule_RegisterBoolConfig)(RedisModuleCtx *ctx, char *name, int default_val, unsigned int flags, RedisModuleConfigGetBoolFunc getfn, RedisModuleConfigSetBoolFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterNumericConfig)(RedisModuleCtx *ctx, const char *name, long long default_val, unsigned int flags, long long min, long long max, RedisModuleConfigGetNumericFunc getfn, RedisModuleConfigSetNumericFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterStringConfig)(RedisModuleCtx *ctx, const char *name, const char *default_val, unsigned int flags, RedisModuleConfigGetStringFunc getfn, RedisModuleConfigSetStringFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterEnumConfig)(RedisModuleCtx *ctx, const char *name, int default_val, unsigned int flags, const char **enum_values, const int *int_values, int num_enum_vals, RedisModuleConfigGetEnumFunc getfn, RedisModuleConfigSetEnumFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_LoadConfigs)(RedisModuleCtx *ctx);
```
The module name will be auto appended along with a "." to the front of the name of the config.
**What RM_Register[...]Config does**:
A RedisModule struct now keeps a list of ModuleConfig objects which look like:
```
typedef struct ModuleConfig {
sds name; /* Name of config without the module name appended to the front */
void *privdata; /* Optional data passed into the module config callbacks */
union get_fn { /* The get callback specificed by the module */
RedisModuleConfigGetStringFunc get_string;
RedisModuleConfigGetNumericFunc get_numeric;
RedisModuleConfigGetBoolFunc get_bool;
RedisModuleConfigGetEnumFunc get_enum;
} get_fn;
union set_fn { /* The set callback specified by the module */
RedisModuleConfigSetStringFunc set_string;
RedisModuleConfigSetNumericFunc set_numeric;
RedisModuleConfigSetBoolFunc set_bool;
RedisModuleConfigSetEnumFunc set_enum;
} set_fn;
RedisModuleConfigApplyFunc apply_fn;
RedisModule *module;
} ModuleConfig;
```
It also registers a standardConfig in the configs array, with a pointer to the
ModuleConfig object associated with it.
**What happens on a CONFIG GET/SET MODULENAME.MODULECONFIG:**
For CONFIG SET, we do the same parsing as is done in config.c and pass that
as the argument to the module set callback. For CONFIG GET, we call the
module get callback and return that value to config.c to return to a client.
**CONFIG REWRITE**:
Starting up a server with module configurations in a .conf file but no module load
directive will fail. The flip side is also true, specifying a module load and a bunch
of module configurations will load those configurations in using the module defined
set callbacks on a RM_LoadConfigs call. Configs being rewritten works the same
way as it does for standard configs, as the module has the ability to specify a
default value. If a module is unloaded with configurations specified in the .conf file
those configurations will be commented out from the .conf file on the next config rewrite.
**RM_LoadConfigs:**
`RedisModule_LoadConfigs(RedisModuleCtx *ctx);`
This last API is used to make configs available within the onLoad() after they have
been registered. The expected usage is that a module will register all of its configs,
then call LoadConfigs to trigger all of the set callbacks, and then can error out if any
of them were malformed. LoadConfigs will attempt to set all configs registered to
either a .conf file argument/loadex argument or their default value if an argument is
not specified. **LoadConfigs is a required function if configs are registered.
** Also note that LoadConfigs **does not** call the apply callbacks, but a module
can do that directly after the LoadConfigs call.
**New Command: MODULE LOADEX [CONFIG NAME VALUE] [ARGS ...]:**
This command provides the ability to provide startup context information to a module.
LOADEX stands for "load extended" similar to GETEX. Note that provided config
names need the full MODULENAME.MODULECONFIG name. Any additional
arguments a module might want are intended to be specified after ARGS.
Everything after ARGS is passed to onLoad as RedisModuleString **argv.
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: Madelyn Olson <matolson@amazon.com>
Co-authored-by: sundb <sundbcn@gmail.com>
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Yossi Gottlieb <yossigo@gmail.com>
2022-03-30 05:47:06 -07:00
|
|
|
/* Module Configuration */
|
|
|
|
typedef struct ModuleConfig ModuleConfig;
|
|
|
|
int performModuleConfigSetFromName(sds name, sds value, const char **err);
|
|
|
|
int performModuleConfigSetDefaultFromName(sds name, const char **err);
|
|
|
|
void addModuleBoolConfig(const char *module_name, const char *name, int flags, void *privdata, int default_val);
|
|
|
|
void addModuleStringConfig(const char *module_name, const char *name, int flags, void *privdata, sds default_val);
|
2024-05-22 23:24:12 -07:00
|
|
|
void addModuleEnumConfig(const char *module_name,
|
|
|
|
const char *name,
|
|
|
|
int flags,
|
|
|
|
void *privdata,
|
|
|
|
int default_val,
|
|
|
|
configEnum *enum_vals);
|
|
|
|
void addModuleNumericConfig(const char *module_name,
|
|
|
|
const char *name,
|
|
|
|
int flags,
|
|
|
|
void *privdata,
|
|
|
|
long long default_val,
|
|
|
|
int conf_flags,
|
|
|
|
long long lower,
|
|
|
|
long long upper);
|
Module Configurations (#10285)
This feature adds the ability to add four different types (Bool, Numeric,
String, Enum) of configurations to a module to be accessed via the redis
config file, and the CONFIG command.
**Configuration Names**:
We impose a restriction that a module configuration always starts with the
module name and contains a '.' followed by the config name. If a module passes
"config1" as the name to a register function, it will be registered as MODULENAME.config1.
**Configuration Persistence**:
Module Configurations exist only as long as a module is loaded. If a module is
unloaded, the configurations are removed.
There is now also a minimal core API for removal of standardConfig objects
from configs by name.
**Get and Set Callbacks**:
Storage of config values is owned by the module that registers them, and provides
callbacks for Redis to access and manipulate the values.
This is exposed through a GET and SET callback.
The get callback returns a typed value of the config to redis. The callback takes
the name of the configuration, and also a privdata pointer. Note that these only
take the CONFIGNAME portion of the config, not the entire MODULENAME.CONFIGNAME.
```
typedef RedisModuleString * (*RedisModuleConfigGetStringFunc)(const char *name, void *privdata);
typedef long long (*RedisModuleConfigGetNumericFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetBoolFunc)(const char *name, void *privdata);
typedef int (*RedisModuleConfigGetEnumFunc)(const char *name, void *privdata);
```
Configs must also must specify a set callback, i.e. what to do on a CONFIG SET XYZ 123
or when loading configurations from cli/.conf file matching these typedefs. *name* is
again just the CONFIGNAME portion, *val* is the parsed value from the core,
*privdata* is the registration time privdata pointer, and *err* is for providing errors to a client.
```
typedef int (*RedisModuleConfigSetStringFunc)(const char *name, RedisModuleString *val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetNumericFunc)(const char *name, long long val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetBoolFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
typedef int (*RedisModuleConfigSetEnumFunc)(const char *name, int val, void *privdata, RedisModuleString **err);
```
Modules can also specify an optional apply callback that will be called after
value(s) have been set via CONFIG SET:
```
typedef int (*RedisModuleConfigApplyFunc)(RedisModuleCtx *ctx, void *privdata, RedisModuleString **err);
```
**Flags:**
We expose 7 new flags to the module, which are used as part of the config registration.
```
#define REDISMODULE_CONFIG_MODIFIABLE 0 /* This is the default for a module config. */
#define REDISMODULE_CONFIG_IMMUTABLE (1ULL<<0) /* Can this value only be set at startup? */
#define REDISMODULE_CONFIG_SENSITIVE (1ULL<<1) /* Does this value contain sensitive information */
#define REDISMODULE_CONFIG_HIDDEN (1ULL<<4) /* This config is hidden in `config get <pattern>` (used for tests/debugging) */
#define REDISMODULE_CONFIG_PROTECTED (1ULL<<5) /* Becomes immutable if enable-protected-configs is enabled. */
#define REDISMODULE_CONFIG_DENY_LOADING (1ULL<<6) /* This config is forbidden during loading. */
/* Numeric Specific Configs */
#define REDISMODULE_CONFIG_MEMORY (1ULL<<7) /* Indicates if this value can be set as a memory value */
```
**Module Registration APIs**:
```
int (*RedisModule_RegisterBoolConfig)(RedisModuleCtx *ctx, char *name, int default_val, unsigned int flags, RedisModuleConfigGetBoolFunc getfn, RedisModuleConfigSetBoolFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterNumericConfig)(RedisModuleCtx *ctx, const char *name, long long default_val, unsigned int flags, long long min, long long max, RedisModuleConfigGetNumericFunc getfn, RedisModuleConfigSetNumericFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterStringConfig)(RedisModuleCtx *ctx, const char *name, const char *default_val, unsigned int flags, RedisModuleConfigGetStringFunc getfn, RedisModuleConfigSetStringFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_RegisterEnumConfig)(RedisModuleCtx *ctx, const char *name, int default_val, unsigned int flags, const char **enum_values, const int *int_values, int num_enum_vals, RedisModuleConfigGetEnumFunc getfn, RedisModuleConfigSetEnumFunc setfn, RedisModuleConfigApplyFunc applyfn, void *privdata);
int (*RedisModule_LoadConfigs)(RedisModuleCtx *ctx);
```
The module name will be auto appended along with a "." to the front of the name of the config.
**What RM_Register[...]Config does**:
A RedisModule struct now keeps a list of ModuleConfig objects which look like:
```
typedef struct ModuleConfig {
sds name; /* Name of config without the module name appended to the front */
void *privdata; /* Optional data passed into the module config callbacks */
union get_fn { /* The get callback specificed by the module */
RedisModuleConfigGetStringFunc get_string;
RedisModuleConfigGetNumericFunc get_numeric;
RedisModuleConfigGetBoolFunc get_bool;
RedisModuleConfigGetEnumFunc get_enum;
} get_fn;
union set_fn { /* The set callback specified by the module */
RedisModuleConfigSetStringFunc set_string;
RedisModuleConfigSetNumericFunc set_numeric;
RedisModuleConfigSetBoolFunc set_bool;
RedisModuleConfigSetEnumFunc set_enum;
} set_fn;
RedisModuleConfigApplyFunc apply_fn;
RedisModule *module;
} ModuleConfig;
```
It also registers a standardConfig in the configs array, with a pointer to the
ModuleConfig object associated with it.
**What happens on a CONFIG GET/SET MODULENAME.MODULECONFIG:**
For CONFIG SET, we do the same parsing as is done in config.c and pass that
as the argument to the module set callback. For CONFIG GET, we call the
module get callback and return that value to config.c to return to a client.
**CONFIG REWRITE**:
Starting up a server with module configurations in a .conf file but no module load
directive will fail. The flip side is also true, specifying a module load and a bunch
of module configurations will load those configurations in using the module defined
set callbacks on a RM_LoadConfigs call. Configs being rewritten works the same
way as it does for standard configs, as the module has the ability to specify a
default value. If a module is unloaded with configurations specified in the .conf file
those configurations will be commented out from the .conf file on the next config rewrite.
**RM_LoadConfigs:**
`RedisModule_LoadConfigs(RedisModuleCtx *ctx);`
This last API is used to make configs available within the onLoad() after they have
been registered. The expected usage is that a module will register all of its configs,
then call LoadConfigs to trigger all of the set callbacks, and then can error out if any
of them were malformed. LoadConfigs will attempt to set all configs registered to
either a .conf file argument/loadex argument or their default value if an argument is
not specified. **LoadConfigs is a required function if configs are registered.
** Also note that LoadConfigs **does not** call the apply callbacks, but a module
can do that directly after the LoadConfigs call.
**New Command: MODULE LOADEX [CONFIG NAME VALUE] [ARGS ...]:**
This command provides the ability to provide startup context information to a module.
LOADEX stands for "load extended" similar to GETEX. Note that provided config
names need the full MODULENAME.MODULECONFIG name. Any additional
arguments a module might want are intended to be specified after ARGS.
Everything after ARGS is passed to onLoad as RedisModuleString **argv.
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: Madelyn Olson <matolson@amazon.com>
Co-authored-by: sundb <sundbcn@gmail.com>
Co-authored-by: Madelyn Olson <34459052+madolson@users.noreply.github.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Yossi Gottlieb <yossigo@gmail.com>
2022-03-30 05:47:06 -07:00
|
|
|
void addModuleConfigApply(list *module_configs, ModuleConfig *module_config);
|
|
|
|
int moduleConfigApplyConfig(list *module_configs, const char **err, const char **err_arg_name);
|
|
|
|
int getModuleBoolConfig(ModuleConfig *module_config);
|
|
|
|
int setModuleBoolConfig(ModuleConfig *config, int val, const char **err);
|
|
|
|
sds getModuleStringConfig(ModuleConfig *module_config);
|
|
|
|
int setModuleStringConfig(ModuleConfig *config, sds strval, const char **err);
|
|
|
|
int getModuleEnumConfig(ModuleConfig *module_config);
|
|
|
|
int setModuleEnumConfig(ModuleConfig *config, int val, const char **err);
|
|
|
|
long long getModuleNumericConfig(ModuleConfig *module_config);
|
|
|
|
int setModuleNumericConfig(ModuleConfig *config, long long val, const char **err);
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* db.c -- Keyspace access API */
|
2024-04-03 10:02:43 +07:00
|
|
|
int removeExpire(serverDb *db, robj *key);
|
|
|
|
void deleteExpiredKeyAndPropagate(serverDb *db, robj *keyobj);
|
|
|
|
void propagateDeletion(serverDb *db, robj *key, int lazy);
|
|
|
|
int keyIsExpired(serverDb *db, robj *key);
|
|
|
|
long long getExpire(serverDb *db, robj *key);
|
|
|
|
void setExpire(client *c, serverDb *db, robj *key, long long when);
|
2020-07-10 10:02:37 +03:00
|
|
|
int checkAlreadyExpired(long long when);
|
2024-04-03 10:02:43 +07:00
|
|
|
robj *lookupKeyRead(serverDb *db, robj *key);
|
|
|
|
robj *lookupKeyWrite(serverDb *db, robj *key);
|
2015-07-26 15:20:46 +02:00
|
|
|
robj *lookupKeyReadOrReply(client *c, robj *key, robj *reply);
|
|
|
|
robj *lookupKeyWriteOrReply(client *c, robj *key, robj *reply);
|
2024-04-03 10:02:43 +07:00
|
|
|
robj *lookupKeyReadWithFlags(serverDb *db, robj *key, int flags);
|
|
|
|
robj *lookupKeyWriteWithFlags(serverDb *db, robj *key, int flags);
|
2016-12-16 09:02:50 +01:00
|
|
|
robj *objectCommandLookup(client *c, robj *key);
|
|
|
|
robj *objectCommandLookupOrReply(client *c, robj *key, robj *reply);
|
2024-05-22 23:24:12 -07:00
|
|
|
int objectSetLRUOrLFU(robj *val, long long lfu_freq, long long lru_idle, long long lru_clock, int lru_multiplier);
|
2016-06-14 15:33:59 +02:00
|
|
|
#define LOOKUP_NONE 0
|
2024-05-22 23:24:12 -07:00
|
|
|
#define LOOKUP_NOTOUCH (1 << 0) /* Don't update LRU. */
|
|
|
|
#define LOOKUP_NONOTIFY (1 << 1) /* Don't trigger keyspace event on key misses. */
|
|
|
|
#define LOOKUP_NOSTATS (1 << 2) /* Don't update keyspace hits/misses counters. */
|
|
|
|
#define LOOKUP_WRITE (1 << 3) /* Delete expired keys even in replicas. */
|
|
|
|
#define LOOKUP_NOEXPIRE (1 << 4) /* Avoid deleting lazy expired keys. */
|
|
|
|
#define LOOKUP_NOEFFECTS \
|
|
|
|
(LOOKUP_NONOTIFY | LOOKUP_NOSTATS | LOOKUP_NOTOUCH | LOOKUP_NOEXPIRE) /* Avoid any effects from fetching the key */
|
2021-11-28 10:26:28 +01:00
|
|
|
|
2024-04-03 10:02:43 +07:00
|
|
|
void dbAdd(serverDb *db, robj *key, robj *val);
|
|
|
|
int dbAddRDBLoad(serverDb *db, sds key, robj *val);
|
|
|
|
void dbReplaceValue(serverDb *db, robj *key, robj *val);
|
2021-11-03 14:12:33 +02:00
|
|
|
|
|
|
|
#define SETKEY_KEEPTTL 1
|
|
|
|
#define SETKEY_NO_SIGNAL 2
|
|
|
|
#define SETKEY_ALREADY_EXIST 4
|
|
|
|
#define SETKEY_DOESNT_EXIST 8
|
2023-05-28 10:58:29 +03:00
|
|
|
#define SETKEY_ADD_OR_UPDATE 16 /* Key most likely doesn't exists */
|
2024-04-03 10:02:43 +07:00
|
|
|
void setKey(client *c, serverDb *db, robj *key, robj *val, int flags);
|
|
|
|
robj *dbRandomKey(serverDb *db);
|
|
|
|
int dbGenericDelete(serverDb *db, robj *key, int async, int flags);
|
|
|
|
int dbSyncDelete(serverDb *db, robj *key);
|
|
|
|
int dbDelete(serverDb *db, robj *key);
|
|
|
|
robj *dbUnshareStringValue(serverDb *db, robj *key, robj *o);
|
2015-09-28 10:47:45 +02:00
|
|
|
|
2024-05-22 23:24:12 -07:00
|
|
|
#define EMPTYDB_NO_FLAGS 0 /* No flags. */
|
|
|
|
#define EMPTYDB_ASYNC (1 << 0) /* Reclaim memory in another thread. */
|
|
|
|
#define EMPTYDB_NOFUNCTIONS (1 << 1) /* Indicate not to flush the functions. */
|
|
|
|
long long emptyData(int dbnum, int flags, void(callback)(dict *));
|
|
|
|
long long emptyDbStructure(serverDb *dbarray, int dbnum, int async, void(callback)(dict *));
|
2019-11-03 17:35:35 +02:00
|
|
|
void flushAllDataAndResetRDB(int flags);
|
2023-05-02 17:31:32 -07:00
|
|
|
long long dbTotalServerKeyCount(void);
|
2024-04-03 10:02:43 +07:00
|
|
|
serverDb *initTempDb(void);
|
2024-05-22 23:24:12 -07:00
|
|
|
void discardTempDb(serverDb *tempDb, void(callback)(dict *));
|
2020-12-02 19:56:11 +08:00
|
|
|
|
2015-09-28 10:47:45 +02:00
|
|
|
|
2015-07-26 15:20:46 +02:00
|
|
|
int selectDb(client *c, int id);
|
2024-04-03 10:02:43 +07:00
|
|
|
void signalModifiedKey(client *c, serverDb *db, robj *key);
|
2020-12-23 19:13:12 -08:00
|
|
|
void signalFlushedDb(int dbid, int async);
|
Replace cluster metadata with slot specific dictionaries (#11695)
This is an implementation of https://github.com/redis/redis/issues/10589 that eliminates 16 bytes per entry in cluster mode, that are currently used to create a linked list between entries in the same slot. Main idea is splitting main dictionary into 16k smaller dictionaries (one per slot), so we can perform all slot specific operations, such as iteration, without any additional info in the `dictEntry`. For Redis cluster, the expectation is that there will be a larger number of keys, so the fixed overhead of 16k dictionaries will be The expire dictionary is also split up so that each slot is logically decoupled, so that in subsequent revisions we will be able to atomically flush a slot of data.
## Important changes
* Incremental rehashing - one big change here is that it's not one, but rather up to 16k dictionaries that can be rehashing at the same time, in order to keep track of them, we introduce a separate queue for dictionaries that are rehashing. Also instead of rehashing a single dictionary, cron job will now try to rehash as many as it can in 1ms.
* getRandomKey - now needs to not only select a random key, from the random bucket, but also needs to select a random dictionary. Fairness is a major concern here, as it's possible that keys can be unevenly distributed across the slots. In order to address this search we introduced binary index tree). With that data structure we are able to efficiently find a random slot using binary search in O(log^2(slot count)) time.
* Iteration efficiency - when iterating dictionary with a lot of empty slots, we want to skip them efficiently. We can do this using same binary index that is used for random key selection, this index allows us to find a slot for a specific key index. For example if there are 10 keys in the slot 0, then we can quickly find a slot that contains 11th key using binary search on top of the binary index tree.
* scan API - in order to perform a scan across the entire DB, the cursor now needs to not only save position within the dictionary but also the slot id. In this change we append slot id into LSB of the cursor so it can be passed around between client and the server. This has interesting side effect, now you'll be able to start scanning specific slot by simply providing slot id as a cursor value. The plan is to not document this as defined behavior, however. It's also worth nothing the SCAN API is now technically incompatible with previous versions, although practically we don't believe it's an issue.
* Checksum calculation optimizations - During command execution, we know that all of the keys are from the same slot (outside of a few notable exceptions such as cross slot scripts and modules). We don't want to compute the checksum multiple multiple times, hence we are relying on cached slot id in the client during the command executions. All operations that access random keys, either should pass in the known slot or recompute the slot.
* Slot info in RDB - in order to resize individual dictionaries correctly, while loading RDB, it's not enough to know total number of keys (of course we could approximate number of keys per slot, but it won't be precise). To address this issue, we've added additional metadata into RDB that contains number of keys in each slot, which can be used as a hint during loading.
* DB size - besides `DBSIZE` API, we need to know size of the DB in many places want, in order to avoid scanning all dictionaries and summing up their sizes in a loop, we've introduced a new field into `redisDb` that keeps track of `key_count`. This way we can keep DBSIZE operation O(1). This is also kept for O(1) expires computation as well.
## Performance
This change improves SET performance in cluster mode by ~5%, most of the gains come from us not having to maintain linked lists for keys in slot, non-cluster mode has same performance. For workloads that rely on evictions, the performance is similar because of the extra overhead for finding keys to evict.
RDB loading performance is slightly reduced, as the slot of each key needs to be computed during the load.
## Interface changes
* Removed `overhead.hashtable.slot-to-keys` to `MEMORY STATS`
* Scan API will now require 64 bits to store the cursor, even on 32 bit systems, as the slot information will be stored.
* New RDB version to support the new op code for SLOT information.
---------
Co-authored-by: Vitaly Arbuzov <arvit@amazon.com>
Co-authored-by: Harkrishn Patro <harkrisp@amazon.com>
Co-authored-by: Roshan Khatri <rvkhatri@amazon.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
2023-10-14 23:58:26 -07:00
|
|
|
void scanGenericCommand(client *c, robj *o, unsigned long long cursor);
|
|
|
|
int parseScanCursorOrReply(client *c, robj *o, unsigned long long *cursor);
|
2024-04-03 10:02:43 +07:00
|
|
|
int dbAsyncDelete(serverDb *db, robj *key);
|
|
|
|
void emptyDbAsync(serverDb *db);
|
2015-09-28 10:51:25 +02:00
|
|
|
size_t lazyfreeGetPendingObjectsCount(void);
|
2020-11-16 16:34:04 +08:00
|
|
|
size_t lazyfreeGetFreedObjectsCount(void);
|
2021-05-17 16:54:37 +03:00
|
|
|
void lazyfreeResetStats(void);
|
2021-06-16 14:45:49 +08:00
|
|
|
void freeObjAsync(robj *key, robj *obj, int dbid);
|
Replication backlog and replicas use one global shared replication buffer (#9166)
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
2021-10-25 14:24:31 +08:00
|
|
|
void freeReplicationBacklogRefMemAsync(list *blocks, rax *index);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2011-03-23 18:09:17 +01:00
|
|
|
/* API to get key arguments from commands */
|
2022-01-20 13:05:27 -08:00
|
|
|
#define GET_KEYSPEC_DEFAULT 0
|
2024-05-22 23:24:12 -07:00
|
|
|
#define GET_KEYSPEC_INCLUDE_NOT_KEYS (1 << 0) /* Consider 'fake' keys as keys */
|
|
|
|
#define GET_KEYSPEC_RETURN_PARTIAL (1 << 1) /* Return all keys that can be found */
|
|
|
|
|
|
|
|
int getKeysFromCommandWithSpecs(struct serverCommand *cmd,
|
|
|
|
robj **argv,
|
|
|
|
int argc,
|
|
|
|
int search_flags,
|
|
|
|
getKeysResult *result);
|
2022-01-20 13:05:27 -08:00
|
|
|
keyReference *getKeysPrepareResult(getKeysResult *result, int numkeys);
|
2024-04-03 23:54:33 +07:00
|
|
|
int getKeysFromCommand(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int doesCommandHaveKeys(struct serverCommand *cmd);
|
|
|
|
int getChannelsFromCommand(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int doesCommandHaveChannelsWithFlags(struct serverCommand *cmd, int flags);
|
2020-10-05 17:03:17 +03:00
|
|
|
void getKeysFreeResult(getKeysResult *result);
|
2024-05-22 23:24:12 -07:00
|
|
|
int sintercardGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int zunionInterDiffGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int zunionInterDiffStoreGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
2024-04-03 23:54:33 +07:00
|
|
|
int evalGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int functionGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int sortGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int sortROGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int migrateGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int georadiusGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int xreadGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int lmpopGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int blmpopGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int zmpopGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int bzmpopGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int setGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
|
|
|
int bitfieldGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result);
|
2011-03-23 18:09:17 +01:00
|
|
|
|
2011-03-29 17:51:15 +02:00
|
|
|
unsigned short crc16(const char *buf, int len);
|
|
|
|
|
2012-07-23 12:54:52 +02:00
|
|
|
/* Sentinel */
|
|
|
|
void initSentinelConfig(void);
|
|
|
|
void initSentinel(void);
|
|
|
|
void sentinelTimer(void);
|
2021-01-21 11:56:08 +02:00
|
|
|
const char *sentinelHandleConfiguration(char **argv, int argc);
|
Sentinel: Fix Config Dependency and Rewrite Sequence (#8271)
This commit fixes a well known and an annoying issue in Sentinel mode.
Cause of this issue:
Currently, Redis rewrite process works well in server mode, however in sentinel mode,
the sentinel config has variant semantics for different configurations, in example configuration
https://github.com/redis/redis/blob/unstable/sentinel.conf, we put comments on these.
However the rewrite process only treat the sentinel config as a single option. During rewrite
process, it will mess up with the lines and comments.
Approaches:
In order to solve this issue, we need to differentiate different subconfig options in sentinel separately,
for example, sentinel monitor <master-name> <ip> <redis-port> <quorum>
we can treat it as sentinel monitor option, instead of the sentinel option.
This commit also fixes the dependency issue when putting configurations in sentinel.conf.
For example before this commit,we must put
`sentinel monitor <master-name> <ip> <redis-port> <quorum>` before
`sentinel auth-pass <master-name> <password>` for a single master,
otherwise the server cannot start and will return error. This commit fixes this issue, as long as
the monitoring master was configured, no matter the sequence is, the sentinel can start and run properly.
2021-01-26 02:31:54 -05:00
|
|
|
void queueSentinelConfig(sds *argv, int argc, int linenum, sds line);
|
|
|
|
void loadSentinelConfigFromQueue(void);
|
2013-11-21 12:27:14 +01:00
|
|
|
void sentinelIsRunning(void);
|
2021-04-01 02:01:05 -04:00
|
|
|
void sentinelCheckConfigFile(void);
|
Treat subcommands as commands (#9504)
## Intro
The purpose is to allow having different flags/ACL categories for
subcommands (Example: CONFIG GET is ok-loading but CONFIG SET isn't)
We create a small command table for every command that has subcommands
and each subcommand has its own flags, etc. (same as a "regular" command)
This commit also unites the Redis and the Sentinel command tables
## Affected commands
CONFIG
Used to have "admin ok-loading ok-stale no-script"
Changes:
1. Dropped "ok-loading" in all except GET (this doesn't change behavior since
there were checks in the code doing that)
XINFO
Used to have "read-only random"
Changes:
1. Dropped "random" in all except CONSUMERS
XGROUP
Used to have "write use-memory"
Changes:
1. Dropped "use-memory" in all except CREATE and CREATECONSUMER
COMMAND
No changes.
MEMORY
Used to have "random read-only"
Changes:
1. Dropped "random" in PURGE and USAGE
ACL
Used to have "admin no-script ok-loading ok-stale"
Changes:
1. Dropped "admin" in WHOAMI, GENPASS, and CAT
LATENCY
No changes.
MODULE
No changes.
SLOWLOG
Used to have "admin random ok-loading ok-stale"
Changes:
1. Dropped "random" in RESET
OBJECT
Used to have "read-only random"
Changes:
1. Dropped "random" in ENCODING and REFCOUNT
SCRIPT
Used to have "may-replicate no-script"
Changes:
1. Dropped "may-replicate" in all except FLUSH and LOAD
CLIENT
Used to have "admin no-script random ok-loading ok-stale"
Changes:
1. Dropped "random" in all except INFO and LIST
2. Dropped "admin" in ID, TRACKING, CACHING, GETREDIR, INFO, SETNAME, GETNAME, and REPLY
STRALGO
No changes.
PUBSUB
No changes.
CLUSTER
Changes:
1. Dropped "admin in countkeysinslots, getkeysinslot, info, nodes, keyslot, myid, and slots
SENTINEL
No changes.
(note that DEBUG also fits, but we decided not to convert it since it's for
debugging and anyway undocumented)
## New sub-command
This commit adds another element to the per-command output of COMMAND,
describing the list of subcommands, if any (in the same structure as "regular" commands)
Also, it adds a new subcommand:
```
COMMAND LIST [FILTERBY (MODULE <module-name>|ACLCAT <cat>|PATTERN <pattern>)]
```
which returns a set of all commands (unless filters), but excluding subcommands.
## Module API
A new module API, RM_CreateSubcommand, was added, in order to allow
module writer to define subcommands
## ACL changes:
1. Now, that each subcommand is actually a command, each has its own ACL id.
2. The old mechanism of allowed_subcommands is redundant
(blocking/allowing a subcommand is the same as blocking/allowing a regular command),
but we had to keep it, to support the widespread usage of allowed_subcommands
to block commands with certain args, that aren't subcommands (e.g. "-select +select|0").
3. I have renamed allowed_subcommands to allowed_firstargs to emphasize the difference.
4. Because subcommands are commands in ACL too, you can now use "-" to block subcommands
(e.g. "+client -client|kill"), which wasn't possible in the past.
5. It is also possible to use the allowed_firstargs mechanism with subcommand.
For example: `+config -config|set +config|set|loglevel` will block all CONFIG SET except
for setting the log level.
6. All of the ACL changes above required some amount of refactoring.
## Misc
1. There are two approaches: Either each subcommand has its own function or all
subcommands use the same function, determining what to do according to argv[0].
For now, I took the former approaches only with CONFIG and COMMAND,
while other commands use the latter approach (for smaller blamelog diff).
2. Deleted memoryGetKeys: It is no longer needed because MEMORY USAGE now uses the "range" key spec.
4. Bugfix: GETNAME was missing from CLIENT's help message.
5. Sentinel and Redis now use the same table, with the same function pointer.
Some commands have a different implementation in Sentinel, so we redirect
them (these are ROLE, PUBLISH, and INFO).
6. Command stats now show the stats per subcommand (e.g. instead of stats just
for "config" you will have stats for "config|set", "config|get", etc.)
7. It is now possible to use COMMAND directly on subcommands:
COMMAND INFO CONFIG|GET (The pipeline syntax was inspired from ACL, and
can be used in functions lookupCommandBySds and lookupCommandByCString)
8. STRALGO is now a container command (has "help")
## Breaking changes:
1. Command stats now show the stats per subcommand (see (5) above)
2021-10-20 10:52:57 +02:00
|
|
|
void sentinelCommand(client *c);
|
|
|
|
void sentinelInfoCommand(client *c);
|
|
|
|
void sentinelPublishCommand(client *c);
|
|
|
|
void sentinelRoleCommand(client *c);
|
2012-07-23 12:54:52 +02:00
|
|
|
|
2024-04-09 01:24:03 -07:00
|
|
|
/* valkey-check-rdb & aof */
|
2017-07-10 13:38:23 +02:00
|
|
|
int redis_check_rdb(char *rdbfilename, FILE *fp);
|
|
|
|
int redis_check_rdb_main(int argc, char **argv, FILE *fp);
|
|
|
|
int redis_check_aof_main(int argc, char **argv);
|
2014-05-09 12:06:06 -04:00
|
|
|
|
2011-04-30 17:46:52 +02:00
|
|
|
/* Scripting */
|
2015-11-05 11:10:46 +01:00
|
|
|
void scriptingInit(int setup);
|
2015-11-13 09:31:01 +01:00
|
|
|
int ldbRemoveChild(pid_t pid);
|
|
|
|
void ldbKillForkedSessions(void);
|
2015-11-14 22:13:32 +01:00
|
|
|
int ldbPendingChildren(void);
|
2021-10-06 09:36:37 +03:00
|
|
|
void luaLdbLineHook(lua_State *lua, lua_Debug *ar);
|
Lua eval scripts first in first out LRU eviction (#13108)
In some cases, users will abuse lua eval. Each EVAL call generates
a new lua script, which is added to the lua interpreter and cached
to redis-server, consuming a large amount of memory over time.
Since EVAL is mostly the one that abuses the lua cache, and these
won't have pipeline issues (i.e. the script won't disappear
unexpectedly,
and cause errors like it would with SCRIPT LOAD and EVALSHA),
we implement a plain FIFO LRU eviction only for these (not for
scripts loaded with SCRIPT LOAD).
### Implementation notes:
When not abused we'll probably have less than 100 scripts, and when
abused we'll have many thousands. So we use a hard coded value of 500
scripts. And considering that we don't have many scripts, then unlike
keys, we don't need to worry about the memory usage of keeping a true
sorted LRU linked list. We compute the SHA of each script anyway,
and put the script in a dict, we can store a listNode there, and use
it for quick removal and re-insertion into an LRU list each time the
script is used.
### New interfaces:
At the same time, a new `evicted_scripts` field is added to
INFO, which represents the number of evicted eval scripts. Users
can check it to see if they are abusing EVAL.
### benchmark:
`./src/redis-benchmark -P 10 -n 1000000 -r 10000000000 eval "return
__rand_int__" 0`
The simple abuse of eval benchmark test that will create 1 million EVAL
scripts. The performance has been improved by 50%, and the max latency
has dropped from 500ms to 13ms (this may be caused by table expansion
inside Lua when the number of scripts is large). And in the INFO memory,
it used to consume 120MB (server cache) + 310MB (lua engine), but now
it only consumes 70KB (server cache) + 210KB (lua_engine) because of
the scripts eviction.
For non-abusive case of about 100 EVAL scripts, there's no noticeable
change in performance or memory usage.
### unlikely potentially breaking change:
in theory, a user can maybe load a
script with EVAL and then use EVALSHA to call it (by calculating the
SHA1 value on the client side), it could be that if we read the docs
carefully we'll realized it's a valid scenario, but we suppose it's
extremely rare. So it may happen that EVALSHA acts on a script created
by EVAL, and the script is evicted and EVALSHA returns a NOSCRIPT error.
that is if you have more than 500 scripts being used in the same
transaction / pipeline.
This solves the second point in #13102.
2024-03-13 14:27:41 +08:00
|
|
|
void freeLuaScriptsSync(dict *lua_scripts, list *lua_scripts_lru_list, lua_State *lua);
|
|
|
|
void freeLuaScriptsAsync(dict *lua_scripts, list *lua_scripts_lru_list, lua_State *lua);
|
Redis Function Libraries (#10004)
# Redis Function Libraries
This PR implements Redis Functions Libraries as describe on: https://github.com/redis/redis/issues/9906.
Libraries purpose is to provide a better code sharing between functions by allowing to create multiple
functions in a single command. Functions that were created together can safely share code between
each other without worrying about compatibility issues and versioning.
Creating a new library is done using 'FUNCTION LOAD' command (full API is described below)
This PR introduces a new struct called libraryInfo, libraryInfo holds information about a library:
* name - name of the library
* engine - engine used to create the library
* code - library code
* description - library description
* functions - the functions exposed by the library
When Redis gets the `FUNCTION LOAD` command it creates a new empty libraryInfo.
Redis passes the `CODE` to the relevant engine alongside the empty libraryInfo.
As a result, the engine will create one or more functions by calling 'libraryCreateFunction'.
The new funcion will be added to the newly created libraryInfo. So far Everything is happening
locally on the libraryInfo so it is easy to abort the operation (in case of an error) by simply
freeing the libraryInfo. After the library info is fully constructed we start the joining phase by
which we will join the new library to the other libraries currently exist on Redis.
The joining phase make sure there is no function collision and add the library to the
librariesCtx (renamed from functionCtx). LibrariesCtx is used all around the code in the exact
same way as functionCtx was used (with respect to RDB loading, replicatio, ...).
The only difference is that apart from function dictionary (maps function name to functionInfo
object), the librariesCtx contains also a libraries dictionary that maps library name to libraryInfo object.
## New API
### FUNCTION LOAD
`FUNCTION LOAD <ENGINE> <LIBRARY NAME> [REPLACE] [DESCRIPTION <DESCRIPTION>] <CODE>`
Create a new library with the given parameters:
* ENGINE - REPLACE Engine name to use to create the library.
* LIBRARY NAME - The new library name.
* REPLACE - If the library already exists, replace it.
* DESCRIPTION - Library description.
* CODE - Library code.
Return "OK" on success, or error on the following cases:
* Library name already taken and REPLACE was not used
* Name collision with another existing library (even if replace was uses)
* Library registration failed by the engine (usually compilation error)
## Changed API
### FUNCTION LIST
`FUNCTION LIST [LIBRARYNAME <LIBRARY NAME PATTERN>] [WITHCODE]`
Command was modified to also allow getting libraries code (so `FUNCTION INFO` command is no longer
needed and removed). In addition the command gets an option argument, `LIBRARYNAME` allows you to
only get libraries that match the given `LIBRARYNAME` pattern. By default, it returns all libraries.
### INFO MEMORY
Added number of libraries to `INFO MEMORY`
### Commands flags
`DENYOOM` flag was set on `FUNCTION LOAD` and `FUNCTION RESTORE`. We consider those commands
as commands that add new data to the dateset (functions are data) and so we want to disallows
to run those commands on OOM.
## Removed API
* FUNCTION CREATE - Decided on https://github.com/redis/redis/issues/9906
* FUNCTION INFO - Decided on https://github.com/redis/redis/issues/9899
## Lua engine changes
When the Lua engine gets the code given on `FUNCTION LOAD` command, it immediately runs it, we call
this run the loading run. Loading run is not a usual script run, it is not possible to invoke any
Redis command from within the load run.
Instead there is a new API provided by `library` object. The new API's:
* `redis.log` - behave the same as `redis.log`
* `redis.register_function` - register a new function to the library
The loading run purpose is to register functions using the new `redis.register_function` API.
Any attempt to use any other API will result in an error. In addition, the load run is has a time
limit of 500ms, error is raise on timeout and the entire operation is aborted.
### `redis.register_function`
`redis.register_function(<function_name>, <callback>, [<description>])`
This new API allows users to register a new function that will be linked to the newly created library.
This API can only be called during the load run (see definition above). Any attempt to use it outside
of the load run will result in an error.
The parameters pass to the API are:
* function_name - Function name (must be a Lua string)
* callback - Lua function object that will be called when the function is invokes using fcall/fcall_ro
* description - Function description, optional (must be a Lua string).
### Example
The following example creates a library called `lib` with 2 functions, `f1` and `f1`, returns 1 and 2 respectively:
```
local function f1(keys, args)
return 1
end
local function f2(keys, args)
return 2
end
redis.register_function('f1', f1)
redis.register_function('f2', f2)
```
Notice: Unlike `eval`, functions inside a library get the KEYS and ARGV as arguments to the
functions and not as global.
### Technical Details
On the load run we only want the user to be able to call a white list on API's. This way, in
the future, if new API's will be added, the new API's will not be available to the load run
unless specifically added to this white list. We put the while list on the `library` object and
make sure the `library` object is only available to the load run by using [lua_setfenv](https://www.lua.org/manual/5.1/manual.html#lua_setfenv) API. This API allows us to set
the `globals` of a function (and all the function it creates). Before starting the load run we
create a new fresh Lua table (call it `g`) that only contains the `library` API (we make sure
to set global protection on this table just like the general global protection already exists
today), then we use [lua_setfenv](https://www.lua.org/manual/5.1/manual.html#lua_setfenv)
to set `g` as the global table of the load run. After the load run finished we update `g`
metatable and set `__index` and `__newindex` functions to be `_G` (Lua default globals),
we also pop out the `library` object as we do not need it anymore.
This way, any function that was created on the load run (and will be invoke using `fcall`) will
see the default globals as it expected to see them and will not have the `library` API anymore.
An important outcome of this new approach is that now we can achieve a distinct global table
for each library (it is not yet like that but it is very easy to achieve it now). In the future we can
decide to remove global protection because global on different libraries will not collide or we
can chose to give different API to different libraries base on some configuration or input.
Notice that this technique was meant to prevent errors and was not meant to prevent malicious
user from exploit it. For example, the load run can still save the `library` object on some local
variable and then using in `fcall` context. To prevent such a malicious use, the C code also make
sure it is running in the right context and if not raise an error.
2022-01-06 13:39:38 +02:00
|
|
|
void freeFunctionsAsync(functionsLibCtx *lib_ctx);
|
2023-05-02 17:31:32 -07:00
|
|
|
int ldbIsEnabled(void);
|
2021-10-05 17:03:12 +03:00
|
|
|
void ldbLog(sds entry);
|
2024-06-06 16:40:55 -07:00
|
|
|
void ldbLogRespReply(char *reply);
|
2021-10-05 17:03:12 +03:00
|
|
|
void sha1hex(char *digest, char *script, size_t len);
|
2023-05-02 17:31:32 -07:00
|
|
|
unsigned long evalMemory(void);
|
2024-05-22 23:24:12 -07:00
|
|
|
dict *evalScriptsDict(void);
|
2023-05-02 17:31:32 -07:00
|
|
|
unsigned long evalScriptsMemory(void);
|
Expose script flags to processCommand for better handling (#10744)
The important part is that read-only scripts (not just EVAL_RO
and FCALL_RO, but also ones with `no-writes` executed by normal EVAL or
FCALL), will now be permitted to run during CLIENT PAUSE WRITE (unlike
before where only the _RO commands would be processed).
Other than that, some errors like OOM, READONLY, MASTERDOWN are now
handled by processCommand, rather than the command itself affects the
error string (and even error code in some cases), and command stats.
Besides that, now the `may-replicate` commands, PFCOUNT and PUBLISH, will
be considered `write` commands in scripts and will be blocked in all
read-only scripts just like other write commands.
They'll also be blocked in EVAL_RO (i.e. even for scripts without the
`no-writes` shebang flag.
This commit also hides the `may_replicate` flag from the COMMAND command
output. this is a **breaking change**.
background about may_replicate:
We don't want to expose a no-may-replicate flag or alike to scripts, since we
consider the may-replicate thing an internal concern of redis, that we may
some day get rid of.
In fact, the may-replicate flag was initially introduced to flag EVAL: since
we didn't know what it's gonna do ahead of execution, before function-flags
existed). PUBLISH and PFCOUNT, both of which because they have side effects
which may some day be fixed differently.
code changes:
The changes in eval.c are mostly code re-ordering:
- evalCalcFunctionName is extracted out of evalGenericCommand
- evalExtractShebangFlags is extracted luaCreateFunction
- evalGetCommandFlags is new code
2022-06-01 14:09:40 +03:00
|
|
|
uint64_t evalGetCommandFlags(client *c, uint64_t orig_flags);
|
|
|
|
uint64_t fcallGetCommandFlags(client *c, uint64_t orig_flags);
|
2023-05-02 17:31:32 -07:00
|
|
|
int isInsideYieldingLongCommand(void);
|
2011-04-30 17:46:52 +02:00
|
|
|
|
2022-02-11 21:58:05 +02:00
|
|
|
typedef struct luaScript {
|
|
|
|
uint64_t flags;
|
|
|
|
robj *body;
|
2024-05-22 23:24:12 -07:00
|
|
|
listNode *node; /* list node in lua_scripts_lru_list list. */
|
2022-02-11 21:58:05 +02:00
|
|
|
} luaScript;
|
2023-02-28 19:38:58 +02:00
|
|
|
/* Cache of recently used small arguments to avoid malloc calls. */
|
|
|
|
#define LUA_CMD_OBJCACHE_SIZE 32
|
|
|
|
#define LUA_CMD_OBJCACHE_MAX_LEN 64
|
2022-02-11 21:58:05 +02:00
|
|
|
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
/* Blocked clients API */
|
2013-12-03 17:43:53 +01:00
|
|
|
void processUnblockedClients(void);
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
void initClientBlockingState(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void blockClient(client *c, int btype);
|
Support for RM_Call on blocking commands (#11568)
Allow running blocking commands from within a module using `RM_Call`.
Today, when `RM_Call` is used, the fake client that is used to run command
is marked with `CLIENT_DENY_BLOCKING` flag. This flag tells the command
that it is not allowed to block the client and in case it needs to block, it must
fallback to some alternative (either return error or perform some default behavior).
For example, `BLPOP` fallback to simple `LPOP` if it is not allowed to block.
All the commands must respect the `CLIENT_DENY_BLOCKING` flag (including
module commands). When the command invocation finished, Redis asserts that
the client was not blocked.
This PR introduces the ability to call blocking command using `RM_Call` by
passing a callback that will be called when the client will get unblocked.
In order to do that, the user must explicitly say that he allow to perform blocking
command by passing a new format specifier argument, `K`, to the `RM_Call`
function. This new flag will tell Redis that it is allow to run blocking command
and block the client. In case the command got blocked, Redis will return a new
type of call reply (`REDISMODULE_REPLY_PROMISE`). This call reply indicates
that the command got blocked and the user can set the on_unblocked handler using
`RM_CallReplyPromiseSetUnblockHandler`.
When clients gets unblocked, it eventually reaches `processUnblockedClients` function.
This is where we check if the client is a fake module client and if it is, we call the unblock
callback instead of performing the usual unblock operations.
**Notice**: `RM_CallReplyPromiseSetUnblockHandler` must be called atomically
along side the command invocation (without releasing the Redis lock in between).
In addition, unlike other CallReply types, the promise call reply must be released
by the module when the Redis GIL is acquired.
The module can abort the execution on the blocking command (if it was not yet
executed) using `RM_CallReplyPromiseAbort`. the API will return `REDISMODULE_OK`
on success and `REDISMODULE_ERR` if the operation is already executed.
**Notice** that in case of misbehave module, Abort might finished successfully but the
operation will not really be aborted. This can only happened if the module do not respect
the disconnect callback of the blocked client.
For pure Redis commands this can not happened.
### Atomicity Guarantees
The API promise that the unblock handler will run atomically as an execution unit.
This means that all the operation performed on the unblock handler will be wrapped
with a multi exec transaction when replicated to the replica and AOF.
The API **do not** grantee any other atomicity properties such as when the unblock
handler will be called. This gives us the flexibility to strengthen the grantees (or not)
in the future if we will decide that we need a better guarantees.
That said, the implementation **does** provide a better guarantees when performing
pure Redis blocking command like `BLPOP`. In this case the unblock handler will run
atomically with the operation that got unblocked (for example, in case of `BLPOP`, the
unblock handler will run atomically with the `LPOP` operation that run when the command
got unblocked). This is an implementation detail that might be change in the future and the
module writer should not count on that.
### Calling blocking commands while running on script mode (`S`)
`RM_Call` script mode (`S`) was introduced on #0372. It is used for usecases where the
command that was invoked on `RM_Call` comes from a user input and we want to make
sure the user will not run dangerous commands like `shutdown`. Some command, such
as `BLPOP`, are marked with `NO_SCRIPT` flag, which means they will not be allowed on
script mode. Those commands are marked with `NO_SCRIPT` just because they are
blocking commands and not because they are dangerous. Now that we can run blocking
commands on RM_Call, there is no real reason not to allow such commands on script mode.
The underline problem is that the `NO_SCRIPT` flag is abused to also mark some of the
blocking commands (notice that those commands know not to block the client if it is not
allowed to do so, and have a fallback logic to such cases. So even if those commands
were not marked with `NO_SCRIPT` flag, it would not harm Redis, and today we can
already run those commands within multi exec).
In addition, not all blocking commands are marked with `NO_SCRIPT` flag, for example
`blmpop` are not marked and can run from within a script.
Those facts shows that there are some ambiguity about the meaning of the `NO_SCRIPT`
flag, and its not fully clear where it should be use.
The PR suggest that blocking commands should not be marked with `NO_SCRIPT` flag,
those commands should handle `CLIENT_DENY_BLOCKING` flag and only block when
it's safe (like they already does today). To achieve that, the PR removes the `NO_SCRIPT`
flag from the following commands:
* `blmove`
* `blpop`
* `brpop`
* `brpoplpush`
* `bzpopmax`
* `bzpopmin`
* `wait`
This might be considered a breaking change as now, on scripts, instead of getting
`command is not allowed from script` error, the user will get some fallback behavior
base on the command implementation. That said, the change matches the behavior
of scripts and multi exec with respect to those commands and allow running them on
`RM_Call` even when script mode is used.
### Additional RedisModule API and changes
* `RM_BlockClientSetPrivateData` - Set private data on the blocked client without the
need to unblock the client. This allows up to set the promise CallReply as the private
data of the blocked client and abort it if the client gets disconnected.
* `RM_BlockClientGetPrivateData` - Return the current private data set on a blocked client.
We need it so we will have access to this private data on the disconnect callback.
* On RM_Call, the returned reply will be added to the auto memory context only if auto
memory is enabled, this allows us to keep the call reply for longer time then the context
lifetime and does not force an unneeded borrow relationship between the CallReply and
the RedisModuleContext.
2023-03-16 14:04:31 +02:00
|
|
|
void unblockClient(client *c, int queue_for_reprocessing);
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
void unblockClientOnTimeout(client *c);
|
|
|
|
void unblockClientOnError(client *c, const char *err_str);
|
2018-09-03 18:39:18 +02:00
|
|
|
void queueClientForReprocessing(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void replyToBlockedClientTimedOut(client *c);
|
|
|
|
int getTimeoutFromObjectOrReply(client *c, robj *object, mstime_t *timeout, int unit);
|
Replication: disconnect blocked clients when switching to slave role.
Bug as old as Redis and blocking operations. It's hard to trigger since
only happens on instance role switch, but the results are quite bad
since an inconsistency between master and slave is created.
How to trigger the bug is a good description of the bug itself.
1. Client does "BLPOP mylist 0" in master.
2. Master is turned into slave, that replicates from New-Master.
3. Client does "LPUSH mylist foo" in New-Master.
4. New-Master propagates write to slave.
5. Slave receives the LPUSH, the blocked client get served.
Now Master "mylist" key has "foo", Slave "mylist" key is empty.
Highlights:
* At step "2" above, the client remains attached, basically escaping any
check performed during command dispatch: read only slave, in that case.
* At step "5" the slave (that was the master), serves the blocked client
consuming a list element, which is not consumed on the master side.
This scenario is technically likely to happen during failovers, however
since Redis Sentinel already disconnects clients using the CLIENT
command when changing the role of the instance, the bug is avoided in
Sentinel deployments.
Closes #2473.
2015-03-24 16:00:09 +01:00
|
|
|
void disconnectAllBlockedClients(void);
|
2017-09-06 15:43:28 +02:00
|
|
|
void handleClientsBlockedOnKeys(void);
|
2024-04-03 10:02:43 +07:00
|
|
|
void signalKeyAsReady(serverDb *db, robj *key, int type);
|
reprocess command when client is unblocked on keys (#11012)
*TL;DR*
---------------------------------------
Following the discussion over the issue [#7551](https://github.com/redis/redis/issues/7551)
We decided to refactor the client blocking code to eliminate some of the code duplications
and to rebuild the infrastructure better for future key blocking cases.
*In this PR*
---------------------------------------
1. reprocess the command once a client becomes unblocked on key (instead of running
custom code for the unblocked path that's different than the one that would have run if
blocking wasn't needed)
2. eliminate some (now) irrelevant code for handling unblocking lists/zsets/streams etc...
3. modify some tests to intercept the error in cases of error on reprocess after unblock (see
details in the notes section below)
4. replace '$' on the client argv with current stream id. Since once we reprocess the stream
XREAD we need to read from the last msg and not wait for new msg in order to prevent
endless block loop.
5. Added statistics to the info "Clients" section to report the:
* `total_blocking_keys` - number of blocking keys
* `total_blocking_keys_on_nokey` - number of blocking keys which have at least 1 client
which would like
to be unblocked on when the key is deleted.
6. Avoid expiring unblocked key during unblock. Previously we used to lookup the unblocked key
which might have been expired during the lookup. Now we lookup the key using NOTOUCH and
NOEXPIRE to avoid deleting it at this point, so propagating commands in blocked.c is no longer needed.
7. deprecated command flags. We decided to remove the CMD_CALL_STATS and CMD_CALL_SLOWLOG
and make an explicit verification in the call() function in order to decide if stats update should take place.
This should simplify the logic and also mitigate existing issues: for example module calls which are
triggered as part of AOF loading might still report stats even though they are called during AOF loading.
*Behavior changes*
---------------------------------------------------
1. As this implementation prevents writing dedicated code handling unblocked streams/lists/zsets,
since we now re-process the command once the client is unblocked some errors will be reported differently.
The old implementation used to issue
``UNBLOCKED the stream key no longer exists``
in the following cases:
- The stream key has been deleted (ie. calling DEL)
- The stream and group existed but the key type was changed by overriding it (ie. with set command)
- The key not longer exists after we swapdb with a db which does not contains this key
- After swapdb when the new db has this key but with different type.
In the new implementation the reported errors will be the same as if the command was processed after effect:
**NOGROUP** - in case key no longer exists, or **WRONGTYPE** in case the key was overridden with a different type.
2. Reprocessing the command means that some checks will be reevaluated once the
client is unblocked.
For example, ACL rules might change since the command originally was executed and
will fail once the client is unblocked.
Another example is OOM condition checks which might enable the command to run and
block but fail the command reprocess once the client is unblocked.
3. One of the changes in this PR is that no command stats are being updated once the
command is blocked (all stats will be updated once the client is unblocked). This implies
that when we have many clients blocked, users will no longer be able to get that information
from the command stats. However the information can still be gathered from the client list.
**Client blocking**
---------------------------------------------------
the blocking on key will still be triggered the same way as it is done today.
in order to block the current client on list of keys, the call to
blockForKeys will still need to be made which will perform the same as it is today:
* add the client to the list of blocked clients on each key
* keep the key with a matching list node (position in the global blocking clients list for that key)
in the client private blocking key dict.
* flag the client with CLIENT_BLOCKED
* update blocking statistics
* register the client on the timeout table
**Key Unblock**
---------------------------------------------------
Unblocking a specific key will be triggered (same as today) by calling signalKeyAsReady.
the implementation in that part will stay the same as today - adding the key to the global readyList.
The reason to maintain the readyList (as apposed to iterating over all clients blocked on the specific key)
is in order to keep the signal operation as short as possible, since it is called during the command processing.
The main change is that instead of going through a dedicated code path that operates the blocked command
we will just call processPendingCommandsAndResetClient.
**ClientUnblock (keys)**
---------------------------------------------------
1. Unblocking clients on keys will be triggered after command is
processed and during the beforeSleep
8. the general schema is:
9. For each key *k* in the readyList:
```
For each client *c* which is blocked on *k*:
in case either:
1. *k* exists AND the *k* type matches the current client blocking type
OR
2. *k* exists and *c* is blocked on module command
OR
3. *k* does not exists and *c* was blocked with the flag
unblock_on_deleted_key
do:
1. remove the client from the list of clients blocked on this key
2. remove the blocking list node from the client blocking key dict
3. remove the client from the timeout list
10. queue the client on the unblocked_clients list
11. *NEW*: call processCommandAndResetClient(c);
```
*NOTE:* for module blocked clients we will still call the moduleUnblockClientByHandle
which will queue the client for processing in moduleUnblockedClients list.
**Process Unblocked clients**
---------------------------------------------------
The process of all unblocked clients is done in the beforeSleep and no change is planned
in that part.
The general schema will be:
For each client *c* in server.unblocked_clients:
* remove client from the server.unblocked_clients
* set back the client readHandler
* continue processing the pending command and input buffer.
*Some notes regarding the new implementation*
---------------------------------------------------
1. Although it was proposed, it is currently difficult to remove the
read handler from the client while it is blocked.
The reason is that a blocked client should be unblocked when it is
disconnected, or we might consume data into void.
2. While this PR mainly keep the current blocking logic as-is, there
might be some future additions to the infrastructure that we would
like to have:
- allow non-preemptive blocking of client - sometimes we can think
that a new kind of blocking can be expected to not be preempt. for
example lets imagine we hold some keys on disk and when a command
needs to process them it will block until the keys are uploaded.
in this case we will want the client to not disconnect or be
unblocked until the process is completed (remove the client read
handler, prevent client timeout, disable unblock via debug command etc...).
- allow generic blocking based on command declared keys - we might
want to add a hook before command processing to check if any of the
declared keys require the command to block. this way it would be
easier to add new kinds of key-based blocking mechanisms.
Co-authored-by: Oran Agra <oran@redislabs.com>
Signed-off-by: Ran Shidlansik <ranshid@amazon.com>
2023-01-01 23:35:42 +02:00
|
|
|
void blockForKeys(client *c, int btype, robj **keys, int numkeys, mstime_t timeout, int unblock_on_nokey);
|
|
|
|
void blockClientShutdown(client *c);
|
|
|
|
void blockPostponeClient(client *c);
|
2024-05-30 23:45:47 -07:00
|
|
|
void blockClientForReplicaAck(client *c, mstime_t timeout, long long offset, long numreplicas, int numlocal);
|
2024-06-07 14:21:33 -07:00
|
|
|
void replicationRequestAckFromReplicas(void);
|
2024-04-03 10:02:43 +07:00
|
|
|
void signalDeletedKeyAsReady(serverDb *db, robj *key, int type);
|
2022-02-21 11:20:41 +02:00
|
|
|
void updateStatsOnUnblock(client *c, long blocked_us, long reply_us, int had_errors);
|
2024-04-03 10:02:43 +07:00
|
|
|
void scanDatabaseForDeletedKeys(serverDb *emptied, serverDb *replaced_with);
|
2024-05-22 23:24:12 -07:00
|
|
|
void totalNumberOfStatefulKeys(unsigned long *blocking_keys,
|
|
|
|
unsigned long *blocking_keys_on_nokey,
|
|
|
|
unsigned long *watched_keys);
|
Modules: Unblock from within a timer coverage (#12337)
Apart from adding the missing coverage, this PR also adds `blockedBeforeSleep`
that gathers all block-related functions from `beforeSleep`
The order inside `blockedBeforeSleep` is different: now `handleClientsBlockedOnKeys`
(which may unblock clients) is called before `processUnblockedClients` (which handles
unblocked clients).
It makes sense to have this order.
There are no visible effects of the wrong ordering, except some cleanups of the now-unblocked
client would have happen in the next `beforeSleep` (will now happen in the current one)
The reason we even got into it is because i triggers an assertion in logresreq.c (breaking
the assumption that `unblockClient` is called **before** actually flushing the reply to the socket):
`handleClientsBlockedOnKeys` is called, then it calls `moduleUnblockClientOnKey`, which calls
`moduleUnblockClient`, which adds the client to `moduleUnblockedClients` back to `beforeSleep`,
we call `handleClientsWithPendingWritesUsingThreads`, it writes the data of buf to the client, so
`client->bufpos` became 0
On the next `beforeSleep`, we call `moduleHandleBlockedClients`, which calls `unblockClient`,
which calls `reqresAppendResponse`, triggering the assert. (because the `bufpos` is 0) - see https://github.com/redis/redis/pull/12301#discussion_r1226386716
2023-06-22 22:15:16 +02:00
|
|
|
void blockedBeforeSleep(void);
|
2020-03-27 16:34:45 +01:00
|
|
|
|
|
|
|
/* timeout.c -- Blocked clients timeout and connections timeout. */
|
2020-03-26 16:02:26 +01:00
|
|
|
void addClientToTimeoutTable(client *c);
|
2020-03-27 11:13:38 +01:00
|
|
|
void removeClientFromTimeoutTable(client *c);
|
2020-03-27 16:34:45 +01:00
|
|
|
void handleBlockedClientsTimeout(void);
|
|
|
|
int clientsCronHandleTimeout(client *c, mstime_t now_ms);
|
2013-12-03 17:43:53 +01:00
|
|
|
|
2016-07-06 15:24:06 +02:00
|
|
|
/* expire.c -- Handling of expired keys */
|
|
|
|
void activeExpireCycle(int type);
|
2024-06-07 14:21:33 -07:00
|
|
|
void expireReplicaKeys(void);
|
|
|
|
void rememberReplicaKeyWithExpire(serverDb *db, robj *key);
|
|
|
|
void flushReplicaKeysWithExpireList(void);
|
|
|
|
size_t getReplicaKeyWithExpireCount(void);
|
2016-07-06 15:24:06 +02:00
|
|
|
|
2016-07-07 15:01:58 +02:00
|
|
|
/* evict.c -- maxmemory handling and LRU eviction. */
|
2016-07-13 10:45:37 +02:00
|
|
|
void evictionPoolAlloc(void);
|
2016-07-15 12:12:52 +02:00
|
|
|
#define LFU_INIT_VAL 5
|
|
|
|
unsigned long LFUGetTimeInMinutes(void);
|
|
|
|
uint8_t LFULogIncr(uint8_t value);
|
2017-10-15 20:17:55 +08:00
|
|
|
unsigned long LFUDecrAndReturn(robj *o);
|
2020-09-15 23:16:01 -07:00
|
|
|
#define EVICT_OK 0
|
|
|
|
#define EVICT_RUNNING 1
|
|
|
|
#define EVICT_FAIL 2
|
|
|
|
int performEvictions(void);
|
2022-01-04 19:08:10 +08:00
|
|
|
void startEvictionTimeProc(void);
|
2016-07-07 15:01:58 +02:00
|
|
|
|
2016-12-13 16:27:13 +01:00
|
|
|
/* Keys hashing / comparison functions for dict.c hash tables. */
|
2017-02-20 16:09:54 +01:00
|
|
|
uint64_t dictSdsHash(const void *key);
|
2021-06-16 06:35:13 +03:00
|
|
|
uint64_t dictSdsCaseHash(const void *key);
|
2021-08-05 08:25:58 +03:00
|
|
|
int dictSdsKeyCompare(dict *d, const void *key1, const void *key2);
|
|
|
|
int dictSdsKeyCaseCompare(dict *d, const void *key1, const void *key2);
|
|
|
|
void dictSdsDestructor(dict *d, void *val);
|
2022-11-16 19:24:18 -08:00
|
|
|
void dictListDestructor(dict *d, void *val);
|
2021-10-07 14:41:26 +03:00
|
|
|
void *dictSdsDup(dict *d, const void *key);
|
2016-12-13 16:27:13 +01:00
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
/* Git SHA1 */
|
2024-04-04 01:46:23 +07:00
|
|
|
char *serverGitSHA1(void);
|
|
|
|
char *serverGitDirty(void);
|
|
|
|
uint64_t serverBuildId(void);
|
|
|
|
const char *serverBuildIdRaw(void);
|
|
|
|
char *serverBuildIdString(void);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
|
|
|
/* Commands prototypes */
|
2015-07-26 15:20:46 +02:00
|
|
|
void authCommand(client *c);
|
|
|
|
void pingCommand(client *c);
|
|
|
|
void echoCommand(client *c);
|
|
|
|
void commandCommand(client *c);
|
Treat subcommands as commands (#9504)
## Intro
The purpose is to allow having different flags/ACL categories for
subcommands (Example: CONFIG GET is ok-loading but CONFIG SET isn't)
We create a small command table for every command that has subcommands
and each subcommand has its own flags, etc. (same as a "regular" command)
This commit also unites the Redis and the Sentinel command tables
## Affected commands
CONFIG
Used to have "admin ok-loading ok-stale no-script"
Changes:
1. Dropped "ok-loading" in all except GET (this doesn't change behavior since
there were checks in the code doing that)
XINFO
Used to have "read-only random"
Changes:
1. Dropped "random" in all except CONSUMERS
XGROUP
Used to have "write use-memory"
Changes:
1. Dropped "use-memory" in all except CREATE and CREATECONSUMER
COMMAND
No changes.
MEMORY
Used to have "random read-only"
Changes:
1. Dropped "random" in PURGE and USAGE
ACL
Used to have "admin no-script ok-loading ok-stale"
Changes:
1. Dropped "admin" in WHOAMI, GENPASS, and CAT
LATENCY
No changes.
MODULE
No changes.
SLOWLOG
Used to have "admin random ok-loading ok-stale"
Changes:
1. Dropped "random" in RESET
OBJECT
Used to have "read-only random"
Changes:
1. Dropped "random" in ENCODING and REFCOUNT
SCRIPT
Used to have "may-replicate no-script"
Changes:
1. Dropped "may-replicate" in all except FLUSH and LOAD
CLIENT
Used to have "admin no-script random ok-loading ok-stale"
Changes:
1. Dropped "random" in all except INFO and LIST
2. Dropped "admin" in ID, TRACKING, CACHING, GETREDIR, INFO, SETNAME, GETNAME, and REPLY
STRALGO
No changes.
PUBSUB
No changes.
CLUSTER
Changes:
1. Dropped "admin in countkeysinslots, getkeysinslot, info, nodes, keyslot, myid, and slots
SENTINEL
No changes.
(note that DEBUG also fits, but we decided not to convert it since it's for
debugging and anyway undocumented)
## New sub-command
This commit adds another element to the per-command output of COMMAND,
describing the list of subcommands, if any (in the same structure as "regular" commands)
Also, it adds a new subcommand:
```
COMMAND LIST [FILTERBY (MODULE <module-name>|ACLCAT <cat>|PATTERN <pattern>)]
```
which returns a set of all commands (unless filters), but excluding subcommands.
## Module API
A new module API, RM_CreateSubcommand, was added, in order to allow
module writer to define subcommands
## ACL changes:
1. Now, that each subcommand is actually a command, each has its own ACL id.
2. The old mechanism of allowed_subcommands is redundant
(blocking/allowing a subcommand is the same as blocking/allowing a regular command),
but we had to keep it, to support the widespread usage of allowed_subcommands
to block commands with certain args, that aren't subcommands (e.g. "-select +select|0").
3. I have renamed allowed_subcommands to allowed_firstargs to emphasize the difference.
4. Because subcommands are commands in ACL too, you can now use "-" to block subcommands
(e.g. "+client -client|kill"), which wasn't possible in the past.
5. It is also possible to use the allowed_firstargs mechanism with subcommand.
For example: `+config -config|set +config|set|loglevel` will block all CONFIG SET except
for setting the log level.
6. All of the ACL changes above required some amount of refactoring.
## Misc
1. There are two approaches: Either each subcommand has its own function or all
subcommands use the same function, determining what to do according to argv[0].
For now, I took the former approaches only with CONFIG and COMMAND,
while other commands use the latter approach (for smaller blamelog diff).
2. Deleted memoryGetKeys: It is no longer needed because MEMORY USAGE now uses the "range" key spec.
4. Bugfix: GETNAME was missing from CLIENT's help message.
5. Sentinel and Redis now use the same table, with the same function pointer.
Some commands have a different implementation in Sentinel, so we redirect
them (these are ROLE, PUBLISH, and INFO).
6. Command stats now show the stats per subcommand (e.g. instead of stats just
for "config" you will have stats for "config|set", "config|get", etc.)
7. It is now possible to use COMMAND directly on subcommands:
COMMAND INFO CONFIG|GET (The pipeline syntax was inspired from ACL, and
can be used in functions lookupCommandBySds and lookupCommandByCString)
8. STRALGO is now a container command (has "help")
## Breaking changes:
1. Command stats now show the stats per subcommand (see (5) above)
2021-10-20 10:52:57 +02:00
|
|
|
void commandCountCommand(client *c);
|
|
|
|
void commandListCommand(client *c);
|
|
|
|
void commandInfoCommand(client *c);
|
|
|
|
void commandGetKeysCommand(client *c);
|
2022-02-08 10:01:35 +02:00
|
|
|
void commandGetKeysAndFlagsCommand(client *c);
|
Treat subcommands as commands (#9504)
## Intro
The purpose is to allow having different flags/ACL categories for
subcommands (Example: CONFIG GET is ok-loading but CONFIG SET isn't)
We create a small command table for every command that has subcommands
and each subcommand has its own flags, etc. (same as a "regular" command)
This commit also unites the Redis and the Sentinel command tables
## Affected commands
CONFIG
Used to have "admin ok-loading ok-stale no-script"
Changes:
1. Dropped "ok-loading" in all except GET (this doesn't change behavior since
there were checks in the code doing that)
XINFO
Used to have "read-only random"
Changes:
1. Dropped "random" in all except CONSUMERS
XGROUP
Used to have "write use-memory"
Changes:
1. Dropped "use-memory" in all except CREATE and CREATECONSUMER
COMMAND
No changes.
MEMORY
Used to have "random read-only"
Changes:
1. Dropped "random" in PURGE and USAGE
ACL
Used to have "admin no-script ok-loading ok-stale"
Changes:
1. Dropped "admin" in WHOAMI, GENPASS, and CAT
LATENCY
No changes.
MODULE
No changes.
SLOWLOG
Used to have "admin random ok-loading ok-stale"
Changes:
1. Dropped "random" in RESET
OBJECT
Used to have "read-only random"
Changes:
1. Dropped "random" in ENCODING and REFCOUNT
SCRIPT
Used to have "may-replicate no-script"
Changes:
1. Dropped "may-replicate" in all except FLUSH and LOAD
CLIENT
Used to have "admin no-script random ok-loading ok-stale"
Changes:
1. Dropped "random" in all except INFO and LIST
2. Dropped "admin" in ID, TRACKING, CACHING, GETREDIR, INFO, SETNAME, GETNAME, and REPLY
STRALGO
No changes.
PUBSUB
No changes.
CLUSTER
Changes:
1. Dropped "admin in countkeysinslots, getkeysinslot, info, nodes, keyslot, myid, and slots
SENTINEL
No changes.
(note that DEBUG also fits, but we decided not to convert it since it's for
debugging and anyway undocumented)
## New sub-command
This commit adds another element to the per-command output of COMMAND,
describing the list of subcommands, if any (in the same structure as "regular" commands)
Also, it adds a new subcommand:
```
COMMAND LIST [FILTERBY (MODULE <module-name>|ACLCAT <cat>|PATTERN <pattern>)]
```
which returns a set of all commands (unless filters), but excluding subcommands.
## Module API
A new module API, RM_CreateSubcommand, was added, in order to allow
module writer to define subcommands
## ACL changes:
1. Now, that each subcommand is actually a command, each has its own ACL id.
2. The old mechanism of allowed_subcommands is redundant
(blocking/allowing a subcommand is the same as blocking/allowing a regular command),
but we had to keep it, to support the widespread usage of allowed_subcommands
to block commands with certain args, that aren't subcommands (e.g. "-select +select|0").
3. I have renamed allowed_subcommands to allowed_firstargs to emphasize the difference.
4. Because subcommands are commands in ACL too, you can now use "-" to block subcommands
(e.g. "+client -client|kill"), which wasn't possible in the past.
5. It is also possible to use the allowed_firstargs mechanism with subcommand.
For example: `+config -config|set +config|set|loglevel` will block all CONFIG SET except
for setting the log level.
6. All of the ACL changes above required some amount of refactoring.
## Misc
1. There are two approaches: Either each subcommand has its own function or all
subcommands use the same function, determining what to do according to argv[0].
For now, I took the former approaches only with CONFIG and COMMAND,
while other commands use the latter approach (for smaller blamelog diff).
2. Deleted memoryGetKeys: It is no longer needed because MEMORY USAGE now uses the "range" key spec.
4. Bugfix: GETNAME was missing from CLIENT's help message.
5. Sentinel and Redis now use the same table, with the same function pointer.
Some commands have a different implementation in Sentinel, so we redirect
them (these are ROLE, PUBLISH, and INFO).
6. Command stats now show the stats per subcommand (e.g. instead of stats just
for "config" you will have stats for "config|set", "config|get", etc.)
7. It is now possible to use COMMAND directly on subcommands:
COMMAND INFO CONFIG|GET (The pipeline syntax was inspired from ACL, and
can be used in functions lookupCommandBySds and lookupCommandByCString)
8. STRALGO is now a container command (has "help")
## Breaking changes:
1. Command stats now show the stats per subcommand (see (5) above)
2021-10-20 10:52:57 +02:00
|
|
|
void commandHelpCommand(client *c);
|
Move doc metadata from COMMAND to COMMAND DOCS (#10056)
Syntax:
`COMMAND DOCS [<command name> ...]`
Background:
Apparently old version of hiredis (and thus also redis-cli) can't
support more than 7 levels of multi-bulk nesting.
The solution is to move all the doc related metadata from COMMAND to a
new COMMAND DOCS sub-command.
The new DOCS sub-command returns a map of commands (not an array like in COMMAND),
And the same goes for the `subcommands` field inside it (also contains a map)
Besides that, the remaining new fields of COMMAND (hints, key-specs, and
sub-commands), are placed in the outer array rather than a nested map.
this was done mainly for consistency with the old format.
Other changes:
---
* Allow COMMAND INFO with no arguments, which returns all commands, so that we can some day deprecated
the plain COMMAND (no args)
* Reduce the amount of deferred replies from both COMMAND and COMMAND
DOCS, especially in the inner loops, since these create many small
reply objects, which lead to many small write syscalls and many small
TCP packets.
To make this easier, when populating the command table, we count the
history, args, and hints so we later know their size in advance.
Additionally, the movablekeys flag was moved into the flags register.
* Update generate-commands-json.py to take the data from both command, it
now executes redis-cli directly, instead of taking input from stdin.
* Sub-commands in both COMMAND (and COMMAND INFO), and also COMMAND DOCS,
show their full name. i.e. CONFIG
* GET will be shown as `config|get` rather than just `get`.
This will be visible both when asking for `COMMAND INFO config` and COMMAND INFO config|get`, but is
especially important for the later.
i.e. imagine someone doing `COMMAND INFO slowlog|get config|get` not being able to distinguish between the two
items in the array response.
2022-01-11 17:16:16 +02:00
|
|
|
void commandDocsCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void setCommand(client *c);
|
|
|
|
void setnxCommand(client *c);
|
|
|
|
void setexCommand(client *c);
|
|
|
|
void psetexCommand(client *c);
|
|
|
|
void getCommand(client *c);
|
2021-01-27 09:47:26 -08:00
|
|
|
void getexCommand(client *c);
|
|
|
|
void getdelCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void delCommand(client *c);
|
2015-07-30 11:46:31 +02:00
|
|
|
void unlinkCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void existsCommand(client *c);
|
|
|
|
void setbitCommand(client *c);
|
|
|
|
void getbitCommand(client *c);
|
BITFIELD command initial implementation.
The new bitfield command is an extension to the Redis bit operations,
where not just single bit operations are performed, but the array of
bits composing a string, can be addressed at random, not aligned
offsets, with any width unsigned and signed integers like u8, s5, u10
(up to 64 bit signed integers and 63 bit unsigned integers).
The BITFIELD command supports subcommands that can SET, GET, or INCRBY
those arbitrary bit counters, with multiple overflow semantics.
Trivial and credits:
A similar command was imagined a few times in the past, but for
some reason looked a bit far fetched or not well specified.
Finally the command was proposed again in a clear form by
Yoav Steinberg from Redis Labs, that proposed a set of commands on
arbitrary sized integers stored at bit offsets.
Starting from this proposal I wrote an initial specification of a single
command with sub-commands similar to what Yoav envisioned, using short
names for types definitions, and adding control on the overflow.
This commit is the resulting implementation.
Examples:
BITFIELD mykey OVERFLOW wrap INCRBY i2 10 -1 GET i2 10
2016-02-25 23:31:45 +01:00
|
|
|
void bitfieldCommand(client *c);
|
2020-03-04 20:51:45 +08:00
|
|
|
void bitfieldroCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void setrangeCommand(client *c);
|
|
|
|
void getrangeCommand(client *c);
|
|
|
|
void incrCommand(client *c);
|
|
|
|
void decrCommand(client *c);
|
|
|
|
void incrbyCommand(client *c);
|
|
|
|
void decrbyCommand(client *c);
|
|
|
|
void incrbyfloatCommand(client *c);
|
|
|
|
void selectCommand(client *c);
|
SWAPDB command.
This new command swaps two Redis databases, so that immediately all the
clients connected to a given DB will see the data of the other DB, and
the other way around. Example:
SWAPDB 0 1
This will swap DB 0 with DB 1. All the clients connected with DB 0 will
immediately see the new data, exactly like all the clients connected
with DB 1 will see the data that was formerly of DB 0.
MOTIVATION AND HISTORY
---
The command was recently demanded by Pedro Melo, but was suggested in
the past multiple times, and always refused by me.
The reason why it was asked: Imagine you have clients operating in DB 0.
At the same time, you create a new version of the dataset in DB 1.
When the new version of the dataset is available, you immediately want
to swap the two views, so that the clients will transparently use the
new version of the data. At the same time you'll likely destroy the
DB 1 dataset (that contains the old data) and start to build a new
version, to repeat the process.
This is an interesting pattern, but the reason why I always opposed to
implement this, was that FLUSHDB was a blocking command in Redis before
Redis 4.0 improvements. Now we have FLUSHDB ASYNC that releases the
old data in O(1) from the point of view of the client, to reclaim memory
incrementally in a different thread.
At this point, the pattern can really be supported without latency
spikes, so I'm providing this implementation for the users to comment.
In case a very compelling argument will be made against this new command
it may be removed.
BEHAVIOR WITH BLOCKING OPERATIONS
---
If a client is blocking for a list in a given DB, after the swap it will
still be blocked in the same DB ID, since this is the most logical thing
to do: if I was blocked for a list push to list "foo", even after the
swap I want still a LPUSH to reach the key "foo" in the same DB in order
to unblock.
However an interesting thing happens when a client is, for instance,
blocked waiting for new elements in list "foo" of DB 0. Then the DB
0 and 1 are swapped with SWAPDB. However the DB 1 happened to have
a list called "foo" containing elements. When this happens, this
implementation can correctly unblock the client.
It is possible that there are subtle corner cases that are not covered
in the implementation, but since the command is self-contained from the
POV of the implementation and the Redis core, it cannot cause anything
bad if not used.
Tests and documentation are yet to be provided.
2016-10-14 15:28:04 +02:00
|
|
|
void swapdbCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void randomkeyCommand(client *c);
|
|
|
|
void keysCommand(client *c);
|
|
|
|
void scanCommand(client *c);
|
|
|
|
void dbsizeCommand(client *c);
|
|
|
|
void lastsaveCommand(client *c);
|
|
|
|
void saveCommand(client *c);
|
|
|
|
void bgsaveCommand(client *c);
|
|
|
|
void bgrewriteaofCommand(client *c);
|
|
|
|
void shutdownCommand(client *c);
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
void slowlogCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void moveCommand(client *c);
|
2020-11-17 19:03:05 +09:00
|
|
|
void copyCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void renameCommand(client *c);
|
|
|
|
void renamenxCommand(client *c);
|
|
|
|
void lpushCommand(client *c);
|
|
|
|
void rpushCommand(client *c);
|
|
|
|
void lpushxCommand(client *c);
|
|
|
|
void rpushxCommand(client *c);
|
|
|
|
void linsertCommand(client *c);
|
|
|
|
void lpopCommand(client *c);
|
|
|
|
void rpopCommand(client *c);
|
2021-09-09 17:02:33 +08:00
|
|
|
void lmpopCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void llenCommand(client *c);
|
|
|
|
void lindexCommand(client *c);
|
|
|
|
void lrangeCommand(client *c);
|
|
|
|
void ltrimCommand(client *c);
|
|
|
|
void typeCommand(client *c);
|
|
|
|
void lsetCommand(client *c);
|
|
|
|
void saddCommand(client *c);
|
|
|
|
void sremCommand(client *c);
|
|
|
|
void smoveCommand(client *c);
|
|
|
|
void sismemberCommand(client *c);
|
2020-08-11 04:55:06 -04:00
|
|
|
void smismemberCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void scardCommand(client *c);
|
|
|
|
void spopCommand(client *c);
|
|
|
|
void srandmemberCommand(client *c);
|
|
|
|
void sinterCommand(client *c);
|
2021-08-03 04:45:27 -04:00
|
|
|
void sinterCardCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void sinterstoreCommand(client *c);
|
|
|
|
void sunionCommand(client *c);
|
|
|
|
void sunionstoreCommand(client *c);
|
|
|
|
void sdiffCommand(client *c);
|
|
|
|
void sdiffstoreCommand(client *c);
|
|
|
|
void sscanCommand(client *c);
|
|
|
|
void syncCommand(client *c);
|
|
|
|
void flushdbCommand(client *c);
|
|
|
|
void flushallCommand(client *c);
|
|
|
|
void sortCommand(client *c);
|
2021-08-09 08:40:29 +02:00
|
|
|
void sortroCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void lremCommand(client *c);
|
2020-06-10 12:40:24 +02:00
|
|
|
void lposCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void rpoplpushCommand(client *c);
|
2020-10-08 02:33:17 -03:00
|
|
|
void lmoveCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void infoCommand(client *c);
|
|
|
|
void mgetCommand(client *c);
|
|
|
|
void monitorCommand(client *c);
|
|
|
|
void expireCommand(client *c);
|
|
|
|
void expireatCommand(client *c);
|
|
|
|
void pexpireCommand(client *c);
|
|
|
|
void pexpireatCommand(client *c);
|
|
|
|
void getsetCommand(client *c);
|
|
|
|
void ttlCommand(client *c);
|
2016-06-14 15:33:59 +02:00
|
|
|
void touchCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void pttlCommand(client *c);
|
2021-05-29 23:20:32 -07:00
|
|
|
void expiretimeCommand(client *c);
|
|
|
|
void pexpiretimeCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void persistCommand(client *c);
|
2018-09-10 10:43:39 +02:00
|
|
|
void replicaofCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void roleCommand(client *c);
|
|
|
|
void debugCommand(client *c);
|
|
|
|
void msetCommand(client *c);
|
|
|
|
void msetnxCommand(client *c);
|
|
|
|
void zaddCommand(client *c);
|
|
|
|
void zincrbyCommand(client *c);
|
|
|
|
void zrangeCommand(client *c);
|
|
|
|
void zrangebyscoreCommand(client *c);
|
|
|
|
void zrevrangebyscoreCommand(client *c);
|
|
|
|
void zrangebylexCommand(client *c);
|
|
|
|
void zrevrangebylexCommand(client *c);
|
|
|
|
void zcountCommand(client *c);
|
|
|
|
void zlexcountCommand(client *c);
|
|
|
|
void zrevrangeCommand(client *c);
|
|
|
|
void zcardCommand(client *c);
|
|
|
|
void zremCommand(client *c);
|
|
|
|
void zscoreCommand(client *c);
|
2020-08-04 10:49:33 -04:00
|
|
|
void zmscoreCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void zremrangebyscoreCommand(client *c);
|
|
|
|
void zremrangebylexCommand(client *c);
|
2018-05-11 17:31:46 +02:00
|
|
|
void zpopminCommand(client *c);
|
|
|
|
void zpopmaxCommand(client *c);
|
2021-09-23 13:34:40 +08:00
|
|
|
void zmpopCommand(client *c);
|
2018-05-11 17:31:46 +02:00
|
|
|
void bzpopminCommand(client *c);
|
|
|
|
void bzpopmaxCommand(client *c);
|
2021-09-23 13:34:40 +08:00
|
|
|
void bzmpopCommand(client *c);
|
2021-01-29 16:47:28 +08:00
|
|
|
void zrandmemberCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void multiCommand(client *c);
|
|
|
|
void execCommand(client *c);
|
|
|
|
void discardCommand(client *c);
|
|
|
|
void blpopCommand(client *c);
|
|
|
|
void brpopCommand(client *c);
|
2021-09-09 17:02:33 +08:00
|
|
|
void blmpopCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void brpoplpushCommand(client *c);
|
2020-10-08 02:33:17 -03:00
|
|
|
void blmoveCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void appendCommand(client *c);
|
|
|
|
void strlenCommand(client *c);
|
|
|
|
void zrankCommand(client *c);
|
|
|
|
void zrevrankCommand(client *c);
|
|
|
|
void hsetCommand(client *c);
|
|
|
|
void hsetnxCommand(client *c);
|
|
|
|
void hgetCommand(client *c);
|
|
|
|
void hmgetCommand(client *c);
|
|
|
|
void hdelCommand(client *c);
|
|
|
|
void hlenCommand(client *c);
|
|
|
|
void hstrlenCommand(client *c);
|
|
|
|
void zremrangebyrankCommand(client *c);
|
|
|
|
void zunionstoreCommand(client *c);
|
|
|
|
void zinterstoreCommand(client *c);
|
2020-11-15 09:14:25 -03:00
|
|
|
void zdiffstoreCommand(client *c);
|
2020-09-11 16:59:15 +08:00
|
|
|
void zunionCommand(client *c);
|
|
|
|
void zinterCommand(client *c);
|
2021-08-03 04:45:27 -04:00
|
|
|
void zinterCardCommand(client *c);
|
2021-01-07 03:58:53 -05:00
|
|
|
void zrangestoreCommand(client *c);
|
2020-11-15 09:14:25 -03:00
|
|
|
void zdiffCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void zscanCommand(client *c);
|
|
|
|
void hkeysCommand(client *c);
|
|
|
|
void hvalsCommand(client *c);
|
|
|
|
void hgetallCommand(client *c);
|
|
|
|
void hexistsCommand(client *c);
|
|
|
|
void hscanCommand(client *c);
|
2021-01-29 16:47:28 +08:00
|
|
|
void hrandfieldCommand(client *c);
|
Treat subcommands as commands (#9504)
## Intro
The purpose is to allow having different flags/ACL categories for
subcommands (Example: CONFIG GET is ok-loading but CONFIG SET isn't)
We create a small command table for every command that has subcommands
and each subcommand has its own flags, etc. (same as a "regular" command)
This commit also unites the Redis and the Sentinel command tables
## Affected commands
CONFIG
Used to have "admin ok-loading ok-stale no-script"
Changes:
1. Dropped "ok-loading" in all except GET (this doesn't change behavior since
there were checks in the code doing that)
XINFO
Used to have "read-only random"
Changes:
1. Dropped "random" in all except CONSUMERS
XGROUP
Used to have "write use-memory"
Changes:
1. Dropped "use-memory" in all except CREATE and CREATECONSUMER
COMMAND
No changes.
MEMORY
Used to have "random read-only"
Changes:
1. Dropped "random" in PURGE and USAGE
ACL
Used to have "admin no-script ok-loading ok-stale"
Changes:
1. Dropped "admin" in WHOAMI, GENPASS, and CAT
LATENCY
No changes.
MODULE
No changes.
SLOWLOG
Used to have "admin random ok-loading ok-stale"
Changes:
1. Dropped "random" in RESET
OBJECT
Used to have "read-only random"
Changes:
1. Dropped "random" in ENCODING and REFCOUNT
SCRIPT
Used to have "may-replicate no-script"
Changes:
1. Dropped "may-replicate" in all except FLUSH and LOAD
CLIENT
Used to have "admin no-script random ok-loading ok-stale"
Changes:
1. Dropped "random" in all except INFO and LIST
2. Dropped "admin" in ID, TRACKING, CACHING, GETREDIR, INFO, SETNAME, GETNAME, and REPLY
STRALGO
No changes.
PUBSUB
No changes.
CLUSTER
Changes:
1. Dropped "admin in countkeysinslots, getkeysinslot, info, nodes, keyslot, myid, and slots
SENTINEL
No changes.
(note that DEBUG also fits, but we decided not to convert it since it's for
debugging and anyway undocumented)
## New sub-command
This commit adds another element to the per-command output of COMMAND,
describing the list of subcommands, if any (in the same structure as "regular" commands)
Also, it adds a new subcommand:
```
COMMAND LIST [FILTERBY (MODULE <module-name>|ACLCAT <cat>|PATTERN <pattern>)]
```
which returns a set of all commands (unless filters), but excluding subcommands.
## Module API
A new module API, RM_CreateSubcommand, was added, in order to allow
module writer to define subcommands
## ACL changes:
1. Now, that each subcommand is actually a command, each has its own ACL id.
2. The old mechanism of allowed_subcommands is redundant
(blocking/allowing a subcommand is the same as blocking/allowing a regular command),
but we had to keep it, to support the widespread usage of allowed_subcommands
to block commands with certain args, that aren't subcommands (e.g. "-select +select|0").
3. I have renamed allowed_subcommands to allowed_firstargs to emphasize the difference.
4. Because subcommands are commands in ACL too, you can now use "-" to block subcommands
(e.g. "+client -client|kill"), which wasn't possible in the past.
5. It is also possible to use the allowed_firstargs mechanism with subcommand.
For example: `+config -config|set +config|set|loglevel` will block all CONFIG SET except
for setting the log level.
6. All of the ACL changes above required some amount of refactoring.
## Misc
1. There are two approaches: Either each subcommand has its own function or all
subcommands use the same function, determining what to do according to argv[0].
For now, I took the former approaches only with CONFIG and COMMAND,
while other commands use the latter approach (for smaller blamelog diff).
2. Deleted memoryGetKeys: It is no longer needed because MEMORY USAGE now uses the "range" key spec.
4. Bugfix: GETNAME was missing from CLIENT's help message.
5. Sentinel and Redis now use the same table, with the same function pointer.
Some commands have a different implementation in Sentinel, so we redirect
them (these are ROLE, PUBLISH, and INFO).
6. Command stats now show the stats per subcommand (e.g. instead of stats just
for "config" you will have stats for "config|set", "config|get", etc.)
7. It is now possible to use COMMAND directly on subcommands:
COMMAND INFO CONFIG|GET (The pipeline syntax was inspired from ACL, and
can be used in functions lookupCommandBySds and lookupCommandByCString)
8. STRALGO is now a container command (has "help")
## Breaking changes:
1. Command stats now show the stats per subcommand (see (5) above)
2021-10-20 10:52:57 +02:00
|
|
|
void configSetCommand(client *c);
|
|
|
|
void configGetCommand(client *c);
|
|
|
|
void configResetStatCommand(client *c);
|
|
|
|
void configRewriteCommand(client *c);
|
|
|
|
void configHelpCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void hincrbyCommand(client *c);
|
|
|
|
void hincrbyfloatCommand(client *c);
|
|
|
|
void subscribeCommand(client *c);
|
|
|
|
void unsubscribeCommand(client *c);
|
|
|
|
void psubscribeCommand(client *c);
|
|
|
|
void punsubscribeCommand(client *c);
|
|
|
|
void publishCommand(client *c);
|
|
|
|
void pubsubCommand(client *c);
|
2022-01-03 01:54:47 +01:00
|
|
|
void spublishCommand(client *c);
|
|
|
|
void ssubscribeCommand(client *c);
|
|
|
|
void sunsubscribeCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void watchCommand(client *c);
|
|
|
|
void unwatchCommand(client *c);
|
|
|
|
void clusterCommand(client *c);
|
2024-06-27 19:58:27 -04:00
|
|
|
void clusterSlotStatsCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void restoreCommand(client *c);
|
|
|
|
void migrateCommand(client *c);
|
|
|
|
void askingCommand(client *c);
|
|
|
|
void readonlyCommand(client *c);
|
|
|
|
void readwriteCommand(client *c);
|
2021-12-26 09:03:37 +02:00
|
|
|
int verifyDumpPayload(unsigned char *p, size_t len, uint16_t *rdbver_ptr);
|
2015-07-26 15:20:46 +02:00
|
|
|
void dumpCommand(client *c);
|
|
|
|
void objectCommand(client *c);
|
2016-09-13 10:26:36 +02:00
|
|
|
void memoryCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void clientCommand(client *c);
|
2018-12-04 12:46:16 +01:00
|
|
|
void helloCommand(client *c);
|
2023-03-22 07:17:20 +01:00
|
|
|
void clientSetinfoCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void evalCommand(client *c);
|
2021-05-12 21:07:34 -07:00
|
|
|
void evalRoCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void evalShaCommand(client *c);
|
2021-05-12 21:07:34 -07:00
|
|
|
void evalShaRoCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void scriptCommand(client *c);
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
void fcallCommand(client *c);
|
|
|
|
void fcallroCommand(client *c);
|
Redis Function Libraries (#10004)
# Redis Function Libraries
This PR implements Redis Functions Libraries as describe on: https://github.com/redis/redis/issues/9906.
Libraries purpose is to provide a better code sharing between functions by allowing to create multiple
functions in a single command. Functions that were created together can safely share code between
each other without worrying about compatibility issues and versioning.
Creating a new library is done using 'FUNCTION LOAD' command (full API is described below)
This PR introduces a new struct called libraryInfo, libraryInfo holds information about a library:
* name - name of the library
* engine - engine used to create the library
* code - library code
* description - library description
* functions - the functions exposed by the library
When Redis gets the `FUNCTION LOAD` command it creates a new empty libraryInfo.
Redis passes the `CODE` to the relevant engine alongside the empty libraryInfo.
As a result, the engine will create one or more functions by calling 'libraryCreateFunction'.
The new funcion will be added to the newly created libraryInfo. So far Everything is happening
locally on the libraryInfo so it is easy to abort the operation (in case of an error) by simply
freeing the libraryInfo. After the library info is fully constructed we start the joining phase by
which we will join the new library to the other libraries currently exist on Redis.
The joining phase make sure there is no function collision and add the library to the
librariesCtx (renamed from functionCtx). LibrariesCtx is used all around the code in the exact
same way as functionCtx was used (with respect to RDB loading, replicatio, ...).
The only difference is that apart from function dictionary (maps function name to functionInfo
object), the librariesCtx contains also a libraries dictionary that maps library name to libraryInfo object.
## New API
### FUNCTION LOAD
`FUNCTION LOAD <ENGINE> <LIBRARY NAME> [REPLACE] [DESCRIPTION <DESCRIPTION>] <CODE>`
Create a new library with the given parameters:
* ENGINE - REPLACE Engine name to use to create the library.
* LIBRARY NAME - The new library name.
* REPLACE - If the library already exists, replace it.
* DESCRIPTION - Library description.
* CODE - Library code.
Return "OK" on success, or error on the following cases:
* Library name already taken and REPLACE was not used
* Name collision with another existing library (even if replace was uses)
* Library registration failed by the engine (usually compilation error)
## Changed API
### FUNCTION LIST
`FUNCTION LIST [LIBRARYNAME <LIBRARY NAME PATTERN>] [WITHCODE]`
Command was modified to also allow getting libraries code (so `FUNCTION INFO` command is no longer
needed and removed). In addition the command gets an option argument, `LIBRARYNAME` allows you to
only get libraries that match the given `LIBRARYNAME` pattern. By default, it returns all libraries.
### INFO MEMORY
Added number of libraries to `INFO MEMORY`
### Commands flags
`DENYOOM` flag was set on `FUNCTION LOAD` and `FUNCTION RESTORE`. We consider those commands
as commands that add new data to the dateset (functions are data) and so we want to disallows
to run those commands on OOM.
## Removed API
* FUNCTION CREATE - Decided on https://github.com/redis/redis/issues/9906
* FUNCTION INFO - Decided on https://github.com/redis/redis/issues/9899
## Lua engine changes
When the Lua engine gets the code given on `FUNCTION LOAD` command, it immediately runs it, we call
this run the loading run. Loading run is not a usual script run, it is not possible to invoke any
Redis command from within the load run.
Instead there is a new API provided by `library` object. The new API's:
* `redis.log` - behave the same as `redis.log`
* `redis.register_function` - register a new function to the library
The loading run purpose is to register functions using the new `redis.register_function` API.
Any attempt to use any other API will result in an error. In addition, the load run is has a time
limit of 500ms, error is raise on timeout and the entire operation is aborted.
### `redis.register_function`
`redis.register_function(<function_name>, <callback>, [<description>])`
This new API allows users to register a new function that will be linked to the newly created library.
This API can only be called during the load run (see definition above). Any attempt to use it outside
of the load run will result in an error.
The parameters pass to the API are:
* function_name - Function name (must be a Lua string)
* callback - Lua function object that will be called when the function is invokes using fcall/fcall_ro
* description - Function description, optional (must be a Lua string).
### Example
The following example creates a library called `lib` with 2 functions, `f1` and `f1`, returns 1 and 2 respectively:
```
local function f1(keys, args)
return 1
end
local function f2(keys, args)
return 2
end
redis.register_function('f1', f1)
redis.register_function('f2', f2)
```
Notice: Unlike `eval`, functions inside a library get the KEYS and ARGV as arguments to the
functions and not as global.
### Technical Details
On the load run we only want the user to be able to call a white list on API's. This way, in
the future, if new API's will be added, the new API's will not be available to the load run
unless specifically added to this white list. We put the while list on the `library` object and
make sure the `library` object is only available to the load run by using [lua_setfenv](https://www.lua.org/manual/5.1/manual.html#lua_setfenv) API. This API allows us to set
the `globals` of a function (and all the function it creates). Before starting the load run we
create a new fresh Lua table (call it `g`) that only contains the `library` API (we make sure
to set global protection on this table just like the general global protection already exists
today), then we use [lua_setfenv](https://www.lua.org/manual/5.1/manual.html#lua_setfenv)
to set `g` as the global table of the load run. After the load run finished we update `g`
metatable and set `__index` and `__newindex` functions to be `_G` (Lua default globals),
we also pop out the `library` object as we do not need it anymore.
This way, any function that was created on the load run (and will be invoke using `fcall`) will
see the default globals as it expected to see them and will not have the `library` API anymore.
An important outcome of this new approach is that now we can achieve a distinct global table
for each library (it is not yet like that but it is very easy to achieve it now). In the future we can
decide to remove global protection because global on different libraries will not collide or we
can chose to give different API to different libraries base on some configuration or input.
Notice that this technique was meant to prevent errors and was not meant to prevent malicious
user from exploit it. For example, the load run can still save the `library` object on some local
variable and then using in `fcall` context. To prevent such a malicious use, the C code also make
sure it is running in the right context and if not raise an error.
2022-01-06 13:39:38 +02:00
|
|
|
void functionLoadCommand(client *c);
|
Auto-generate the command table from JSON files (#9656)
Delete the hardcoded command table and replace it with an auto-generated table, based
on a JSON file that describes the commands (each command must have a JSON file).
These JSON files are the SSOT of everything there is to know about Redis commands,
and it is reflected fully in COMMAND INFO.
These JSON files are used to generate commands.c (using a python script), which is then
committed to the repo and compiled.
The purpose is:
* Clients and proxies will be able to get much more info from redis, instead of relying on hard coded logic.
* drop the dependency between Redis-user and the commands.json in redis-doc.
* delete help.h and have redis-cli learn everything it needs to know just by issuing COMMAND (will be
done in a separate PR)
* redis.io should stop using commands.json and learn everything from Redis (ultimately one of the release
artifacts should be a large JSON, containing all the information about all of the commands, which will be
generated from COMMAND's reply)
* the byproduct of this is:
* module commands will be able to provide that info and possibly be more of a first-class citizens
* in theory, one may be able to generate a redis client library for a strictly typed language, by using this info.
### Interface changes
#### COMMAND INFO's reply change (and arg-less COMMAND)
Before this commit the reply at index 7 contained the key-specs list
and reply at index 8 contained the sub-commands list (Both unreleased).
Now, reply at index 7 is a map of:
- summary - short command description
- since - debut version
- group - command group
- complexity - complexity string
- doc-flags - flags used for documentation (e.g. "deprecated")
- deprecated-since - if deprecated, from which version?
- replaced-by - if deprecated, which command replaced it?
- history - a list of (version, what-changed) tuples
- hints - a list of strings, meant to provide hints for clients/proxies. see https://github.com/redis/redis/issues/9876
- arguments - an array of arguments. each element is a map, with the possibility of nesting (sub-arguments)
- key-specs - an array of keys specs (already in unstable, just changed location)
- subcommands - a list of sub-commands (already in unstable, just changed location)
- reply-schema - will be added in the future (see https://github.com/redis/redis/issues/9845)
more details on these can be found in https://github.com/redis/redis-doc/pull/1697
only the first three fields are mandatory
#### API changes (unreleased API obviously)
now they take RedisModuleCommand opaque pointer instead of looking up the command by name
- RM_CreateSubcommand
- RM_AddCommandKeySpec
- RM_SetCommandKeySpecBeginSearchIndex
- RM_SetCommandKeySpecBeginSearchKeyword
- RM_SetCommandKeySpecFindKeysRange
- RM_SetCommandKeySpecFindKeysKeynum
Currently, we did not add module API to provide additional information about their commands because
we couldn't agree on how the API should look like, see https://github.com/redis/redis/issues/9944.
### Somehow related changes
1. Literals should be in uppercase while placeholder in lowercase. Now all the GEO* command
will be documented with M|KM|FT|MI and can take both lowercase and uppercase
### Unrelated changes
1. Bugfix: no_madaory_keys was absent in COMMAND's reply
2. expose CMD_MODULE as "module" via COMMAND
3. have a dedicated uint64 for ACL categories (instead of having them in the same uint64 as command flags)
Co-authored-by: Itamar Haber <itamar@garantiadata.com>
2021-12-15 20:23:15 +01:00
|
|
|
void functionDeleteCommand(client *c);
|
|
|
|
void functionKillCommand(client *c);
|
|
|
|
void functionStatsCommand(client *c);
|
|
|
|
void functionListCommand(client *c);
|
|
|
|
void functionHelpCommand(client *c);
|
2021-12-16 17:58:25 +02:00
|
|
|
void functionFlushCommand(client *c);
|
2021-12-26 09:03:37 +02:00
|
|
|
void functionRestoreCommand(client *c);
|
|
|
|
void functionDumpCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void timeCommand(client *c);
|
|
|
|
void bitopCommand(client *c);
|
|
|
|
void bitcountCommand(client *c);
|
|
|
|
void bitposCommand(client *c);
|
|
|
|
void replconfCommand(client *c);
|
|
|
|
void waitCommand(client *c);
|
Implementing the WAITAOF command (issue #10505) (#11713)
Implementing the WAITAOF functionality which would allow the user to
block until a specified number of Redises have fsynced all previous write
commands to the AOF.
Syntax: `WAITAOF <num_local> <num_replicas> <timeout>`
Response: Array containing two elements: num_local, num_replicas
num_local is always either 0 or 1 representing the local AOF on the master.
num_replicas is the number of replicas that acknowledged the a replication
offset of the last write being fsynced to the AOF.
Returns an error when called on replicas, or when called with non-zero
num_local on a master with AOF disabled, in all other cases the response
just contains number of fsync copies.
Main changes:
* Added code to keep track of replication offsets that are confirmed to have
been fsynced to disk.
* Keep advancing master_repl_offset even when replication is disabled (and
there's no replication backlog, only if there's an AOF enabled).
This way we can use this command and it's mechanisms even when replication
is disabled.
* Extend REPLCONF ACK to `REPLCONF ACK <ofs> FACK <ofs>`, the FACK
will be appended only if there's an AOF on the replica, and already ignored on
old masters (thus backwards compatible)
* WAIT now no longer wait for the replication offset after your last command, but
rather the replication offset after your last write (or read command that caused
propagation, e.g. lazy expiry).
Unrelated changes:
* WAIT command respects CLIENT_DENY_BLOCKING (not just CLIENT_MULTI)
Implementation details:
* Add an atomic var named `fsynced_reploff_pending` that's updated
(usually by the bio thread) and later copied to the main `fsynced_reploff`
variable (only if the AOF base file exists).
I.e. during the initial AOF rewrite it will not be used as the fsynced offset
since the AOF base is still missing.
* Replace close+fsync bio job with new BIO_CLOSE_AOF (AOF specific)
job that will also update fsync offset the field.
* Handle all AOF jobs (BIO_CLOSE_AOF, BIO_AOF_FSYNC) in the same bio
worker thread, to impose ordering on their execution. This solves a
race condition where a job could set `fsynced_reploff_pending` to a higher
value than another pending fsync job, resulting in indicating an offset
for which parts of the data have not yet actually been fsynced.
Imposing an ordering on the jobs guarantees that fsync jobs are executed
in increasing order of replication offset.
* Drain bio jobs when switching `appendfsync` to "always"
This should prevent a write race between updates to `fsynced_reploff_pending`
in the main thread (`flushAppendOnlyFile` when set to ALWAYS fsync), and
those done in the bio thread.
* Drain the pending fsync when starting over a new AOF to avoid race conditions
with the previous AOF offsets overriding the new one (e.g. after switching to
replicate from a new master).
* Make sure to update the fsynced offset at the end of the initial AOF rewrite.
a must in case there are no additional writes that trigger a periodic fsync,
specifically for a replica that does a full sync.
Limitations:
It is possible to write a module and a Lua script that propagate to the AOF and doesn't
propagate to the replication stream. see REDISMODULE_ARGV_NO_REPLICAS and luaRedisSetReplCommand.
These features are incompatible with the WAITAOF command, and can result
in two bad cases. The scenario is that the user executes command that only
propagates to AOF, and then immediately
issues a WAITAOF, and there's no further writes on the replication stream after that.
1. if the the last thing that happened on the replication stream is a PING
(which increased the replication offset but won't trigger an fsync on the replica),
then the client would hang forever (will wait for an fack that the replica will never
send sine it doesn't trigger any fsyncs).
2. if the last thing that happened is a write command that got propagated properly,
then WAITAOF will be released immediately, without waiting for an fsync (since
the offset didn't change)
Refactoring:
* Plumbing to allow bio worker to handle multiple job types
This introduces infrastructure necessary to allow BIO workers to
not have a 1-1 mapping of worker to job-type. This allows in the
future to assign multiple job types to a single worker, either as
a performance/resource optimization, or as a way of enforcing
ordering between specific classes of jobs.
Co-authored-by: Oran Agra <oran@redislabs.com>
2023-03-14 20:26:21 +02:00
|
|
|
void waitaofCommand(client *c);
|
2017-06-30 10:03:37 +02:00
|
|
|
void georadiusbymemberCommand(client *c);
|
|
|
|
void georadiusbymemberroCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void georadiusCommand(client *c);
|
2017-06-30 10:03:37 +02:00
|
|
|
void georadiusroCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void geoaddCommand(client *c);
|
|
|
|
void geohashCommand(client *c);
|
|
|
|
void geoposCommand(client *c);
|
|
|
|
void geodistCommand(client *c);
|
2020-12-12 08:21:05 +08:00
|
|
|
void geosearchCommand(client *c);
|
|
|
|
void geosearchstoreCommand(client *c);
|
2015-07-26 15:20:46 +02:00
|
|
|
void pfselftestCommand(client *c);
|
|
|
|
void pfaddCommand(client *c);
|
|
|
|
void pfcountCommand(client *c);
|
|
|
|
void pfmergeCommand(client *c);
|
|
|
|
void pfdebugCommand(client *c);
|
|
|
|
void latencyCommand(client *c);
|
2016-03-06 13:44:24 +01:00
|
|
|
void moduleCommand(client *c);
|
2016-08-03 11:12:13 +02:00
|
|
|
void securityWarningCommand(client *c);
|
2017-08-30 12:40:27 +02:00
|
|
|
void xaddCommand(client *c);
|
|
|
|
void xrangeCommand(client *c);
|
2017-11-20 11:25:05 +01:00
|
|
|
void xrevrangeCommand(client *c);
|
2017-09-06 12:03:17 +02:00
|
|
|
void xlenCommand(client *c);
|
2017-09-08 11:40:16 +02:00
|
|
|
void xreadCommand(client *c);
|
2018-01-16 15:38:22 +01:00
|
|
|
void xgroupCommand(client *c);
|
2018-10-16 13:17:14 +02:00
|
|
|
void xsetidCommand(client *c);
|
2018-01-25 16:39:49 +01:00
|
|
|
void xackCommand(client *c);
|
2018-01-26 17:27:34 +01:00
|
|
|
void xpendingCommand(client *c);
|
2018-02-21 11:42:51 +01:00
|
|
|
void xclaimCommand(client *c);
|
2021-01-06 10:34:27 +02:00
|
|
|
void xautoclaimCommand(client *c);
|
2018-03-07 16:08:06 +01:00
|
|
|
void xinfoCommand(client *c);
|
2018-04-18 13:12:09 +02:00
|
|
|
void xdelCommand(client *c);
|
2018-04-19 16:25:29 +02:00
|
|
|
void xtrimCommand(client *c);
|
2018-09-12 11:34:10 +02:00
|
|
|
void lolwutCommand(client *c);
|
2019-01-15 09:36:12 +01:00
|
|
|
void aclCommand(client *c);
|
2021-11-18 10:47:49 +02:00
|
|
|
void lcsCommand(client *c);
|
2021-11-23 09:38:25 +01:00
|
|
|
void quitCommand(client *c);
|
2020-11-05 10:51:26 +02:00
|
|
|
void resetCommand(client *c);
|
2021-01-28 13:18:05 -08:00
|
|
|
void failoverCommand(client *c);
|
2010-06-22 00:07:48 +02:00
|
|
|
|
2010-07-27 09:36:42 +02:00
|
|
|
#if defined(__GNUC__)
|
2024-05-22 23:24:12 -07:00
|
|
|
void *calloc(size_t count, size_t size) __attribute__((deprecated));
|
|
|
|
void free(void *ptr) __attribute__((deprecated));
|
|
|
|
void *malloc(size_t size) __attribute__((deprecated));
|
|
|
|
void *realloc(void *ptr, size_t size) __attribute__((deprecated));
|
2010-07-27 09:36:42 +02:00
|
|
|
#endif
|
|
|
|
|
2011-10-04 17:22:29 +02:00
|
|
|
/* Debugging stuff */
|
2016-06-20 23:08:06 +03:00
|
|
|
void _serverAssertWithInfo(const client *c, const robj *o, const char *estr, const char *file, int line);
|
|
|
|
void _serverAssert(const char *estr, const char *file, int line);
|
2020-09-16 20:21:04 +03:00
|
|
|
#ifdef __GNUC__
|
2024-05-22 23:24:12 -07:00
|
|
|
void _serverPanic(const char *file, int line, const char *msg, ...) __attribute__((format(printf, 3, 4)));
|
2020-09-16 20:21:04 +03:00
|
|
|
#else
|
2017-01-18 17:05:10 +01:00
|
|
|
void _serverPanic(const char *file, int line, const char *msg, ...);
|
2020-09-16 20:21:04 +03:00
|
|
|
#endif
|
2016-06-20 23:08:06 +03:00
|
|
|
void serverLogObjectDebugInfo(const robj *o);
|
Print stack trace from all threads in crash report (#12453)
In this PR we are adding the functionality to collect all the process's threads' backtraces.
## Changes made in this PR
### **introduce threads mngr API**
The **threads mngr API** which has 2 abilities:
* `ThreadsManager_init() `- register to SIGUSR2. called on the server start-up.
* ` ThreadsManager_runOnThreads()` - receives a list of a pid_t and a callback, tells every
thread in the list to invoke the callback, and returns the output collected by each invocation.
**Elaborating atomicvar API**
* `atomicIncrGet(var,newvalue_var,count) `-- Increment and get the atomic counter new value
* `atomicFlagGetSet` -- Get and set the atomic counter value to 1
### **Always set SIGALRM handler**
SIGALRM handler prints the process's stacktrace to the log file. Up until now, it was set only if the
`server.watchdog_period` > 0. This can be also useful if debugging is needed. However, in situations
where the server can't get requests, (a deadlock, for example) we weren't able to change the signal handler.
To make it available at run time we set SIGALRM handler on server startup. The signal handler name was
changed to a more general `sigalrmSignalHandler`.
### **Print all the process' threads' stacktraces**
`logStackTrace()` now calls `writeStacktraces()`, instead of logging the current thread stacktrace.
`writeStacktraces()`:
* On Linux systems we use the threads manager API to collect the backtraces of all the process' threads.
To get the `tids` list (threads ids) we read the `/proc/<redis-server-pid>/tasks` file which includes a list of directories.
Each directory name corresponds to one tid (including the main thread). For each thread, we also need to check if it
can get the signal from the threads manager (meaning it is not blocking/ignoring that signal). We send the threads
manager this tids list and `collect_stacktrace_data()` callback, which collects the thread's backtrace addresses,
its name, and tid.
* On other systems, the behavior remained as it was (writing only the current thread stacktrace to the log file).
## compatibility notes
1. **The threads mngr API is only supported in linux.**
2. glibc earlier than 2.3 We use `syscall(SYS_gettid)` and `syscall(SYS_tgkill...)` because their dedicated
alternatives (`gettid()` and `tgkill`) were added in glibc 2.3.
## Output example
Each thread backtrace will have the following format:
`<tid> <thread_name> [additional_info]`
* **tid**: as read from the `/proc/<redis-server-pid>/tasks` file
* **thread_name**: the tread name as it is registered in the os/
* **additional_info**: Sometimes we want to add specific information about one of the threads. currently.
it is only used to mark the thread that handles the backtraces collection by adding "*".
In case of crash - this also indicates which thread caused the crash. The handling thread in won't
necessarily appear first.
```
------ STACK TRACE ------
EIP:
/lib/aarch64-linux-gnu/libc.so.6(epoll_pwait+0x9c)[0xffffb9295ebc]
67089 redis-server *
linux-vdso.so.1(__kernel_rt_sigreturn+0x0)[0xffffb9437790]
/lib/aarch64-linux-gnu/libc.so.6(epoll_pwait+0x9c)[0xffffb9295ebc]
redis-server *:6379(+0x75e0c)[0xaaaac2fe5e0c]
redis-server *:6379(aeProcessEvents+0x18c)[0xaaaac2fe6c00]
redis-server *:6379(aeMain+0x24)[0xaaaac2fe7038]
redis-server *:6379(main+0xe0c)[0xaaaac3001afc]
/lib/aarch64-linux-gnu/libc.so.6(+0x273fc)[0xffffb91d73fc]
/lib/aarch64-linux-gnu/libc.so.6(__libc_start_main+0x98)[0xffffb91d74cc]
redis-server *:6379(_start+0x30)[0xaaaac2fe0370]
67093 bio_lazy_free
/lib/aarch64-linux-gnu/libc.so.6(+0x79dfc)[0xffffb9229dfc]
/lib/aarch64-linux-gnu/libc.so.6(pthread_cond_wait+0x208)[0xffffb922c8fc]
redis-server *:6379(bioProcessBackgroundJobs+0x174)[0xaaaac30976e8]
/lib/aarch64-linux-gnu/libc.so.6(+0x7d5c8)[0xffffb922d5c8]
/lib/aarch64-linux-gnu/libc.so.6(+0xe5d1c)[0xffffb9295d1c]
67091 bio_close_file
/lib/aarch64-linux-gnu/libc.so.6(+0x79dfc)[0xffffb9229dfc]
/lib/aarch64-linux-gnu/libc.so.6(pthread_cond_wait+0x208)[0xffffb922c8fc]
redis-server *:6379(bioProcessBackgroundJobs+0x174)[0xaaaac30976e8]
/lib/aarch64-linux-gnu/libc.so.6(+0x7d5c8)[0xffffb922d5c8]
/lib/aarch64-linux-gnu/libc.so.6(+0xe5d1c)[0xffffb9295d1c]
67092 bio_aof
/lib/aarch64-linux-gnu/libc.so.6(+0x79dfc)[0xffffb9229dfc]
/lib/aarch64-linux-gnu/libc.so.6(pthread_cond_wait+0x208)[0xffffb922c8fc]
redis-server *:6379(bioProcessBackgroundJobs+0x174)[0xaaaac30976e8]
/lib/aarch64-linux-gnu/libc.so.6(+0x7d5c8)[0xffffb922d5c8]
/lib/aarch64-linux-gnu/libc.so.6(+0xe5d1c)[0xffffb9295d1c]
67089:signal-handler (1693824528) --------
```
2023-09-24 09:47:23 +03:00
|
|
|
void setupDebugSigHandlers(void);
|
Signal handler attributes (#12426)
This PR purpose is to make the crash report process thread safe.
main changes include:
1. `setupSigSegvHandler()` is introduced to initialize the signal handler.
This function first initializes the signal handler mutex (if not initialized yet)
and then registers the process to the signal handler.
2. **sigsegvHandler** flags :
SA_NODEFER - don't add the signal to the process signal mask. We use this
flag because we want to be able to handle a second call to the signal manually.
removed SA_RESETHAND: this flag resets the signal handler function upon the first
entrance to the registered function. The reason to use this flag is to protect from
recursively entering the signal handler by the same thread. But, it also means
that if a second thread crashes while handling a signal, the process will be
terminated immediately and we won't get the crash report.
In this PR we discard this flag. The signal handler guard described below purpose
is to solve the above issues.
3. Add a **signal handler lock** with ERRORCHECK attributes.
The lock's purpose is to ensure that only one thread generates a crash report.
Once a second thread enters the signal handler it will be blocked.
We use the ERRORCHECK lock in order to protect from possible deadlock in
case the thread handling the crash gets a signal. In the latest scenario, we log
what we have collected until the handler crashed.
At the end of the crash report we reset the signal handler SIG_DFL, with no flags, and
rethrow the signal to generate a core dump (if enabled) and exit the process.
During the work on this PR we wanted to understand the historical reasons for
how crash is handled.
With respect to the choice of the flag, we believe the **SA_RESETHAND** was not
added for any specific purpose.
**SA_ONSTACK** which is removed here from bugReportEnd(), was originally also
set in the initial registration to signal handler, but removed in 3ada43e73. In addition,
it was removed from another location in deee2c1ef with the following description,
which is also relevant to why it should be removed from bugReportEnd:
> it seems to be some valgrind bug with SA_ONSTACK.
> SA_ONSTACK seems unneeded since WD is not recursive (SA_NODEFER was removed),
> also, not sure if it's even valid without a call to sigaltstack()
2023-08-20 19:16:45 +03:00
|
|
|
void setupSigSegvHandler(void);
|
|
|
|
void removeSigSegvHandlers(void);
|
2021-02-15 17:08:53 +02:00
|
|
|
const char *getSafeInfoString(const char *s, size_t len, char **tmp);
|
Make INFO command variadic (#6891)
This is an enhancement for INFO command, previously INFO only support one argument
for different info section , if user want to get more categories information, either perform
INFO all / default or calling INFO for multiple times.
**Description of the feature**
The goal of adding this feature is to let the user retrieve multiple categories via the INFO
command, and still avoid emitting the same section twice.
A use case for this is like Redis Sentinel, which periodically calling INFO command to refresh
info from monitored Master/Slaves, only Server and Replication part categories are used for
parsing information. If the INFO command can return just enough categories that client side
needs, it can save a lot of time for client side parsing it as well as network bandwidth.
**Implementation**
To share code between redis, sentinel, and other users of INFO (DEBUG and modules),
we have a new `genInfoSectionDict` function that returns a dict and some boolean flags
(e.g. `all`) to the caller (built from user input).
Sentinel is later purging unwanted sections from that, and then it is forwarded to the info `genRedisInfoString`.
**Usage Examples**
INFO Server Replication
INFO CPU Memory
INFO default commandstats
Co-authored-by: Oran Agra <oran@redislabs.com>
2022-02-08 06:14:42 -05:00
|
|
|
dict *genInfoSectionDict(robj **argv, int argc, char **defaults, int *out_all, int *out_everything);
|
|
|
|
void releaseInfoSectionDict(dict *sec);
|
2024-05-08 09:44:05 -04:00
|
|
|
sds genValkeyInfoString(dict *section_dict, int all_sections, int everything);
|
2019-04-16 22:16:12 +03:00
|
|
|
sds genModulesInfoString(sds info);
|
2023-05-02 17:31:32 -07:00
|
|
|
void applyWatchdogPeriod(void);
|
2012-03-27 11:47:51 +02:00
|
|
|
void watchdogScheduleSignal(int period);
|
2015-07-26 15:17:43 +02:00
|
|
|
void serverLogHexDump(int level, char *descr, void *value, size_t len);
|
2015-12-16 17:41:20 +01:00
|
|
|
int memtest_preserving_test(unsigned long *m, size_t bytes, int passes);
|
2022-01-18 21:55:20 +08:00
|
|
|
void mixDigest(unsigned char *digest, const void *ptr, size_t len);
|
|
|
|
void xorDigest(unsigned char *digest, const void *ptr, size_t len);
|
2022-01-23 16:05:06 +08:00
|
|
|
sds catSubCommandFullname(const char *parent_name, const char *sub_name);
|
2024-04-03 23:54:33 +07:00
|
|
|
void commandAddSubcommand(struct serverCommand *parent, struct serverCommand *subcommand, const char *declared_name);
|
2020-09-03 08:47:29 +03:00
|
|
|
void debugDelay(int usec);
|
2020-09-20 05:06:17 -04:00
|
|
|
void killThreads(void);
|
2020-09-18 16:08:52 +08:00
|
|
|
void makeThreadKillable(void);
|
2024-04-03 10:02:43 +07:00
|
|
|
void swapMainDbWithTempDb(serverDb *tempDb);
|
2023-12-21 19:51:46 +08:00
|
|
|
sds getVersion(void);
|
2024-07-25 19:34:39 +03:00
|
|
|
void debugPauseProcess(void);
|
2012-07-23 12:54:52 +02:00
|
|
|
|
2021-03-24 08:22:12 +02:00
|
|
|
/* Use macro for checking log level to avoid evaluating arguments in cases log
|
|
|
|
* should be ignored due to low level. */
|
2024-05-22 23:24:12 -07:00
|
|
|
#define serverLog(level, ...) \
|
|
|
|
do { \
|
|
|
|
if (((level) & 0xff) < server.verbosity) break; \
|
|
|
|
_serverLog(level, __VA_ARGS__); \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
#define serverDebug(fmt, ...) printf("DEBUG %s:%d > " fmt "\n", __FILE__, __LINE__, __VA_ARGS__)
|
|
|
|
#define serverDebugMark() printf("-- MARK %s:%d --\n", __FILE__, __LINE__)
|
2012-07-23 12:54:52 +02:00
|
|
|
|
2024-06-07 14:21:33 -07:00
|
|
|
int iAmPrimary(void);
|
Fix master replica inconsistency for upgrading scenario.
Before this commit, when upgrading a replica, expired keys will not
be loaded, thus causing replica having less keys in db. To this point,
master and replica's keys is logically consistent. However, before
the keys in master and replica are physically consistent, that is,
they have the same dbsize, if master got a problem and the replica
got promoted and becomes new master of that partition, and master
updates a key which does not exist on master, but physically exists
on the old master(new replica), the old master would refuse to update
the key, thus causing master and replica data inconsistent.
How could this happen?
That's all because of the wrong judgement of roles while starting up
the server. We can not use server.masterhost to judge if the server
is master or replica, since it fails in cluster mode.
When we start the server, we load rdb and do want to load expired keys,
and do not want to have the ability to active expire keys, if it is
a replica.
2020-03-18 16:20:10 +08:00
|
|
|
|
2022-01-02 19:48:29 -08:00
|
|
|
#define STRINGIFY_(x) #x
|
|
|
|
#define STRINGIFY(x) STRINGIFY_(x)
|
|
|
|
|
2010-06-22 00:07:48 +02:00
|
|
|
#endif
|