futriix/src/db.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

2689 lines
101 KiB
C
Raw Normal View History

/*
* Copyright (c) 2009-2012, Redis Ltd.
* 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.
*/
#include "server.h"
#include "cluster.h"
#include "latency.h"
#include "script.h"
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
#include "functions.h"
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
#include "io_threads.h"
#include <signal.h>
#include <ctype.h>
/*-----------------------------------------------------------------------------
* C-level DB API
*----------------------------------------------------------------------------*/
/* Flags for expireIfNeeded */
#define EXPIRE_FORCE_DELETE_EXPIRED 1
#define EXPIRE_AVOID_DELETE_EXPIRED 2
/* Return values for expireIfNeeded */
typedef enum {
KEY_VALID = 0, /* Could be volatile and not yet expired, non-volatile, or even non-existing key. */
KEY_EXPIRED, /* Logically expired but not yet deleted. */
KEY_DELETED /* The key was deleted now. */
} keyStatus;
keyStatus expireIfNeededWithDictIndex(serverDb *db, robj *key, int flags, int dict_index);
keyStatus expireIfNeeded(serverDb *db, robj *key, int flags);
int keyIsExpiredWithDictIndex(serverDb *db, robj *key, int dict_index);
int keyIsExpired(serverDb *db, robj *key);
static void dbSetValue(serverDb *db, robj *key, robj *val, int overwrite, dictEntry *de);
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
static int getKVStoreIndexForKey(sds key);
dictEntry *dbFindExpiresWithDictIndex(serverDb *db, void *key, int dict_index);
/* Update LFU when an object is accessed.
* Firstly, decrement the counter if the decrement time is reached.
* Then logarithmically increment the counter, and update the access time. */
void updateLFU(robj *val) {
unsigned long counter = LFUDecrAndReturn(val);
counter = LFULogIncr(counter);
val->lru = (LFUGetTimeInMinutes() << 8) | counter;
}
/* Lookup a key for read or write operations, or return NULL if the key is not
* found in the specified DB. This function implements the functionality of
* lookupKeyRead(), lookupKeyWrite() and their ...WithFlags() variants.
*
* Side-effects of calling this function:
*
* 1. A key gets expired if it reached it's TTL.
* 2. The key's last access time is updated.
* 3. The global keys hits/misses stats are updated (reported in INFO).
* 4. If keyspace notifications are enabled, a "keymiss" notification is fired.
*
* Flags change the behavior of this command:
*
* LOOKUP_NONE (or zero): No special flags are passed.
* LOOKUP_NOTOUCH: Don't alter the last access time of the key.
* LOOKUP_NONOTIFY: Don't trigger keyspace event on key miss.
* LOOKUP_NOSTATS: Don't increment key hits/misses counters.
* LOOKUP_WRITE: Prepare the key for writing (delete expired keys even on
* replicas, use separate keyspace stats and events (TODO)).
* LOOKUP_NOEXPIRE: Perform expiration check, but avoid deleting the key,
* so that we don't have to propagate the deletion.
*
* Note: this function also returns NULL if the key is logically expired but
* still existing, in case this is a replica and the LOOKUP_WRITE is not set.
* Even if the key expiry is primary-driven, we can correctly report a key is
* expired on replicas even if the primary is lagging expiring our key via DELs
* in the replication link. */
robj *lookupKey(serverDb *db, robj *key, int flags) {
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
dictEntry *de = dbFind(db, key->ptr);
robj *val = NULL;
if (de) {
val = dictGetVal(de);
/* Forcing deletion of expired keys on a replica makes the replica
* inconsistent with the primary. We forbid it on readonly replicas, but
* we have to allow it on writable replicas to make write commands
* behave consistently.
*
* It's possible that the WRITE flag is set even during a readonly
* command, since the command may trigger events that cause modules to
* perform additional writes. */
int is_ro_replica = server.primary_host && server.repl_replica_ro;
int expire_flags = 0;
if (flags & LOOKUP_WRITE && !is_ro_replica) expire_flags |= EXPIRE_FORCE_DELETE_EXPIRED;
if (flags & LOOKUP_NOEXPIRE) expire_flags |= EXPIRE_AVOID_DELETE_EXPIRED;
if (expireIfNeeded(db, key, expire_flags) != KEY_VALID) {
/* The key is no longer valid. */
val = NULL;
}
}
if (val) {
2013-01-17 01:00:20 +08:00
/* Update the access time for the ageing algorithm.
* Don't do it if we have a saving child, as this will trigger
* a copy on write madness. */
if (server.current_client && server.current_client->flag.no_touch &&
server.current_client->cmd->proc != touchCommand)
flags |= LOOKUP_NOTOUCH;
if (!hasActiveChildProcess() && !(flags & LOOKUP_NOTOUCH)) {
if (!canUseSharedObject() && val->refcount == OBJ_SHARED_REFCOUNT) {
val = dupStringObject(val);
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
kvstoreDictSetVal(db->keys, getKVStoreIndexForKey(key->ptr), de, val);
}
if (server.maxmemory_policy & MAXMEMORY_FLAG_LFU) {
updateLFU(val);
} else {
val->lru = LRU_CLOCK();
}
}
if (!(flags & (LOOKUP_NOSTATS | LOOKUP_WRITE))) server.stat_keyspace_hits++;
/* TODO: Use separate hits stats for WRITE */
} else {
if (!(flags & (LOOKUP_NONOTIFY | LOOKUP_WRITE))) notifyKeyspaceEvent(NOTIFY_KEY_MISS, "keymiss", key, db->id);
if (!(flags & (LOOKUP_NOSTATS | LOOKUP_WRITE))) server.stat_keyspace_misses++;
/* TODO: Use separate misses stats and notify event for WRITE */
}
return val;
}
/* Lookup a key for read operations, or return NULL if the key is not found
* in the specified DB.
*
* This API should not be used when we write to the key after obtaining
* the object linked to the key, but only for read only operations.
*
* This function is equivalent to lookupKey(). The point of using this function
* rather than lookupKey() directly is to indicate that the purpose is to read
* the key. */
robj *lookupKeyReadWithFlags(serverDb *db, robj *key, int flags) {
serverAssert(!(flags & LOOKUP_WRITE));
return lookupKey(db, key, flags);
}
/* Like lookupKeyReadWithFlags(), but does not use any flag, which is the
* common case. */
robj *lookupKeyRead(serverDb *db, robj *key) {
return lookupKeyReadWithFlags(db, key, LOOKUP_NONE);
}
/* Lookup a key for write operations, and as a side effect, if needed, expires
* the key if its TTL is reached. It's equivalent to lookupKey() with the
* LOOKUP_WRITE flag added.
*
* Returns the linked value object if the key exists or NULL if the key
* does not exist in the specified DB. */
robj *lookupKeyWriteWithFlags(serverDb *db, robj *key, int flags) {
return lookupKey(db, key, flags | LOOKUP_WRITE);
}
robj *lookupKeyWrite(serverDb *db, robj *key) {
return lookupKeyWriteWithFlags(db, key, LOOKUP_NONE);
}
robj *lookupKeyReadOrReply(client *c, robj *key, robj *reply) {
robj *o = lookupKeyRead(c->db, key);
if (!o) addReplyOrErrorObject(c, reply);
return o;
}
robj *lookupKeyWriteOrReply(client *c, robj *key, robj *reply) {
robj *o = lookupKeyWrite(c->db, key);
if (!o) addReplyOrErrorObject(c, reply);
return o;
}
Embed key into dict entry (#541) This PR incorporates changes related to key embedding described in the https://github.com/redis/redis/issues/12216 With this change there will be no `key` pointer and embedded the key within the `dictEntry`. 1 byte is used for additional bookkeeping. Overall the saving would be 7 bytes on average. Key changes: New dict entry type introduced, which is now used as an entry for the main dictionary: ```c typedef struct { union { void *val; uint64_t u64; int64_t s64; double d; } v; struct dictEntry *next; /* Next entry in the same hash bucket. */ uint8_t key_header_size; /* offset into key_buf where the key is located at. */ unsigned char key_buf[]; /* buffer with embedded key. */ } embeddedDictEntry; ``` One new function has been added to the dictType: ```c size_t (*embedKey)(unsigned char *buf, size_t buf_len, const void *key, unsigned char *header_size); ``` Change is opt-in per dict type, hence sets, hashes and other types that are using dictionary are not impacted. With this change main dictionary now owns the data, so copy on insert in dbAdd is no longer needed. ### Benchmarking results TLDR; Around 9-10% memory usage reduction in overall memory usage for scenario with key of 16 bytes and value of 8 bytes and 16 bytes. The throughput per second varies but is similar or greater in most of the run(s) with the changes against unstable (ae2d421). --------- Signed-off-by: Harkrishn Patro <harkrisp@amazon.com> Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-02 15:45:37 -07:00
/* Add the key to the DB.
*
* In this case a copy of `key` is copied in kvstore, the caller must ensure the `key` is properly freed.
*
* It's up to the caller to increment the reference
2013-01-17 01:00:20 +08:00
* counter of the value if needed.
*
* If the update_if_existing argument is false, the program is aborted
* if the key already exists, otherwise, it can fall back to dbOverwrite. */
static void dbAddInternal(serverDb *db, robj *key, robj *val, int update_if_existing) {
dictEntry *existing;
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
int dict_index = getKVStoreIndexForKey(key->ptr);
dictEntry *de = kvstoreDictAddRaw(db->keys, dict_index, key->ptr, &existing);
if (update_if_existing && existing) {
dbSetValue(db, key, val, 1, existing);
return;
}
serverAssertWithInfo(NULL, key, de != NULL);
initObjectLRUOrLFU(val);
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
kvstoreDictSetVal(db->keys, dict_index, de, val);
signalKeyAsReady(db, key, val->type);
notifyKeyspaceEvent(NOTIFY_NEW, "new", key, db->id);
}
void dbAdd(serverDb *db, robj *key, robj *val) {
dbAddInternal(db, key, val, 0);
}
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
/* Returns which dict index should be used with kvstore for a given key. */
static int getKVStoreIndexForKey(sds key) {
return server.cluster_enabled ? getKeySlot(key) : 0;
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
}
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
/* Returns the cluster hash slot for a given key, trying to use the cached slot that
* stored on the server.current_client first. If there is no cached value, it will compute the hash slot
* and then cache the value.*/
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) {
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
serverAssert(server.cluster_enabled);
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
/* This is performance optimization that uses pre-set slot id from the current command,
* in order to avoid calculation of the key hash.
*
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
* This optimization is only used when current_client flag `CLIENT_EXECUTING_COMMAND` is set.
* It only gets set during the execution of command under `call` method. Other flows requesting
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
* the key slot would fallback to keyHashSlot.
*
* Modules and scripts executed on the primary may get replicated as multi-execs that operate on multiple slots,
* so we must always recompute the slot for commands coming from the primary.
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
*/
if (server.current_client && server.current_client->slot >= 0 && server.current_client->flag.executing_command &&
!server.current_client->flag.primary) {
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
debugServerAssertWithInfo(server.current_client, NULL,
(int)keyHashSlot(key, (int)sdslen(key)) == server.current_client->slot);
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
return server.current_client->slot;
}
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
int slot = keyHashSlot(key, (int)sdslen(key));
/* For the case of replicated commands from primary, getNodeByQuery() never gets called,
* and thus c->slot never gets populated. That said, if this command ends up accessing a key,
* we are able to backfill c->slot here, where the key's hash calculation is made. */
if (server.current_client && server.current_client->flag.primary) {
server.current_client->slot = slot;
}
return slot;
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
}
/* This is a special version of dbAdd() that is used only when loading
* keys from the RDB file: the key is passed as an SDS string that is
Embed key into dict entry (#541) This PR incorporates changes related to key embedding described in the https://github.com/redis/redis/issues/12216 With this change there will be no `key` pointer and embedded the key within the `dictEntry`. 1 byte is used for additional bookkeeping. Overall the saving would be 7 bytes on average. Key changes: New dict entry type introduced, which is now used as an entry for the main dictionary: ```c typedef struct { union { void *val; uint64_t u64; int64_t s64; double d; } v; struct dictEntry *next; /* Next entry in the same hash bucket. */ uint8_t key_header_size; /* offset into key_buf where the key is located at. */ unsigned char key_buf[]; /* buffer with embedded key. */ } embeddedDictEntry; ``` One new function has been added to the dictType: ```c size_t (*embedKey)(unsigned char *buf, size_t buf_len, const void *key, unsigned char *header_size); ``` Change is opt-in per dict type, hence sets, hashes and other types that are using dictionary are not impacted. With this change main dictionary now owns the data, so copy on insert in dbAdd is no longer needed. ### Benchmarking results TLDR; Around 9-10% memory usage reduction in overall memory usage for scenario with key of 16 bytes and value of 8 bytes and 16 bytes. The throughput per second varies but is similar or greater in most of the run(s) with the changes against unstable (ae2d421). --------- Signed-off-by: Harkrishn Patro <harkrisp@amazon.com> Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-02 15:45:37 -07:00
* copied by the function and freed by the caller.
*
* Moreover this function will not abort if the key is already busy, to
* give more control to the caller, nor will signal the key as ready
* since it is not useful in this context.
*
Embed key into dict entry (#541) This PR incorporates changes related to key embedding described in the https://github.com/redis/redis/issues/12216 With this change there will be no `key` pointer and embedded the key within the `dictEntry`. 1 byte is used for additional bookkeeping. Overall the saving would be 7 bytes on average. Key changes: New dict entry type introduced, which is now used as an entry for the main dictionary: ```c typedef struct { union { void *val; uint64_t u64; int64_t s64; double d; } v; struct dictEntry *next; /* Next entry in the same hash bucket. */ uint8_t key_header_size; /* offset into key_buf where the key is located at. */ unsigned char key_buf[]; /* buffer with embedded key. */ } embeddedDictEntry; ``` One new function has been added to the dictType: ```c size_t (*embedKey)(unsigned char *buf, size_t buf_len, const void *key, unsigned char *header_size); ``` Change is opt-in per dict type, hence sets, hashes and other types that are using dictionary are not impacted. With this change main dictionary now owns the data, so copy on insert in dbAdd is no longer needed. ### Benchmarking results TLDR; Around 9-10% memory usage reduction in overall memory usage for scenario with key of 16 bytes and value of 8 bytes and 16 bytes. The throughput per second varies but is similar or greater in most of the run(s) with the changes against unstable (ae2d421). --------- Signed-off-by: Harkrishn Patro <harkrisp@amazon.com> Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
2024-07-02 15:45:37 -07:00
* The function returns 1 if the key was added to the database, otherwise 0 is returned.
*
* In this case a copy of `key` is copied in kvstore, the caller must ensure the `key` is properly freed.
*/
int dbAddRDBLoad(serverDb *db, sds key, robj *val) {
int dict_index = getKVStoreIndexForKey(key);
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
dictEntry *de = kvstoreDictAddRaw(db->keys, dict_index, key, NULL);
if (de == NULL) return 0;
initObjectLRUOrLFU(val);
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
kvstoreDictSetVal(db->keys, dict_index, de, val);
return 1;
}
/* Overwrite an existing key with a new value. Incrementing the reference
* count of the new value is up to the caller.
* This function does not modify the expire time of the existing key.
*
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
* The 'overwrite' flag is an indication whether this is done as part of a
* complete replacement of their key, which can be thought as a deletion and
* replacement (in which case we need to emit deletion signals), or just an
* update of a value of an existing key (when false).
*
* The dictEntry input is optional, can be used if we already have one.
*
* The program is aborted if the key was not already present. */
static void dbSetValue(serverDb *db, robj *key, robj *val, int overwrite, dictEntry *de) {
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
int dict_index = getKVStoreIndexForKey(key->ptr);
if (!de) de = kvstoreDictFind(db->keys, dict_index, key->ptr);
2015-07-26 15:29:53 +02:00
serverAssertWithInfo(NULL, key, de != NULL);
robj *old = dictGetVal(de);
val->lru = old->lru;
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
if (overwrite) {
/* RM_StringDMA may call dbUnshareStringValue which may free val, so we
* need to incr to retain old */
incrRefCount(old);
/* Although the key is not really deleted from the database, we regard
* overwrite as two steps of unlink+add, so we still need to call the unlink
* callback of the module. */
moduleNotifyKeyUnlink(key, old, db->id, DB_FLAG_KEY_OVERWRITE);
/* We want to try to unblock any module clients or clients using a blocking XREADGROUP */
signalDeletedKeyAsReady(db, key, old->type);
decrRefCount(old);
/* Because of RM_StringDMA, old may be changed, so we need get old again */
old = dictGetVal(de);
}
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
kvstoreDictSetVal(db->keys, dict_index, de, val);
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
/* For efficiency, let the I/O thread that allocated an object also deallocate it. */
if (tryOffloadFreeObjToIOThreads(old) == C_OK) {
/* OK */
} else if (server.lazyfree_lazy_server_del) {
freeObjAsync(key, old, db->id);
} else {
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
decrRefCount(old);
}
}
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
/* Replace an existing key with a new value, we just replace value and don't
* emit any events */
void dbReplaceValue(serverDb *db, robj *key, robj *val) {
dbSetValue(db, key, val, 0, NULL);
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
}
/* High level Set operation. This function can be used in order to set
* a key, whatever it was existing or not, to a new object.
*
* 1) The ref count of the value object is incremented.
* 2) clients WATCHing for the destination key notified.
* 3) The expire time of the key is reset (the key is made persistent),
* unless 'SETKEY_KEEPTTL' is enabled in flags.
* 4) The key lookup can take place outside this interface outcome will be
* delivered with 'SETKEY_ALREADY_EXIST' or 'SETKEY_DOESNT_EXIST'
Replication: fix the infamous key leakage of writable slaves + EXPIRE. BACKGROUND AND USE CASEj Redis slaves are normally write only, however the supprot a "writable" mode which is very handy when scaling reads on slaves, that actually need write operations in order to access data. For instance imagine having slaves replicating certain Sets keys from the master. When accessing the data on the slave, we want to peform intersections between such Sets values. However we don't want to intersect each time: to cache the intersection for some time often is a good idea. To do so, it is possible to setup a slave as a writable slave, and perform the intersection on the slave side, perhaps setting a TTL on the resulting key so that it will expire after some time. THE BUG Problem: in order to have a consistent replication, expiring of keys in Redis replication is up to the master, that synthesize DEL operations to send in the replication stream. However slaves logically expire keys by hiding them from read attempts from clients so that if the master did not promptly sent a DEL, the client still see logically expired keys as non existing. Because slaves don't actively expire keys by actually evicting them but just masking from the POV of read operations, if a key is created in a writable slave, and an expire is set, the key will be leaked forever: 1. No DEL will be received from the master, which does not know about such a key at all. 2. No eviction will be performed by the slave, since it needs to disable eviction because it's up to masters, otherwise consistency of data is lost. THE FIX In order to fix the problem, the slave should be able to tag keys that were created in the slave side and have an expire set in some way. My solution involved using an unique additional dictionary created by the writable slave only if needed. The dictionary is obviously keyed by the key name that we need to track: all the keys that are set with an expire directly by a client writing to the slave are tracked. The value in the dictionary is a bitmap of all the DBs where such a key name need to be tracked, so that we can use a single dictionary to track keys in all the DBs used by the slave (actually this limits the solution to the first 64 DBs, but the default with Redis is to use 16 DBs). This solution allows to pay both a small complexity and CPU penalty, which is zero when the feature is not used, actually. The slave-side eviction is encapsulated in code which is not coupled with the rest of the Redis core, if not for the hook to track the keys. TODO I'm doing the first smoke tests to see if the feature works as expected: so far so good. Unit tests should be added before merging into the 4.0 branch.
2016-12-13 10:20:06 +01:00
*
* All the new keys in the database should be created via this interface.
* The client 'c' argument may be set to NULL if the operation is performed
* in a context where there is no clear client performing the operation. */
void setKey(client *c, serverDb *db, robj *key, robj *val, int flags) {
int keyfound = 0;
if (flags & SETKEY_ALREADY_EXIST)
keyfound = 1;
else if (flags & SETKEY_ADD_OR_UPDATE)
keyfound = -1;
else if (!(flags & SETKEY_DOESNT_EXIST))
keyfound = (lookupKeyWrite(db, key) != NULL);
if (!keyfound) {
dbAdd(db, key, val);
} else if (keyfound < 0) {
dbAddInternal(db, key, val, 1);
} else {
dbSetValue(db, key, val, 1, NULL);
}
incrRefCount(val);
if (!(flags & SETKEY_KEEPTTL)) removeExpire(db, key);
if (!(flags & SETKEY_NO_SIGNAL)) signalModifiedKey(c, db, key);
}
/* Return a random key, in form of an Object.
* If there are no keys, NULL is returned.
*
* The function makes sure to return keys not already expired. */
robj *dbRandomKey(serverDb *db) {
2014-03-20 16:20:37 +01:00
dictEntry *de;
int maxtries = 100;
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 allvolatile = kvstoreSize(db->keys) == kvstoreSize(db->expires);
while (1) {
sds key;
robj *keyobj;
int randomDictIndex = kvstoreGetFairRandomDictIndex(db->keys);
de = kvstoreDictGetFairRandomKey(db->keys, randomDictIndex);
if (de == NULL) return NULL;
key = dictGetKey(de);
keyobj = createStringObject(key, sdslen(key));
if (dbFindExpiresWithDictIndex(db, key, randomDictIndex)) {
if (allvolatile && server.primary_host && --maxtries == 0) {
/* If the DB is composed only of keys with an expire set,
* it could happen that all the keys are already logically
* expired in the repilca, so the function cannot stop because
* expireIfNeeded() is false, nor it can stop because
* dictGetFairRandomKey() returns NULL (there are keys to return).
* To prevent the infinite loop we do some tries, but if there
* are the conditions for an infinite loop, eventually we
* return a key name that may be already expired. */
return keyobj;
}
if (expireIfNeededWithDictIndex(db, keyobj, 0, randomDictIndex) != KEY_VALID) {
decrRefCount(keyobj);
continue; /* search for another key. This expired. */
}
}
return keyobj;
}
}
int dbGenericDeleteWithDictIndex(serverDb *db, robj *key, int async, int flags, int dict_index) {
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
dictEntry **plink;
int table;
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
dictEntry *de = kvstoreDictTwoPhaseUnlinkFind(db->keys, dict_index, key->ptr, &plink, &table);
if (de) {
robj *val = dictGetVal(de);
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
/* RM_StringDMA may call dbUnshareStringValue which may free val, so we
* need to incr to retain val */
incrRefCount(val);
/* Tells the module that the key has been unlinked from the database. */
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
moduleNotifyKeyUnlink(key, val, db->id, flags);
Blocked module clients should be aware when a key is deleted (#11310) The use case is a module that wants to implement a blocking command on a key that necessarily exists and wants to unblock the client in case the key is deleted (much like what we implemented for XREADGROUP in #10306) New module API: * RedisModule_BlockClientOnKeysWithFlags Flags: * REDISMODULE_BLOCK_UNBLOCK_NONE * REDISMODULE_BLOCK_UNBLOCK_DELETED ### Detailed description of code changes blocked.c: 1. Both module and stream functions are called whether the key exists or not, regardless of its type. We do that in order to allow modules/stream to unblock the client in case the key is no longer present or has changed type (the behavior for streams didn't change, just code that moved into serveClientsBlockedOnStreamKey) 2. Make sure afterCommand is called in serveClientsBlockedOnKeyByModule, in order to propagate actions from moduleTryServeClientBlockedOnKey. 3. handleClientsBlockedOnKeys: call propagatePendingCommands directly after lookupKeyReadWithFlags to prevent a possible lazy-expire DEL from being mixed with any command propagated by the preceding functions. 4. blockForKeys: Caller can specifiy that it wants to be awakened if key is deleted. Minor optimizations (use dictAddRaw). 5. signalKeyAsReady became signalKeyAsReadyLogic which can take a boolean in case the key is deleted. It will only signal if there's at least one client that awaits key deletion (to save calls to handleClientsBlockedOnKeys). Minor optimizations (use dictAddRaw) db.c: 1. scanDatabaseForDeletedStreams is now scanDatabaseForDeletedKeys and will signalKeyAsReady for any key that was removed from the database or changed type. It is the responsibility of the code in blocked.c to ignore or act on deleted/type-changed keys. 2. Use the new signalDeletedKeyAsReady where needed blockedonkey.c + tcl: 1. Added test of new capabilities (FSL.BPOPGT now requires the key to exist in order to work)
2022-10-18 18:50:02 +02:00
/* We want to try to unblock any module clients or clients using a blocking XREADGROUP */
signalDeletedKeyAsReady(db, key, val->type);
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
/* We should call decr before freeObjAsync. If not, the refcount may be
* greater than 1, so freeObjAsync doesn't work */
decrRefCount(val);
if (async) {
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
/* Because of dbUnshareStringValue, the val in de may change. */
freeObjAsync(key, dictGetVal(de), db->id);
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
kvstoreDictSetVal(db->keys, dict_index, de, NULL);
}
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
/* Deleting an entry from the expires dict will not free the sds of
* the key, because it is shared with the main dictionary. */
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
kvstoreDictDelete(db->expires, dict_index, key->ptr);
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
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
kvstoreDictTwoPhaseUnlinkFree(db->keys, dict_index, de, plink, table);
return 1;
} else {
return 0;
}
}
/* Helper for sync and async delete. */
int dbGenericDelete(serverDb *db, robj *key, int async, int flags) {
int dict_index = getKVStoreIndexForKey(key->ptr);
return dbGenericDeleteWithDictIndex(db, key, async, flags, dict_index);
}
/* Delete a key, value, and associated expiration entry if any, from the DB */
int dbSyncDelete(serverDb *db, robj *key) {
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
return dbGenericDelete(db, key, 0, DB_FLAG_KEY_DELETED);
}
/* Delete a key, value, and associated expiration entry if any, from the DB. If
* the value consists of many allocations, it may be freed asynchronously. */
int dbAsyncDelete(serverDb *db, robj *key) {
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
return dbGenericDelete(db, key, 1, DB_FLAG_KEY_DELETED);
}
/* This is a wrapper whose behavior depends on the server lazy free
* configuration. Deletes the key synchronously or asynchronously. */
int dbDelete(serverDb *db, robj *key) {
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
return dbGenericDelete(db, key, server.lazyfree_lazy_server_del, DB_FLAG_KEY_DELETED);
}
/* Prepare the string object stored at 'key' to be modified destructively
* to implement commands like SETBIT or APPEND.
*
* An object is usually ready to be modified unless one of the two conditions
* are true:
*
* 1) The object 'o' is shared (refcount > 1), we don't want to affect
* other users.
* 2) The object encoding is not "RAW".
*
* If the object is found in one of the above conditions (or both) by the
* function, an unshared / not-encoded copy of the string object is stored
* at 'key' in the specified 'db'. Otherwise the object 'o' itself is
* returned.
*
* USAGE:
*
* The object 'o' is what the caller already obtained by looking up 'key'
* in 'db', the usage pattern looks like this:
*
* o = lookupKeyWrite(db,key);
* if (checkType(c,o,OBJ_STRING)) return;
* o = dbUnshareStringValue(db,key,o);
*
* At this point the caller is ready to modify the object, for example
* using an sdscat() call to append some data, or anything else.
*/
robj *dbUnshareStringValue(serverDb *db, robj *key, robj *o) {
2015-07-26 15:29:53 +02:00
serverAssert(o->type == OBJ_STRING);
if (o->refcount != 1 || o->encoding != OBJ_ENCODING_RAW) {
robj *decoded = getDecodedObject(o);
o = createRawStringObject(decoded->ptr, sdslen(decoded->ptr));
decrRefCount(decoded);
Add a special notification unlink available only for modules (#9406) Add a new module event `RedisModule_Event_Key`, this event is fired when a key is removed from the keyspace. The event includes an open key that can be used for reading the key before it is removed. Modules can also extract the key-name, and use RM_Open or RM_Call to access key from within that event, but shouldn't modify anything from within this event. The following sub events are available: - `REDISMODULE_SUBEVENT_KEY_DELETED` - `REDISMODULE_SUBEVENT_KEY_EXPIRED` - `REDISMODULE_SUBEVENT_KEY_EVICTED` - `REDISMODULE_SUBEVENT_KEY_OVERWRITE` The data pointer can be casted to a RedisModuleKeyInfo structure with the following fields: ``` RedisModuleKey *key; // Opened Key ``` ### internals * We also add two dict functions: `dictTwoPhaseUnlinkFind` finds an element from the table, also get the plink of the entry. The entry is returned if the element is found. The user should later call `dictTwoPhaseUnlinkFree` with it in order to unlink and release it. Otherwise if the key is not found, NULL is returned. These two functions should be used in pair. `dictTwoPhaseUnlinkFind` pauses rehash and `dictTwoPhaseUnlinkFree` resumes rehash. * We change `dbOverwrite` to `dbReplaceValue` which just replaces the value of the key and doesn't fire any events. The "overwrite" part (which emits events) is just when called from `setKey`, the other places that called dbOverwrite were ones that just update the value in-place (INCR*, SPOP, and dbUnshareStringValue). This should not have any real impact since `moduleNotifyKeyUnlink` and `signalDeletedKeyAsReady` wouldn't have mattered in these cases anyway (i.e. module keys and stream keys didn't have direct calls to dbOverwrite) * since we allow doing RM_OpenKey from withing these callbacks, we temporarily disable lazy expiry. * We also temporarily disable lazy expiry when we are in unlink/unlink2 callback and keyspace notification callback. * Move special definitions to the top of redismodule.h This is needed to resolve compilation errors with RedisModuleKeyInfoV1 that carries a RedisModuleKey member. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-11-30 17:56:36 +08:00
dbReplaceValue(db, key, o);
}
return o;
}
/* Remove all keys from the database(s) structure. The dbarray argument
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
* may not be the server main DBs (could be a temporary DB).
*
* The dbnum can be -1 if all the DBs should be emptied, or the specified
* DB index if we want to empty only a single database.
* The function returns the number of keys removed from the database(s). */
long long emptyDbStructure(serverDb *dbarray, int dbnum, int async, void(callback)(dict *)) {
long long removed = 0;
int startdb, enddb;
if (dbnum == -1) {
startdb = 0;
enddb = server.dbnum - 1;
} else {
startdb = enddb = dbnum;
}
for (int j = startdb; j <= enddb; j++) {
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
removed += kvstoreSize(dbarray[j].keys);
if (async) {
emptyDbAsync(&dbarray[j]);
} else {
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
kvstoreEmpty(dbarray[j].keys, callback);
kvstoreEmpty(dbarray[j].expires, callback);
}
/* Because all keys of database are removed, reset average ttl. */
dbarray[j].avg_ttl = 0;
dbarray[j].expires_cursor = 0;
}
return removed;
}
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
/* Remove all data (keys and functions) from all the databases in a
* server. If callback is given the function is called from
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
* time to time to signal that work is in progress.
*
2018-07-01 13:24:50 +08:00
* The dbnum can be -1 if all the DBs should be flushed, or the specified
* DB number if we want to flush only a single database number.
*
* Flags are be EMPTYDB_NO_FLAGS if no special flags are specified or
* EMPTYDB_ASYNC if we want the memory to be freed in a different thread
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
* and the function to return ASAP. EMPTYDB_NOFUNCTIONS can also be set
* to specify that we do not want to delete the functions.
*
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
* On success the function returns the number of keys removed from the
* database(s). Otherwise -1 is returned in the specific case the
* DB number is out of range, and errno is set to EINVAL. */
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
long long emptyData(int dbnum, int flags, void(callback)(dict *)) {
2018-07-25 18:13:34 +08:00
int async = (flags & EMPTYDB_ASYNC);
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
int with_functions = !(flags & EMPTYDB_NOFUNCTIONS);
ValkeyModuleFlushInfoV1 fi = {VALKEYMODULE_FLUSHINFO_VERSION, !async, dbnum};
long long removed = 0;
if (dbnum < -1 || dbnum >= server.dbnum) {
errno = EINVAL;
return -1;
}
/* Fire the flushdb modules event. */
moduleFireServerEvent(VALKEYMODULE_EVENT_FLUSHDB, VALKEYMODULE_SUBEVENT_FLUSHDB_START, &fi);
/* Make sure the WATCHed keys are affected by the FLUSH* commands.
* Note that we need to call the function while the keys are still
* there. */
signalFlushedDb(dbnum, async);
/* Empty the database structure. */
removed = emptyDbStructure(server.db, dbnum, async, callback);
if (dbnum == -1) flushReplicaKeysWithExpireList();
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
if (with_functions) {
serverAssert(dbnum == -1);
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
functionsLibCtxClearCurrent(async);
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
}
/* Also fire the end event. Note that this event will fire almost
* immediately after the start event if the flush is asynchronous. */
moduleFireServerEvent(VALKEYMODULE_EVENT_FLUSHDB, VALKEYMODULE_SUBEVENT_FLUSHDB_END, &fi);
return removed;
}
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
/* Initialize temporary db on replica for use during diskless replication. */
serverDb *initTempDb(void) {
int slot_count_bits = 0;
int flags = KVSTORE_ALLOCATE_DICTS_ON_DEMAND;
if (server.cluster_enabled) {
slot_count_bits = CLUSTER_SLOT_MASK_BITS;
flags |= KVSTORE_FREE_EMPTY_DICTS;
}
serverDb *tempDb = zcalloc(sizeof(serverDb) * server.dbnum);
for (int i = 0; i < server.dbnum; i++) {
tempDb[i].id = i;
tempDb[i].keys = kvstoreCreate(&kvstoreKeysDictType, slot_count_bits, flags);
tempDb[i].expires = kvstoreCreate(&kvstoreExpiresDictType, slot_count_bits, flags);
}
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
return tempDb;
}
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
/* Discard tempDb, this can be slow (similar to FLUSHALL), but it's always async. */
void discardTempDb(serverDb *tempDb, void(callback)(dict *)) {
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
int async = 1;
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
/* Release temp DBs. */
emptyDbStructure(tempDb, -1, async, callback);
for (int i = 0; i < server.dbnum; i++) {
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
kvstoreRelease(tempDb[i].keys);
kvstoreRelease(tempDb[i].expires);
}
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
zfree(tempDb);
}
int selectDb(client *c, int id) {
if (id < 0 || id >= server.dbnum) return C_ERR;
c->db = &server.db[id];
return C_OK;
}
long long dbTotalServerKeyCount(void) {
long long total = 0;
int j;
for (j = 0; j < server.dbnum; j++) {
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
total += kvstoreSize(server.db[j].keys);
}
return total;
}
/*-----------------------------------------------------------------------------
* Hooks for key space changes.
*
* Every time a key in the database is modified the function
* signalModifiedKey() is called.
*
* Every time a DB is flushed the function signalFlushDb() is called.
*----------------------------------------------------------------------------*/
/* Note that the 'c' argument may be NULL if the key was modified out of
* a context of a client. */
void signalModifiedKey(client *c, serverDb *db, robj *key) {
touchWatchedKey(db, key);
trackingInvalidateKey(c, key, 1);
}
void signalFlushedDb(int dbid, int async) {
int startdb, enddb;
if (dbid == -1) {
startdb = 0;
enddb = server.dbnum - 1;
} else {
startdb = enddb = dbid;
}
for (int j = startdb; j <= enddb; j++) {
Blocked module clients should be aware when a key is deleted (#11310) The use case is a module that wants to implement a blocking command on a key that necessarily exists and wants to unblock the client in case the key is deleted (much like what we implemented for XREADGROUP in #10306) New module API: * RedisModule_BlockClientOnKeysWithFlags Flags: * REDISMODULE_BLOCK_UNBLOCK_NONE * REDISMODULE_BLOCK_UNBLOCK_DELETED ### Detailed description of code changes blocked.c: 1. Both module and stream functions are called whether the key exists or not, regardless of its type. We do that in order to allow modules/stream to unblock the client in case the key is no longer present or has changed type (the behavior for streams didn't change, just code that moved into serveClientsBlockedOnStreamKey) 2. Make sure afterCommand is called in serveClientsBlockedOnKeyByModule, in order to propagate actions from moduleTryServeClientBlockedOnKey. 3. handleClientsBlockedOnKeys: call propagatePendingCommands directly after lookupKeyReadWithFlags to prevent a possible lazy-expire DEL from being mixed with any command propagated by the preceding functions. 4. blockForKeys: Caller can specifiy that it wants to be awakened if key is deleted. Minor optimizations (use dictAddRaw). 5. signalKeyAsReady became signalKeyAsReadyLogic which can take a boolean in case the key is deleted. It will only signal if there's at least one client that awaits key deletion (to save calls to handleClientsBlockedOnKeys). Minor optimizations (use dictAddRaw) db.c: 1. scanDatabaseForDeletedStreams is now scanDatabaseForDeletedKeys and will signalKeyAsReady for any key that was removed from the database or changed type. It is the responsibility of the code in blocked.c to ignore or act on deleted/type-changed keys. 2. Use the new signalDeletedKeyAsReady where needed blockedonkey.c + tcl: 1. Added test of new capabilities (FSL.BPOPGT now requires the key to exist in order to work)
2022-10-18 18:50:02 +02:00
scanDatabaseForDeletedKeys(&server.db[j], NULL);
touchAllWatchedKeysInDb(&server.db[j], NULL);
}
trackingInvalidateKeysOnFlush(async);
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
/* Changes in this method may take place in swapMainDbWithTempDb as well,
* where we execute similar calls, but with subtle differences as it's
* not simply flushing db. */
}
/*-----------------------------------------------------------------------------
* Type agnostic commands operating on the key space
*----------------------------------------------------------------------------*/
/* Return the set of flags to use for the emptyData() call for FLUSHALL
* and FLUSHDB commands.
*
* sync: flushes the database in an sync manner.
* async: flushes the database in an async manner.
* no option: determine sync or async according to the value of lazyfree-lazy-user-flush.
*
* On success C_OK is returned and the flags are stored in *flags, otherwise
* C_ERR is returned and the function sends an error to the client. */
int getFlushCommandFlags(client *c, int *flags) {
/* Parse the optional ASYNC option. */
if (c->argc == 2 && !strcasecmp(c->argv[1]->ptr, "sync")) {
*flags = EMPTYDB_NO_FLAGS;
} else if (c->argc == 2 && !strcasecmp(c->argv[1]->ptr, "async")) {
*flags = EMPTYDB_ASYNC;
} else if (c->argc == 1) {
*flags = server.lazyfree_lazy_user_flush ? EMPTYDB_ASYNC : EMPTYDB_NO_FLAGS;
} else {
addReplyErrorObject(c, shared.syntaxerr);
return C_ERR;
}
return C_OK;
}
/* Flushes the whole server data set. */
void flushAllDataAndResetRDB(int flags) {
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
server.dirty += emptyData(-1, flags, NULL);
if (server.child_type == CHILD_TYPE_RDB) killRDBChild();
if (server.saveparamslen > 0) {
rdbSaveInfo rsi, *rsiptr;
rsiptr = rdbPopulateSaveInfo(&rsi);
Reclaim page cache of RDB file (#11248) # Background The RDB file is usually generated and used once and seldom used again, but the content would reside in page cache until OS evicts it. A potential problem is that once the free memory exhausts, the OS have to reclaim some memory from page cache or swap anonymous page out, which may result in a jitters to the Redis service. Supposing an exact scenario, a high-capacity machine hosts many redis instances, and we're upgrading the Redis together. The page cache in host machine increases as RDBs are generated. Once the free memory drop into low watermark(which is more likely to happen in older Linux kernel like 3.10, before [watermark_scale_factor](https://lore.kernel.org/lkml/1455813719-2395-1-git-send-email-hannes@cmpxchg.org/) is introduced, the `low watermark` is linear to `min watermark`, and there'is not too much buffer space for `kswapd` to be wake up to reclaim memory), a `direct reclaim` happens, which means the process would stall to wait for memory allocation. # What the PR does The PR introduces a capability to reclaim the cache when the RDB is operated. Generally there're two cases, read and write the RDB. For read it's a little messy to address the incremental reclaim, so the reclaim is done in one go in background after the load is finished to avoid blocking the work thread. For write, incremental reclaim amortizes the work of reclaim so no need to put it into background, and the peak watermark of cache can be reduced in this way. Two cases are addresses specially, replication and restart, for both of which the cache is leveraged to speed up the processing, so the reclaim is postponed to a right time. To do this, a flag is added to`rdbSave` and `rdbLoad` to control whether the cache need to be kept, with the default value false. # Something deserve noting 1. Though `posix_fadvise` is the POSIX standard, but only few platform support it, e.g. Linux, FreeBSD 10.0. 2. In Linux `posix_fadvise` only take effect on writeback-ed pages, so a `sync`(or `fsync`, `fdatasync`) is needed to flush the dirty page before `posix_fadvise` if we reclaim write cache. # About test A unit test is added to verify the effect of `posix_fadvise`. In integration test overall cache increase is checked, as well as the cache backed by RDB as a specific TCL test is executed in isolated Github action job.
2023-02-12 15:23:29 +08:00
rdbSave(REPLICA_REQ_NONE, server.rdb_filename, rsiptr, RDBFLAGS_NONE);
}
#if defined(USE_JEMALLOC)
/* jemalloc 5 doesn't release pages back to the OS when there's no traffic.
* for large databases, flushdb blocks for long anyway, so a bit more won't
* harm and this way the flush and purge will be synchronous. */
if (!(flags & EMPTYDB_ASYNC)) jemalloc_purge();
#endif
}
/* FLUSHDB [ASYNC]
*
* Flushes the currently SELECTed DB. */
void flushdbCommand(client *c) {
int flags;
if (getFlushCommandFlags(c, &flags) == C_ERR) return;
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
/* flushdb should not flush the functions */
server.dirty += emptyData(c->db->id, flags | EMPTYDB_NOFUNCTIONS, NULL);
FLUSHDB and FLUSHALL add call forceCommandPropagation / FLUSHALL reset dirty counter to 0 if we enable save (#10691) ## FLUSHALL We used to restore the dirty counter after `rdbSave` zeroed it if we enable save. Otherwise FLUSHALL will not be replicated nor put into the AOF. And then we do increment it again below. Without that extra dirty++, when db was already empty, FLUSHALL will not be replicated nor put into the AOF. We now gonna replace all that dirty counter magic with a call to forceCommandPropagation (REPL and AOF), instead of all the messing around with the dirty counter. Added tests to cover three part (dirty counter, REPL, AOF). One benefit other than cleaner code is that the `rdb_changes_since_last_save` is correct in this case. ## FLUSHDB FLUSHDB was not replicated nor put into the AOF when db was already empty. Unlike DEL on a non-existing key, FLUSHDB always does something, and that's to call the module hook. So basically FLUSHDB is never a NOP, and thus it should always be propagated. Not doing that, could mean that if a module does something in that hook, and wants to avoid issues of that hook being missing on the replica if the db is empty, it'll need to do complicated things. So now FLUSHDB add call forceCommandPropagation, we will always propagate FLUSHDB. Always propagating FLUSHDB seems like a safe approach that shouldn't have any drawbacks (other than looking odd) This was mentioned in #8972 ## Test section: We actually found it while solving a race condition in the BGSAVE test (other.tcl). It was found in extra_ci Daily Arm64 (test-libc-malloc). ``` [exception]: Executing test client: ERR Background save already in progress. ERR Background save already in progress ``` It look like `r flushdb` trigger (schedule) a bgsave right after `waitForBgsave r` and before `r save`. Changing flushdb to flushall, FLUSHALL will do a foreground save and then set the dirty counter to 0.
2022-05-11 16:21:16 +08:00
/* Without the forceCommandPropagation, when DB was already empty,
* FLUSHDB will not be replicated nor put into the AOF. */
forceCommandPropagation(c, PROPAGATE_REPL | PROPAGATE_AOF);
addReply(c, shared.ok);
FLUSHDB and FLUSHALL add call forceCommandPropagation / FLUSHALL reset dirty counter to 0 if we enable save (#10691) ## FLUSHALL We used to restore the dirty counter after `rdbSave` zeroed it if we enable save. Otherwise FLUSHALL will not be replicated nor put into the AOF. And then we do increment it again below. Without that extra dirty++, when db was already empty, FLUSHALL will not be replicated nor put into the AOF. We now gonna replace all that dirty counter magic with a call to forceCommandPropagation (REPL and AOF), instead of all the messing around with the dirty counter. Added tests to cover three part (dirty counter, REPL, AOF). One benefit other than cleaner code is that the `rdb_changes_since_last_save` is correct in this case. ## FLUSHDB FLUSHDB was not replicated nor put into the AOF when db was already empty. Unlike DEL on a non-existing key, FLUSHDB always does something, and that's to call the module hook. So basically FLUSHDB is never a NOP, and thus it should always be propagated. Not doing that, could mean that if a module does something in that hook, and wants to avoid issues of that hook being missing on the replica if the db is empty, it'll need to do complicated things. So now FLUSHDB add call forceCommandPropagation, we will always propagate FLUSHDB. Always propagating FLUSHDB seems like a safe approach that shouldn't have any drawbacks (other than looking odd) This was mentioned in #8972 ## Test section: We actually found it while solving a race condition in the BGSAVE test (other.tcl). It was found in extra_ci Daily Arm64 (test-libc-malloc). ``` [exception]: Executing test client: ERR Background save already in progress. ERR Background save already in progress ``` It look like `r flushdb` trigger (schedule) a bgsave right after `waitForBgsave r` and before `r save`. Changing flushdb to flushall, FLUSHALL will do a foreground save and then set the dirty counter to 0.
2022-05-11 16:21:16 +08:00
#if defined(USE_JEMALLOC)
/* jemalloc 5 doesn't release pages back to the OS when there's no traffic.
* for large databases, flushdb blocks for long anyway, so a bit more won't
* harm and this way the flush and purge will be synchronous. */
if (!(flags & EMPTYDB_ASYNC)) jemalloc_purge();
#endif
}
/* FLUSHALL [ASYNC]
*
* Flushes the whole server data set. */
void flushallCommand(client *c) {
int flags;
if (getFlushCommandFlags(c, &flags) == C_ERR) return;
Change FUNCTION CREATE, DELETE and FLUSH to be WRITE commands instead of MAY_REPLICATE. (#9953) The issue with MAY_REPLICATE is that all automatic mechanisms to handle write commands will not work. This require have a special treatment for: * Not allow those commands to be executed on RO replica. * Allow those commands to be executed on RO replica from primary connection. * Allow those commands to be executed on the RO replica from AOF. By setting those commands as WRITE commands we are getting all those properties from Redis. Test was added to verify that those properties work as expected. In addition, rearrange when and where functions are flushed. Before this PR functions were flushed manually on `rdbLoadRio` and cleaned manually on failure. This contradicts the assumptions that functions are data and need to be created/deleted alongside with the data. A side effect of this, for example, `debug reload noflush` did not flush the data but did flush the functions, `debug loadaof` flush the data but not the functions. This PR move functions deletion into `emptyDb`. `emptyDb` (renamed to `emptyData`) will now accept an additional flag, `NOFUNCTIONS` which specifically indicate that we do not want to flush the functions (on all other cases, functions will be flushed). Used the new flag on FLUSHALL and FLUSHDB only! Tests were added to `debug reload` and `debug loadaof` to verify that functions behave the same as the data. Notice that because now functions will be deleted along side with the data we can not allow `CLUSTER RESET` to be called from within a function (it will cause the function to be released while running), this PR adds `NO_SCRIPT` flag to `CLUSTER RESET` so it will not be possible to be called from within a function. The other cluster commands are allowed from within a function (there are use-cases that uses `GETKEYSINSLOT` to iterate over all the keys on a given slot). Tests was added to verify `CLUSTER RESET` is denied from within a script. Another small change on this PR is that `RDBFLAGS_ALLOW_DUP` is also applicable on functions. When loading functions, if this flag is set, we will replace old functions with new ones on collisions.
2021-12-21 16:13:29 +02:00
/* flushall should not flush the functions */
flushAllDataAndResetRDB(flags | EMPTYDB_NOFUNCTIONS);
FLUSHDB and FLUSHALL add call forceCommandPropagation / FLUSHALL reset dirty counter to 0 if we enable save (#10691) ## FLUSHALL We used to restore the dirty counter after `rdbSave` zeroed it if we enable save. Otherwise FLUSHALL will not be replicated nor put into the AOF. And then we do increment it again below. Without that extra dirty++, when db was already empty, FLUSHALL will not be replicated nor put into the AOF. We now gonna replace all that dirty counter magic with a call to forceCommandPropagation (REPL and AOF), instead of all the messing around with the dirty counter. Added tests to cover three part (dirty counter, REPL, AOF). One benefit other than cleaner code is that the `rdb_changes_since_last_save` is correct in this case. ## FLUSHDB FLUSHDB was not replicated nor put into the AOF when db was already empty. Unlike DEL on a non-existing key, FLUSHDB always does something, and that's to call the module hook. So basically FLUSHDB is never a NOP, and thus it should always be propagated. Not doing that, could mean that if a module does something in that hook, and wants to avoid issues of that hook being missing on the replica if the db is empty, it'll need to do complicated things. So now FLUSHDB add call forceCommandPropagation, we will always propagate FLUSHDB. Always propagating FLUSHDB seems like a safe approach that shouldn't have any drawbacks (other than looking odd) This was mentioned in #8972 ## Test section: We actually found it while solving a race condition in the BGSAVE test (other.tcl). It was found in extra_ci Daily Arm64 (test-libc-malloc). ``` [exception]: Executing test client: ERR Background save already in progress. ERR Background save already in progress ``` It look like `r flushdb` trigger (schedule) a bgsave right after `waitForBgsave r` and before `r save`. Changing flushdb to flushall, FLUSHALL will do a foreground save and then set the dirty counter to 0.
2022-05-11 16:21:16 +08:00
/* Without the forceCommandPropagation, when DBs were already empty,
* FLUSHALL will not be replicated nor put into the AOF. */
forceCommandPropagation(c, PROPAGATE_REPL | PROPAGATE_AOF);
addReply(c, shared.ok);
}
/* This command implements DEL and UNLINK. */
void delGenericCommand(client *c, int lazy) {
int numdel = 0, j;
for (j = 1; j < c->argc; j++) {
if (expireIfNeeded(c->db, c->argv[j], 0) == KEY_DELETED) continue;
int deleted = lazy ? dbAsyncDelete(c->db, c->argv[j]) : dbSyncDelete(c->db, c->argv[j]);
if (deleted) {
signalModifiedKey(c, c->db, c->argv[j]);
2015-07-27 09:41:48 +02:00
notifyKeyspaceEvent(NOTIFY_GENERIC, "del", c->argv[j], c->db->id);
server.dirty++;
numdel++;
}
}
addReplyLongLong(c, numdel);
}
void delCommand(client *c) {
delGenericCommand(c, server.lazyfree_lazy_user_del);
}
void unlinkCommand(client *c) {
delGenericCommand(c, 1);
}
/* EXISTS key1 key2 ... key_N.
* Return value is the number of keys existing. */
void existsCommand(client *c) {
long long count = 0;
int j;
for (j = 1; j < c->argc; j++) {
if (lookupKeyReadWithFlags(c->db, c->argv[j], LOOKUP_NOTOUCH)) count++;
}
addReplyLongLong(c, count);
}
void selectCommand(client *c) {
int id;
if (getIntFromObjectOrReply(c, c->argv[1], &id, NULL) != C_OK) return;
if (server.cluster_enabled && id != 0) {
2011-03-29 17:51:15 +02:00
addReplyError(c, "SELECT is not allowed in cluster mode");
return;
}
if (selectDb(c, id) == C_ERR) {
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
addReplyError(c, "DB index is out of range");
} else {
addReply(c, shared.ok);
}
}
void randomkeyCommand(client *c) {
robj *key;
if ((key = dbRandomKey(c->db)) == NULL) {
2018-11-30 09:41:54 +01:00
addReplyNull(c);
return;
}
addReplyBulk(c, key);
decrRefCount(key);
}
void keysCommand(client *c) {
dictEntry *de;
sds pattern = c->argv[1]->ptr;
int plen = sdslen(pattern), allkeys, pslot = -1;
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
unsigned long numkeys = 0;
void *replylen = addReplyDeferredLen(c);
allkeys = (pattern[0] == '*' && plen == 1);
if (server.cluster_enabled && !allkeys) {
pslot = patternHashSlot(pattern, plen);
}
kvstoreDictIterator *kvs_di = NULL;
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
kvstoreIterator *kvs_it = NULL;
if (pslot != -1) {
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
if (!kvstoreDictSize(c->db->keys, pslot)) {
/* Requested slot is empty */
setDeferredArrayLen(c, replylen, 0);
return;
}
kvs_di = kvstoreGetDictSafeIterator(c->db->keys, pslot);
} else {
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
kvs_it = kvstoreIteratorInit(c->db->keys);
}
robj keyobj;
while ((de = kvs_di ? kvstoreDictIteratorNext(kvs_di) : kvstoreIteratorNext(kvs_it)) != NULL) {
sds key = dictGetKey(de);
2010-08-30 11:51:45 +02:00
if (allkeys || stringmatchlen(pattern, plen, key, sdslen(key), 0)) {
initStaticStringObject(keyobj, key);
if (!keyIsExpired(c->db, &keyobj)) {
addReplyBulkCBuffer(c, key, sdslen(key));
numkeys++;
}
}
if (c->flag.close_asap) break;
}
if (kvs_di) kvstoreReleaseDictIterator(kvs_di);
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
if (kvs_it) kvstoreIteratorRelease(kvs_it);
setDeferredArrayLen(c, replylen, numkeys);
}
/* Data used by the dict scan callback. */
typedef struct {
list *keys; /* elements that collect from dict */
robj *o; /* o must be a hash/set/zset object, NULL means current db */
long long type; /* the particular type when scan the db */
sds pattern; /* pattern string, NULL means no pattern */
long sampled; /* cumulative number of keys sampled */
int only_keys; /* set to 1 means to return keys only */
} scanData;
/* Helper function to compare key type in scan commands */
int objectTypeCompare(robj *o, long long target) {
if (o->type != OBJ_MODULE) {
if (o->type != target)
return 0;
else
return 1;
}
/* module type compare */
long long mt = (long long)VALKEYMODULE_TYPE_SIGN(((moduleValue *)o->ptr)->type->id);
if (target != -mt)
return 0;
else
return 1;
}
/* This callback is used by scanGenericCommand in order to collect elements
* returned by the dictionary iterator into a list. */
2012-07-09 01:00:26 -07:00
void scanCallback(void *privdata, const dictEntry *de) {
scanData *data = (scanData *)privdata;
list *keys = data->keys;
robj *o = data->o;
sds val = NULL;
sds key = NULL;
data->sampled++;
/* o and typename can not have values at the same time. */
serverAssert(!((data->type != LLONG_MAX) && o));
/* Filter an element if it isn't the type we want. */
if (!o && data->type != LLONG_MAX) {
robj *rval = dictGetVal(de);
if (!objectTypeCompare(rval, data->type)) return;
}
/* Filter element if it does not match the pattern. */
sds keysds = dictGetKey(de);
if (data->pattern) {
if (!stringmatchlen(data->pattern, sdslen(data->pattern), keysds, sdslen(keysds), 0)) {
return;
}
}
if (o == NULL) {
key = keysds;
} else if (o->type == OBJ_SET) {
key = keysds;
} else if (o->type == OBJ_HASH) {
key = keysds;
if (!data->only_keys) {
val = dictGetVal(de);
}
} else if (o->type == OBJ_ZSET) {
key = sdsdup(keysds);
if (!data->only_keys) {
char buf[MAX_LONG_DOUBLE_CHARS];
int len = ld2string(buf, sizeof(buf), *(double *)dictGetVal(de), LD_STR_AUTO);
val = sdsnewlen(buf, len);
}
} else {
2015-07-27 09:41:48 +02:00
serverPanic("Type not handled in SCAN callback.");
}
listAddNodeTail(keys, key);
if (val) listAddNodeTail(keys, val);
2012-07-09 01:00:26 -07:00
}
/* Try to parse a SCAN cursor stored at object 'o':
* if the cursor is valid, store it as unsigned integer into *cursor and
* returns C_OK. Otherwise return C_ERR and send an error to the
* client. */
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 parseScanCursorOrReply(client *c, robj *o, unsigned long long *cursor) {
if (!string2ull(o->ptr, cursor)) {
addReplyError(c, "invalid cursor");
return C_ERR;
}
return C_OK;
}
char *obj_type_name[OBJ_TYPE_MAX] = {"string", "list", "set", "zset", "hash", NULL, /* module type is special */
"stream"};
/* Helper function to get type from a string in scan commands */
long long getObjectTypeByName(char *name) {
for (long long i = 0; i < OBJ_TYPE_MAX; i++) {
if (obj_type_name[i] && !strcasecmp(name, obj_type_name[i])) {
return i;
}
}
moduleType *mt = moduleTypeLookupModuleByNameIgnoreCase(name);
if (mt != NULL) return -(VALKEYMODULE_TYPE_SIGN(mt->id));
return LLONG_MAX;
}
char *getObjectTypeName(robj *o) {
if (o == NULL) {
return "none";
}
serverAssert(o->type >= 0 && o->type < OBJ_TYPE_MAX);
if (o->type == OBJ_MODULE) {
moduleValue *mv = o->ptr;
return mv->type->name;
} else {
return obj_type_name[o->type];
}
}
/* This command implements SCAN, HSCAN and SSCAN commands.
* If object 'o' is passed, then it must be a Hash, Set or Zset object, otherwise
* if 'o' is NULL the command will operate on the dictionary associated with
* the current database.
*
* When 'o' is not NULL the function assumes that the first argument in
* the client arguments vector is a key so it skips it before iterating
* in order to parse options.
*
* In the case of a Hash object the function returns both the field and value
* of every element on the Hash. */
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) {
2012-07-09 01:00:26 -07:00
int i, j;
listNode *node;
long count = 10;
2014-12-12 12:02:39 +09:00
sds pat = NULL;
sds typename = NULL;
long long type = LLONG_MAX;
int patlen = 0, use_pattern = 0, only_keys = 0;
dict *ht;
/* Object must be NULL (to iterate keys names), or the type of the object
* must be Set, Sorted Set, or Hash. */
2015-07-26 15:29:53 +02:00
serverAssert(o == NULL || o->type == OBJ_SET || o->type == OBJ_HASH || o->type == OBJ_ZSET);
/* Set i to the first option argument. The previous one is the cursor. */
i = (o == NULL) ? 2 : 3; /* Skip the key argument if needed. */
2012-07-09 01:00:26 -07:00
/* Step 1: Parse options. */
2012-07-09 01:00:26 -07:00
while (i < c->argc) {
j = c->argc - i;
if (!strcasecmp(c->argv[i]->ptr, "count") && j >= 2) {
2013-10-25 12:01:49 +02:00
if (getLongFromObjectOrReply(c, c->argv[i + 1], &count, NULL) != C_OK) {
return;
2012-07-09 01:00:26 -07:00
}
if (count < 1) {
addReplyErrorObject(c, shared.syntaxerr);
return;
2012-07-09 01:00:26 -07:00
}
i += 2;
} else if (!strcasecmp(c->argv[i]->ptr, "match") && j >= 2) {
2012-07-09 01:00:26 -07:00
pat = c->argv[i + 1]->ptr;
patlen = sdslen(pat);
/* The pattern always matches if it is exactly "*", so it is
* equivalent to disabling it. */
use_pattern = !(patlen == 1 && pat[0] == '*');
2012-07-09 01:00:26 -07:00
i += 2;
} else if (!strcasecmp(c->argv[i]->ptr, "type") && o == NULL && j >= 2) {
/* SCAN for a particular type only applies to the db dict */
typename = c->argv[i + 1]->ptr;
type = getObjectTypeByName(typename);
if (type == LLONG_MAX) {
addReplyErrorFormat(c, "unknown type name '%s'", typename);
return;
}
i += 2;
} else if (!strcasecmp(c->argv[i]->ptr, "novalues")) {
if (!o || o->type != OBJ_HASH) {
addReplyError(c, "NOVALUES option can only be used in HSCAN");
return;
}
only_keys = 1;
i++;
} else if (!strcasecmp(c->argv[i]->ptr, "noscores")) {
if (!o || o->type != OBJ_ZSET) {
addReplyError(c, "NOSCORES option can only be used in ZSCAN");
return;
}
only_keys = 1;
i++;
2012-07-09 01:00:26 -07:00
} else {
addReplyErrorObject(c, shared.syntaxerr);
return;
2012-07-09 01:00:26 -07:00
}
}
/* Step 2: Iterate the collection.
*
2021-11-24 19:34:13 +08:00
* Note that if the object is encoded with a listpack, intset, or any other
* representation that is not a hash table, we are sure that it is also
* composed of a small number of elements. So to avoid taking state we
* just return everything inside the object in a single call, setting the
* cursor to zero to signal the end of the iteration. */
/* Handle the case of a hash table. */
ht = NULL;
if (o == NULL) {
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
ht = NULL;
} else if (o->type == OBJ_SET && o->encoding == OBJ_ENCODING_HT) {
ht = o->ptr;
} else if (o->type == OBJ_HASH && o->encoding == OBJ_ENCODING_HT) {
ht = o->ptr;
} else if (o->type == OBJ_ZSET && o->encoding == OBJ_ENCODING_SKIPLIST) {
zset *zs = o->ptr;
ht = zs->dict;
}
list *keys = listCreate();
/* Set a free callback for the contents of the collected keys list.
* For the main keyspace dict, and when we scan a key that's dict encoded
* (we have 'ht'), we don't need to define free method because the strings
* in the list are just a shallow copy from the pointer in the dictEntry.
* When scanning a key with other encodings (e.g. listpack), we need to
* free the temporary strings we add to that list.
* The exception to the above is ZSET, where we do allocate temporary
* strings even when scanning a dict. */
if (o && (!ht || o->type == OBJ_ZSET)) {
listSetFreeMethod(keys, (void (*)(void *))sdsfree);
}
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
/* For main dictionary scan or data structure using hashtable. */
if (!o || ht) {
/* We set the max number of iterations to ten times the specified
* COUNT, so if the hash table is in a pathological state (very
* sparsely populated) we avoid to block too much time at the cost
* of returning no or very few elements. */
long maxiterations = count * 10;
/* We pass scanData which have three pointers to the callback:
* 1. data.keys: the list to which it will add new elements;
* 2. data.o: the object containing the dictionary so that
* it is possible to fetch more data in a type-dependent way;
* 3. data.type: the specified type scan in the db, LLONG_MAX means
* type matching is no needed;
* 4. data.pattern: the pattern string;
* 5. data.sampled: the maxiteration limit is there in case we're
* working on an empty dict, one with a lot of empty buckets, and
* for the buckets are not empty, we need to limit the spampled number
* to prevent a long hang time caused by filtering too many keys;
* 6. data.only_keys: to control whether values will be returned or
* only keys are returned. */
scanData data = {
.keys = keys,
.o = o,
.type = type,
.pattern = use_pattern ? pat : NULL,
.sampled = 0,
.only_keys = only_keys,
};
/* A pattern may restrict all matching keys to one cluster slot. */
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 onlydidx = -1;
if (o == NULL && use_pattern && server.cluster_enabled) {
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
onlydidx = patternHashSlot(pat, patlen);
}
do {
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
/* In cluster mode there is a separate dictionary for each slot.
* If cursor is empty, we should try exploring next non-empty slot. */
if (o == NULL) {
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
cursor = kvstoreScan(c->db->keys, cursor, onlydidx, scanCallback, NULL, &data);
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
} else {
cursor = dictScan(ht, cursor, scanCallback, &data);
}
} while (cursor && maxiterations-- && data.sampled < count);
} else if (o->type == OBJ_SET) {
char *str;
char buf[LONG_STR_SIZE];
size_t len;
int64_t llele;
setTypeIterator *si = setTypeInitIterator(o);
while (setTypeNext(si, &str, &len, &llele) != -1) {
if (str == NULL) {
len = ll2string(buf, sizeof(buf), llele);
}
char *key = str ? str : buf;
if (use_pattern && !stringmatchlen(pat, sdslen(pat), key, len, 0)) {
continue;
}
listAddNodeTail(keys, sdsnewlen(key, len));
}
setTypeReleaseIterator(si);
cursor = 0;
} else if ((o->type == OBJ_HASH || o->type == OBJ_ZSET) && o->encoding == OBJ_ENCODING_LISTPACK) {
unsigned char *p = lpFirst(o->ptr);
unsigned char *str;
int64_t len;
unsigned char intbuf[LP_INTBUF_SIZE];
while (p) {
str = lpGet(p, &len, intbuf);
/* point to the value */
p = lpNext(o->ptr, p);
if (use_pattern && !stringmatchlen(pat, sdslen(pat), (char *)str, len, 0)) {
/* jump to the next key/val pair */
p = lpNext(o->ptr, p);
continue;
}
/* add key object */
listAddNodeTail(keys, sdsnewlen(str, len));
/* add value object */
if (!only_keys) {
str = lpGet(p, &len, intbuf);
listAddNodeTail(keys, sdsnewlen(str, len));
}
p = lpNext(o->ptr, p);
}
cursor = 0;
} else {
2015-07-27 09:41:48 +02:00
serverPanic("Not handled encoding in SCAN.");
}
2012-07-09 01:00:26 -07:00
/* Step 3: Filter the expired keys */
if (o == NULL && listLength(keys)) {
robj kobj;
listIter li;
listNode *ln;
listRewind(keys, &li);
while ((ln = listNext(&li))) {
sds key = listNodeValue(ln);
initStaticStringObject(kobj, key);
if (expireIfNeeded(c->db, &kobj, 0) != KEY_VALID) {
listDelNode(keys, ln);
}
2012-07-09 01:00:26 -07:00
}
}
/* Step 4: Reply to the client. */
addReplyArrayLen(c, 2);
addReplyBulkLongLong(c, cursor);
2012-07-09 01:00:26 -07:00
addReplyArrayLen(c, listLength(keys));
while ((node = listFirst(keys)) != NULL) {
sds key = listNodeValue(node);
addReplyBulkCBuffer(c, key, sdslen(key));
listDelNode(keys, node);
2012-07-09 01:00:26 -07:00
}
listRelease(keys);
}
/* The SCAN command completely relies on scanGenericCommand. */
void scanCommand(client *c) {
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
unsigned long long cursor;
if (parseScanCursorOrReply(c, c->argv[1], &cursor) == C_ERR) return;
scanGenericCommand(c, NULL, cursor);
}
void dbsizeCommand(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
addReplyLongLong(c, kvstoreSize(c->db->keys));
}
void lastsaveCommand(client *c) {
addReplyLongLong(c, server.lastsave);
}
void typeCommand(client *c) {
robj *o;
o = lookupKeyReadWithFlags(c->db, c->argv[1], LOOKUP_NOTOUCH);
addReplyStatus(c, getObjectTypeName(o));
}
/* SHUTDOWN [[NOSAVE | SAVE] [NOW] [FORCE] | ABORT] */
void shutdownCommand(client *c) {
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
int flags = SHUTDOWN_NOFLAGS;
int abort = 0;
for (int i = 1; i < c->argc; i++) {
if (!strcasecmp(c->argv[i]->ptr, "nosave")) {
2015-07-27 09:41:48 +02:00
flags |= SHUTDOWN_NOSAVE;
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
} else if (!strcasecmp(c->argv[i]->ptr, "save")) {
2015-07-27 09:41:48 +02:00
flags |= SHUTDOWN_SAVE;
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
} else if (!strcasecmp(c->argv[i]->ptr, "now")) {
flags |= SHUTDOWN_NOW;
} else if (!strcasecmp(c->argv[i]->ptr, "force")) {
flags |= SHUTDOWN_FORCE;
} else if (!strcasecmp(c->argv[i]->ptr, "abort")) {
abort = 1;
} else {
addReplyErrorObject(c, shared.syntaxerr);
return;
}
}
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
if ((abort && flags != SHUTDOWN_NOFLAGS) || (flags & SHUTDOWN_NOSAVE && flags & SHUTDOWN_SAVE)) {
/* Illegal combo. */
addReplyErrorObject(c, shared.syntaxerr);
return;
}
if (abort) {
if (abortShutdown() == C_OK)
addReply(c, shared.ok);
else
addReplyError(c, "No shutdown in progress.");
return;
}
if (!(flags & SHUTDOWN_NOW) && c->flag.deny_blocking) {
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
addReplyError(c, "SHUTDOWN without NOW or ABORT isn't allowed for DENY BLOCKING client");
return;
}
Fix broken protocol in MISCONF error, RM_Yield bugs, RM_Call(EVAL) OOM check bug, and new RM_Call checks. (#10786) * Fix broken protocol when redis can't persist to RDB (general commands, not modules), excessive newline. regression of #10372 (7.0 RC3) * Fix broken protocol when Redis can't persist to AOF (modules and scripts), missing newline. * Fix bug in OOM check of EVAL scripts called from RM_Call. set the cached OOM state for scripts before executing module commands too, so that it can serve scripts that are executed by modules. i.e. in the past EVAL executed by RM_Call could have either falsely fail or falsely succeeded because of a wrong cached OOM state flag. * Fix bugs with RM_Yield: 1. SHUTDOWN should only accept the NOSAVE mode 2. Avoid eviction during yield command processing. 3. Avoid processing master client commands while yielding from another client * Add new two more checks to RM_Call script mode. 1. READONLY You can't write against a read only replica 2. MASTERDOWN Link with MASTER is down and `replica-serve-stale-data` is set to `no` * Add new RM_Call flag to let redis automatically refuse `deny-oom` commands while over the memory limit. * Add tests to cover various errors from Scripts, Modules, Modules calling scripts, and Modules calling commands in script mode. Add tests: * Looks like the MISCONF error was completely uncovered by the tests, add tests for it, including from scripts, and modules * Add tests for NOREPLICAS from scripts * Add tests for the various errors in module RM_Call, including RM_Call that calls EVAL, and RM_call in "eval mode". that includes: NOREPLICAS, READONLY, MASTERDOWN, MISCONF
2022-06-01 13:04:22 +03:00
if (!(flags & SHUTDOWN_NOSAVE) && isInsideYieldingLongCommand()) {
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
/* Script timed out. Shutdown allowed only with the NOSAVE flag. See
* also processCommand where these errors are returned. */
if (server.busy_module_yield_flags && server.busy_module_yield_reply) {
addReplyErrorFormat(c, "-BUSY %s", server.busy_module_yield_reply);
} else if (server.busy_module_yield_flags) {
addReplyErrorObject(c, shared.slowmoduleerr);
} else if (scriptIsEval()) {
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
addReplyErrorObject(c, shared.slowevalerr);
} else {
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
addReplyErrorObject(c, shared.slowscripterr);
}
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
return;
}
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
blockClientShutdown(c);
if (prepareForShutdown(c, flags) == C_OK) exit(0);
Wait for replicas when shutting down (#9872) To avoid data loss, this commit adds a grace period for lagging replicas to catch up the replication offset. Done: * Wait for replicas when shutdown is triggered by SIGTERM and SIGINT. * Wait for replicas when shutdown is triggered by the SHUTDOWN command. A new blocked client type BLOCKED_SHUTDOWN is introduced, allowing multiple clients to call SHUTDOWN in parallel. Note that they don't expect a response unless an error happens and shutdown is aborted. * Log warning for each replica lagging behind when finishing shutdown. * CLIENT_PAUSE_WRITE while waiting for replicas. * Configurable grace period 'shutdown-timeout' in seconds (default 10). * New flags for the SHUTDOWN command: - NOW disables the grace period for lagging replicas. - FORCE ignores errors writing the RDB or AOF files which would normally prevent a shutdown. - ABORT cancels ongoing shutdown. Can't be combined with other flags. * New field in the output of the INFO command: 'shutdown_in_milliseconds'. The value is the remaining maximum time to wait for lagging replicas before finishing the shutdown. This field is present in the Server section **only** during shutdown. Not directly related: * When shutting down, if there is an AOF saving child, it is killed **even** if AOF is disabled. This can happen if BGREWRITEAOF is used when AOF is off. * Client pause now has end time and type (WRITE or ALL) per purpose. The different pause purposes are *CLIENT PAUSE command*, *failover* and *shutdown*. If clients are unpaused for one purpose, it doesn't affect client pause for other purposes. For example, the CLIENT UNPAUSE command doesn't affect client pause initiated by the failover or shutdown procedures. A completed failover or a failed shutdown doesn't unpause clients paused by the CLIENT PAUSE command. Notes: * DEBUG RESTART doesn't wait for replicas. * We already have a warning logged when a replica disconnects. This means that if any replica connection is lost during the shutdown, it is either logged as disconnected or as lagging at the time of exit. Co-authored-by: Oran Agra <oran@redislabs.com>
2022-01-02 08:50:15 +01:00
/* If we're here, then shutdown is ongoing (the client is still blocked) or
* failed (the client has received an error). */
}
void renameGenericCommand(client *c, int nx) {
robj *o;
long long expire;
int samekey = 0;
/* When source and dest key is the same, no operation is performed,
* if the key exists, however we still return an error on unexisting key. */
if (sdscmp(c->argv[1]->ptr, c->argv[2]->ptr) == 0) samekey = 1;
if ((o = lookupKeyWriteOrReply(c, c->argv[1], shared.nokeyerr)) == NULL) return;
if (samekey) {
addReply(c, nx ? shared.czero : shared.ok);
return;
}
incrRefCount(o);
expire = getExpire(c->db, c->argv[1]);
if (lookupKeyWrite(c->db, c->argv[2]) != NULL) {
if (nx) {
decrRefCount(o);
addReply(c, shared.czero);
return;
}
2013-01-23 16:44:45 +01:00
/* Overwrite: delete the old key before creating the new one
* with the same name. */
dbDelete(c->db, c->argv[2]);
}
dbAdd(c->db, c->argv[2], o);
Replication: fix the infamous key leakage of writable slaves + EXPIRE. BACKGROUND AND USE CASEj Redis slaves are normally write only, however the supprot a "writable" mode which is very handy when scaling reads on slaves, that actually need write operations in order to access data. For instance imagine having slaves replicating certain Sets keys from the master. When accessing the data on the slave, we want to peform intersections between such Sets values. However we don't want to intersect each time: to cache the intersection for some time often is a good idea. To do so, it is possible to setup a slave as a writable slave, and perform the intersection on the slave side, perhaps setting a TTL on the resulting key so that it will expire after some time. THE BUG Problem: in order to have a consistent replication, expiring of keys in Redis replication is up to the master, that synthesize DEL operations to send in the replication stream. However slaves logically expire keys by hiding them from read attempts from clients so that if the master did not promptly sent a DEL, the client still see logically expired keys as non existing. Because slaves don't actively expire keys by actually evicting them but just masking from the POV of read operations, if a key is created in a writable slave, and an expire is set, the key will be leaked forever: 1. No DEL will be received from the master, which does not know about such a key at all. 2. No eviction will be performed by the slave, since it needs to disable eviction because it's up to masters, otherwise consistency of data is lost. THE FIX In order to fix the problem, the slave should be able to tag keys that were created in the slave side and have an expire set in some way. My solution involved using an unique additional dictionary created by the writable slave only if needed. The dictionary is obviously keyed by the key name that we need to track: all the keys that are set with an expire directly by a client writing to the slave are tracked. The value in the dictionary is a bitmap of all the DBs where such a key name need to be tracked, so that we can use a single dictionary to track keys in all the DBs used by the slave (actually this limits the solution to the first 64 DBs, but the default with Redis is to use 16 DBs). This solution allows to pay both a small complexity and CPU penalty, which is zero when the feature is not used, actually. The slave-side eviction is encapsulated in code which is not coupled with the rest of the Redis core, if not for the hook to track the keys. TODO I'm doing the first smoke tests to see if the feature works as expected: so far so good. Unit tests should be added before merging into the 4.0 branch.
2016-12-13 10:20:06 +01:00
if (expire != -1) setExpire(c, c->db, c->argv[2], expire);
dbDelete(c->db, c->argv[1]);
signalModifiedKey(c, c->db, c->argv[1]);
signalModifiedKey(c, c->db, c->argv[2]);
2015-07-27 09:41:48 +02:00
notifyKeyspaceEvent(NOTIFY_GENERIC, "rename_from", c->argv[1], c->db->id);
notifyKeyspaceEvent(NOTIFY_GENERIC, "rename_to", c->argv[2], c->db->id);
server.dirty++;
addReply(c, nx ? shared.cone : shared.ok);
}
void renameCommand(client *c) {
renameGenericCommand(c, 0);
}
void renamenxCommand(client *c) {
renameGenericCommand(c, 1);
}
void moveCommand(client *c) {
robj *o;
serverDb *src, *dst;
int srcid, dbid;
long long expire;
2011-03-29 17:51:15 +02:00
if (server.cluster_enabled) {
addReplyError(c, "MOVE is not allowed in cluster mode");
return;
}
/* Obtain source and target DB pointers */
src = c->db;
srcid = c->db->id;
if (getIntFromObjectOrReply(c, c->argv[2], &dbid, NULL) != C_OK) return;
if (selectDb(c, dbid) == C_ERR) {
addReplyError(c, "DB index is out of range");
return;
}
dst = c->db;
selectDb(c, srcid); /* Back to the source DB */
/* If the user is moving using as target the same
* DB as the source DB it is probably an error. */
if (src == dst) {
addReplyErrorObject(c, shared.sameobjecterr);
return;
}
/* Check if the element exists and get a reference */
o = lookupKeyWrite(c->db, c->argv[1]);
if (!o) {
addReply(c, shared.czero);
return;
}
expire = getExpire(c->db, c->argv[1]);
/* Return zero if the key already exists in the target DB */
if (lookupKeyWrite(dst, c->argv[1]) != NULL) {
addReply(c, shared.czero);
return;
}
dbAdd(dst, c->argv[1], o);
Replication: fix the infamous key leakage of writable slaves + EXPIRE. BACKGROUND AND USE CASEj Redis slaves are normally write only, however the supprot a "writable" mode which is very handy when scaling reads on slaves, that actually need write operations in order to access data. For instance imagine having slaves replicating certain Sets keys from the master. When accessing the data on the slave, we want to peform intersections between such Sets values. However we don't want to intersect each time: to cache the intersection for some time often is a good idea. To do so, it is possible to setup a slave as a writable slave, and perform the intersection on the slave side, perhaps setting a TTL on the resulting key so that it will expire after some time. THE BUG Problem: in order to have a consistent replication, expiring of keys in Redis replication is up to the master, that synthesize DEL operations to send in the replication stream. However slaves logically expire keys by hiding them from read attempts from clients so that if the master did not promptly sent a DEL, the client still see logically expired keys as non existing. Because slaves don't actively expire keys by actually evicting them but just masking from the POV of read operations, if a key is created in a writable slave, and an expire is set, the key will be leaked forever: 1. No DEL will be received from the master, which does not know about such a key at all. 2. No eviction will be performed by the slave, since it needs to disable eviction because it's up to masters, otherwise consistency of data is lost. THE FIX In order to fix the problem, the slave should be able to tag keys that were created in the slave side and have an expire set in some way. My solution involved using an unique additional dictionary created by the writable slave only if needed. The dictionary is obviously keyed by the key name that we need to track: all the keys that are set with an expire directly by a client writing to the slave are tracked. The value in the dictionary is a bitmap of all the DBs where such a key name need to be tracked, so that we can use a single dictionary to track keys in all the DBs used by the slave (actually this limits the solution to the first 64 DBs, but the default with Redis is to use 16 DBs). This solution allows to pay both a small complexity and CPU penalty, which is zero when the feature is not used, actually. The slave-side eviction is encapsulated in code which is not coupled with the rest of the Redis core, if not for the hook to track the keys. TODO I'm doing the first smoke tests to see if the feature works as expected: so far so good. Unit tests should be added before merging into the 4.0 branch.
2016-12-13 10:20:06 +01:00
if (expire != -1) setExpire(c, dst, c->argv[1], expire);
incrRefCount(o);
/* OK! key moved, free the entry in the source DB */
dbDelete(src, c->argv[1]);
signalModifiedKey(c, src, c->argv[1]);
signalModifiedKey(c, dst, c->argv[1]);
2019-11-19 00:02:45 -05:00
notifyKeyspaceEvent(NOTIFY_GENERIC, "move_from", c->argv[1], src->id);
notifyKeyspaceEvent(NOTIFY_GENERIC, "move_to", c->argv[1], dst->id);
server.dirty++;
addReply(c, shared.cone);
}
void copyCommand(client *c) {
robj *o;
serverDb *src, *dst;
int srcid, dbid;
long long expire;
int j, replace = 0, delete = 0;
/* Obtain source and target DB pointers
* Default target DB is the same as the source DB
* Parse the REPLACE option and targetDB option. */
src = c->db;
dst = c->db;
srcid = c->db->id;
dbid = c->db->id;
for (j = 3; j < c->argc; j++) {
int additional = c->argc - j - 1;
if (!strcasecmp(c->argv[j]->ptr, "replace")) {
replace = 1;
} else if (!strcasecmp(c->argv[j]->ptr, "db") && additional >= 1) {
if (getIntFromObjectOrReply(c, c->argv[j + 1], &dbid, NULL) != C_OK) return;
if (selectDb(c, dbid) == C_ERR) {
addReplyError(c, "DB index is out of range");
return;
}
dst = c->db;
selectDb(c, srcid); /* Back to the source DB */
j++; /* Consume additional arg. */
} else {
addReplyErrorObject(c, shared.syntaxerr);
return;
}
}
if ((server.cluster_enabled == 1) && (srcid != 0 || dbid != 0)) {
addReplyError(c, "Copying to another database is not allowed in cluster mode");
return;
}
/* If the user select the same DB as
* the source DB and using newkey as the same key
* it is probably an error. */
robj *key = c->argv[1];
robj *newkey = c->argv[2];
if (src == dst && (sdscmp(key->ptr, newkey->ptr) == 0)) {
addReplyErrorObject(c, shared.sameobjecterr);
return;
}
/* Check if the element exists and get a reference */
o = lookupKeyRead(c->db, key);
if (!o) {
addReply(c, shared.czero);
return;
}
expire = getExpire(c->db, key);
/* Return zero if the key already exists in the target DB.
* If REPLACE option is selected, delete newkey from targetDB. */
if (lookupKeyWrite(dst, newkey) != NULL) {
if (replace) {
delete = 1;
} else {
addReply(c, shared.czero);
return;
}
}
/* Duplicate object according to object's type. */
robj *newobj;
switch (o->type) {
case OBJ_STRING: newobj = dupStringObject(o); break;
case OBJ_LIST: newobj = listTypeDup(o); break;
case OBJ_SET: newobj = setTypeDup(o); break;
case OBJ_ZSET: newobj = zsetDup(o); break;
case OBJ_HASH: newobj = hashTypeDup(o); break;
case OBJ_STREAM: newobj = streamDup(o); break;
case OBJ_MODULE:
newobj = moduleTypeDupOrReply(c, key, newkey, dst->id, o);
if (!newobj) return;
break;
default: addReplyError(c, "unknown type object"); return;
}
if (delete) {
dbDelete(dst, newkey);
}
dbAdd(dst, newkey, newobj);
if (expire != -1) setExpire(c, dst, newkey, expire);
/* OK! key copied */
signalModifiedKey(c, dst, c->argv[2]);
notifyKeyspaceEvent(NOTIFY_GENERIC, "copy_to", c->argv[2], dst->id);
server.dirty++;
addReply(c, shared.cone);
}
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
/* Helper function for dbSwapDatabases(): scans the list of keys that have
* one or more blocked clients for B[LR]POP or other blocking commands
* and signal the keys as ready if they are of the right type. See the comment
* where the function is used for more info. */
void scanDatabaseForReadyKeys(serverDb *db) {
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
dictEntry *de;
dictIterator *di = dictGetSafeIterator(db->blocking_keys);
while ((de = dictNext(di)) != NULL) {
robj *key = dictGetKey(de);
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
dictEntry *kde = dbFind(db, key->ptr);
if (kde) {
robj *value = dictGetVal(kde);
signalKeyAsReady(db, key, value->type);
}
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
}
dictReleaseIterator(di);
}
/* Since we are unblocking XREADGROUP clients in the event the
* key was deleted/overwritten we must do the same in case the
* database was flushed/swapped. */
void scanDatabaseForDeletedKeys(serverDb *emptied, serverDb *replaced_with) {
dictEntry *de;
dictIterator *di = dictGetSafeIterator(emptied->blocking_keys);
while ((de = dictNext(di)) != NULL) {
robj *key = dictGetKey(de);
Blocked module clients should be aware when a key is deleted (#11310) The use case is a module that wants to implement a blocking command on a key that necessarily exists and wants to unblock the client in case the key is deleted (much like what we implemented for XREADGROUP in #10306) New module API: * RedisModule_BlockClientOnKeysWithFlags Flags: * REDISMODULE_BLOCK_UNBLOCK_NONE * REDISMODULE_BLOCK_UNBLOCK_DELETED ### Detailed description of code changes blocked.c: 1. Both module and stream functions are called whether the key exists or not, regardless of its type. We do that in order to allow modules/stream to unblock the client in case the key is no longer present or has changed type (the behavior for streams didn't change, just code that moved into serveClientsBlockedOnStreamKey) 2. Make sure afterCommand is called in serveClientsBlockedOnKeyByModule, in order to propagate actions from moduleTryServeClientBlockedOnKey. 3. handleClientsBlockedOnKeys: call propagatePendingCommands directly after lookupKeyReadWithFlags to prevent a possible lazy-expire DEL from being mixed with any command propagated by the preceding functions. 4. blockForKeys: Caller can specifiy that it wants to be awakened if key is deleted. Minor optimizations (use dictAddRaw). 5. signalKeyAsReady became signalKeyAsReadyLogic which can take a boolean in case the key is deleted. It will only signal if there's at least one client that awaits key deletion (to save calls to handleClientsBlockedOnKeys). Minor optimizations (use dictAddRaw) db.c: 1. scanDatabaseForDeletedStreams is now scanDatabaseForDeletedKeys and will signalKeyAsReady for any key that was removed from the database or changed type. It is the responsibility of the code in blocked.c to ignore or act on deleted/type-changed keys. 2. Use the new signalDeletedKeyAsReady where needed blockedonkey.c + tcl: 1. Added test of new capabilities (FSL.BPOPGT now requires the key to exist in order to work)
2022-10-18 18:50:02 +02:00
int existed = 0, exists = 0;
int original_type = -1, curr_type = -1;
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
dictEntry *kde = dbFind(emptied, key->ptr);
if (kde) {
robj *value = dictGetVal(kde);
Blocked module clients should be aware when a key is deleted (#11310) The use case is a module that wants to implement a blocking command on a key that necessarily exists and wants to unblock the client in case the key is deleted (much like what we implemented for XREADGROUP in #10306) New module API: * RedisModule_BlockClientOnKeysWithFlags Flags: * REDISMODULE_BLOCK_UNBLOCK_NONE * REDISMODULE_BLOCK_UNBLOCK_DELETED ### Detailed description of code changes blocked.c: 1. Both module and stream functions are called whether the key exists or not, regardless of its type. We do that in order to allow modules/stream to unblock the client in case the key is no longer present or has changed type (the behavior for streams didn't change, just code that moved into serveClientsBlockedOnStreamKey) 2. Make sure afterCommand is called in serveClientsBlockedOnKeyByModule, in order to propagate actions from moduleTryServeClientBlockedOnKey. 3. handleClientsBlockedOnKeys: call propagatePendingCommands directly after lookupKeyReadWithFlags to prevent a possible lazy-expire DEL from being mixed with any command propagated by the preceding functions. 4. blockForKeys: Caller can specifiy that it wants to be awakened if key is deleted. Minor optimizations (use dictAddRaw). 5. signalKeyAsReady became signalKeyAsReadyLogic which can take a boolean in case the key is deleted. It will only signal if there's at least one client that awaits key deletion (to save calls to handleClientsBlockedOnKeys). Minor optimizations (use dictAddRaw) db.c: 1. scanDatabaseForDeletedStreams is now scanDatabaseForDeletedKeys and will signalKeyAsReady for any key that was removed from the database or changed type. It is the responsibility of the code in blocked.c to ignore or act on deleted/type-changed keys. 2. Use the new signalDeletedKeyAsReady where needed blockedonkey.c + tcl: 1. Added test of new capabilities (FSL.BPOPGT now requires the key to exist in order to work)
2022-10-18 18:50:02 +02:00
original_type = value->type;
existed = 1;
}
Blocked module clients should be aware when a key is deleted (#11310) The use case is a module that wants to implement a blocking command on a key that necessarily exists and wants to unblock the client in case the key is deleted (much like what we implemented for XREADGROUP in #10306) New module API: * RedisModule_BlockClientOnKeysWithFlags Flags: * REDISMODULE_BLOCK_UNBLOCK_NONE * REDISMODULE_BLOCK_UNBLOCK_DELETED ### Detailed description of code changes blocked.c: 1. Both module and stream functions are called whether the key exists or not, regardless of its type. We do that in order to allow modules/stream to unblock the client in case the key is no longer present or has changed type (the behavior for streams didn't change, just code that moved into serveClientsBlockedOnStreamKey) 2. Make sure afterCommand is called in serveClientsBlockedOnKeyByModule, in order to propagate actions from moduleTryServeClientBlockedOnKey. 3. handleClientsBlockedOnKeys: call propagatePendingCommands directly after lookupKeyReadWithFlags to prevent a possible lazy-expire DEL from being mixed with any command propagated by the preceding functions. 4. blockForKeys: Caller can specifiy that it wants to be awakened if key is deleted. Minor optimizations (use dictAddRaw). 5. signalKeyAsReady became signalKeyAsReadyLogic which can take a boolean in case the key is deleted. It will only signal if there's at least one client that awaits key deletion (to save calls to handleClientsBlockedOnKeys). Minor optimizations (use dictAddRaw) db.c: 1. scanDatabaseForDeletedStreams is now scanDatabaseForDeletedKeys and will signalKeyAsReady for any key that was removed from the database or changed type. It is the responsibility of the code in blocked.c to ignore or act on deleted/type-changed keys. 2. Use the new signalDeletedKeyAsReady where needed blockedonkey.c + tcl: 1. Added test of new capabilities (FSL.BPOPGT now requires the key to exist in order to work)
2022-10-18 18:50:02 +02:00
if (replaced_with) {
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
kde = dbFind(replaced_with, key->ptr);
if (kde) {
robj *value = dictGetVal(kde);
Blocked module clients should be aware when a key is deleted (#11310) The use case is a module that wants to implement a blocking command on a key that necessarily exists and wants to unblock the client in case the key is deleted (much like what we implemented for XREADGROUP in #10306) New module API: * RedisModule_BlockClientOnKeysWithFlags Flags: * REDISMODULE_BLOCK_UNBLOCK_NONE * REDISMODULE_BLOCK_UNBLOCK_DELETED ### Detailed description of code changes blocked.c: 1. Both module and stream functions are called whether the key exists or not, regardless of its type. We do that in order to allow modules/stream to unblock the client in case the key is no longer present or has changed type (the behavior for streams didn't change, just code that moved into serveClientsBlockedOnStreamKey) 2. Make sure afterCommand is called in serveClientsBlockedOnKeyByModule, in order to propagate actions from moduleTryServeClientBlockedOnKey. 3. handleClientsBlockedOnKeys: call propagatePendingCommands directly after lookupKeyReadWithFlags to prevent a possible lazy-expire DEL from being mixed with any command propagated by the preceding functions. 4. blockForKeys: Caller can specifiy that it wants to be awakened if key is deleted. Minor optimizations (use dictAddRaw). 5. signalKeyAsReady became signalKeyAsReadyLogic which can take a boolean in case the key is deleted. It will only signal if there's at least one client that awaits key deletion (to save calls to handleClientsBlockedOnKeys). Minor optimizations (use dictAddRaw) db.c: 1. scanDatabaseForDeletedStreams is now scanDatabaseForDeletedKeys and will signalKeyAsReady for any key that was removed from the database or changed type. It is the responsibility of the code in blocked.c to ignore or act on deleted/type-changed keys. 2. Use the new signalDeletedKeyAsReady where needed blockedonkey.c + tcl: 1. Added test of new capabilities (FSL.BPOPGT now requires the key to exist in order to work)
2022-10-18 18:50:02 +02:00
curr_type = value->type;
exists = 1;
}
}
/* We want to try to unblock any client using a blocking XREADGROUP */
Blocked module clients should be aware when a key is deleted (#11310) The use case is a module that wants to implement a blocking command on a key that necessarily exists and wants to unblock the client in case the key is deleted (much like what we implemented for XREADGROUP in #10306) New module API: * RedisModule_BlockClientOnKeysWithFlags Flags: * REDISMODULE_BLOCK_UNBLOCK_NONE * REDISMODULE_BLOCK_UNBLOCK_DELETED ### Detailed description of code changes blocked.c: 1. Both module and stream functions are called whether the key exists or not, regardless of its type. We do that in order to allow modules/stream to unblock the client in case the key is no longer present or has changed type (the behavior for streams didn't change, just code that moved into serveClientsBlockedOnStreamKey) 2. Make sure afterCommand is called in serveClientsBlockedOnKeyByModule, in order to propagate actions from moduleTryServeClientBlockedOnKey. 3. handleClientsBlockedOnKeys: call propagatePendingCommands directly after lookupKeyReadWithFlags to prevent a possible lazy-expire DEL from being mixed with any command propagated by the preceding functions. 4. blockForKeys: Caller can specifiy that it wants to be awakened if key is deleted. Minor optimizations (use dictAddRaw). 5. signalKeyAsReady became signalKeyAsReadyLogic which can take a boolean in case the key is deleted. It will only signal if there's at least one client that awaits key deletion (to save calls to handleClientsBlockedOnKeys). Minor optimizations (use dictAddRaw) db.c: 1. scanDatabaseForDeletedStreams is now scanDatabaseForDeletedKeys and will signalKeyAsReady for any key that was removed from the database or changed type. It is the responsibility of the code in blocked.c to ignore or act on deleted/type-changed keys. 2. Use the new signalDeletedKeyAsReady where needed blockedonkey.c + tcl: 1. Added test of new capabilities (FSL.BPOPGT now requires the key to exist in order to work)
2022-10-18 18:50:02 +02:00
if ((existed && !exists) || original_type != curr_type) signalDeletedKeyAsReady(emptied, key, original_type);
}
dictReleaseIterator(di);
}
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
/* Swap two databases at runtime so that all clients will magically see
* the new database even if already connected. Note that the client
* structure c->db points to a given DB, so we need to be smarter and
* swap the underlying referenced structures, otherwise we would need
* to fix all the references to the DB structure.
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
*
* Returns C_ERR if at least one of the DB ids are out of range, otherwise
* C_OK is returned. */
int dbSwapDatabases(int id1, int id2) {
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
if (id1 < 0 || id1 >= server.dbnum || id2 < 0 || id2 >= server.dbnum) return C_ERR;
if (id1 == id2) return C_OK;
serverDb aux = server.db[id1];
serverDb *db1 = &server.db[id1], *db2 = &server.db[id2];
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
/* Swapdb should make transaction fail if there is any
* client watching keys */
touchAllWatchedKeysInDb(db1, db2);
touchAllWatchedKeysInDb(db2, db1);
/* Try to unblock any XREADGROUP clients if the key no longer exists. */
Blocked module clients should be aware when a key is deleted (#11310) The use case is a module that wants to implement a blocking command on a key that necessarily exists and wants to unblock the client in case the key is deleted (much like what we implemented for XREADGROUP in #10306) New module API: * RedisModule_BlockClientOnKeysWithFlags Flags: * REDISMODULE_BLOCK_UNBLOCK_NONE * REDISMODULE_BLOCK_UNBLOCK_DELETED ### Detailed description of code changes blocked.c: 1. Both module and stream functions are called whether the key exists or not, regardless of its type. We do that in order to allow modules/stream to unblock the client in case the key is no longer present or has changed type (the behavior for streams didn't change, just code that moved into serveClientsBlockedOnStreamKey) 2. Make sure afterCommand is called in serveClientsBlockedOnKeyByModule, in order to propagate actions from moduleTryServeClientBlockedOnKey. 3. handleClientsBlockedOnKeys: call propagatePendingCommands directly after lookupKeyReadWithFlags to prevent a possible lazy-expire DEL from being mixed with any command propagated by the preceding functions. 4. blockForKeys: Caller can specifiy that it wants to be awakened if key is deleted. Minor optimizations (use dictAddRaw). 5. signalKeyAsReady became signalKeyAsReadyLogic which can take a boolean in case the key is deleted. It will only signal if there's at least one client that awaits key deletion (to save calls to handleClientsBlockedOnKeys). Minor optimizations (use dictAddRaw) db.c: 1. scanDatabaseForDeletedStreams is now scanDatabaseForDeletedKeys and will signalKeyAsReady for any key that was removed from the database or changed type. It is the responsibility of the code in blocked.c to ignore or act on deleted/type-changed keys. 2. Use the new signalDeletedKeyAsReady where needed blockedonkey.c + tcl: 1. Added test of new capabilities (FSL.BPOPGT now requires the key to exist in order to work)
2022-10-18 18:50:02 +02:00
scanDatabaseForDeletedKeys(db1, db2);
scanDatabaseForDeletedKeys(db2, db1);
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
/* Swap hash tables. Note that we don't swap blocking_keys,
* ready_keys and watched_keys, since we want clients to
* remain in the same DB they were. */
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
db1->keys = db2->keys;
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
db1->expires = db2->expires;
db1->avg_ttl = db2->avg_ttl;
db1->expires_cursor = db2->expires_cursor;
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
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
db2->keys = aux.keys;
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
db2->expires = aux.expires;
db2->avg_ttl = aux.avg_ttl;
db2->expires_cursor = aux.expires_cursor;
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
/* Now we need to handle clients blocked on lists: as an effect
* of swapping the two DBs, a client that was waiting for list
* X in a given DB, may now actually be unblocked if X happens
* to exist in the new version of the DB, after the swap.
*
* However normally we only do this check for efficiency reasons
* in dbAdd() when a list is created. So here we need to rescan
* the list of clients blocked on lists and signal lists as ready
* if needed. */
scanDatabaseForReadyKeys(db1);
scanDatabaseForReadyKeys(db2);
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
return C_OK;
}
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
/* Logically, this discards (flushes) the old main database, and apply the newly loaded
* database (temp) as the main (active) database, the actual freeing of old database
* (which will now be placed in the temp one) is done later. */
void swapMainDbWithTempDb(serverDb *tempDb) {
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
for (int i = 0; i < server.dbnum; i++) {
serverDb aux = server.db[i];
serverDb *activedb = &server.db[i], *newdb = &tempDb[i];
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
/* Swapping databases should make transaction fail if there is any
* client watching keys. */
touchAllWatchedKeysInDb(activedb, newdb);
/* Try to unblock any XREADGROUP clients if the key no longer exists. */
Blocked module clients should be aware when a key is deleted (#11310) The use case is a module that wants to implement a blocking command on a key that necessarily exists and wants to unblock the client in case the key is deleted (much like what we implemented for XREADGROUP in #10306) New module API: * RedisModule_BlockClientOnKeysWithFlags Flags: * REDISMODULE_BLOCK_UNBLOCK_NONE * REDISMODULE_BLOCK_UNBLOCK_DELETED ### Detailed description of code changes blocked.c: 1. Both module and stream functions are called whether the key exists or not, regardless of its type. We do that in order to allow modules/stream to unblock the client in case the key is no longer present or has changed type (the behavior for streams didn't change, just code that moved into serveClientsBlockedOnStreamKey) 2. Make sure afterCommand is called in serveClientsBlockedOnKeyByModule, in order to propagate actions from moduleTryServeClientBlockedOnKey. 3. handleClientsBlockedOnKeys: call propagatePendingCommands directly after lookupKeyReadWithFlags to prevent a possible lazy-expire DEL from being mixed with any command propagated by the preceding functions. 4. blockForKeys: Caller can specifiy that it wants to be awakened if key is deleted. Minor optimizations (use dictAddRaw). 5. signalKeyAsReady became signalKeyAsReadyLogic which can take a boolean in case the key is deleted. It will only signal if there's at least one client that awaits key deletion (to save calls to handleClientsBlockedOnKeys). Minor optimizations (use dictAddRaw) db.c: 1. scanDatabaseForDeletedStreams is now scanDatabaseForDeletedKeys and will signalKeyAsReady for any key that was removed from the database or changed type. It is the responsibility of the code in blocked.c to ignore or act on deleted/type-changed keys. 2. Use the new signalDeletedKeyAsReady where needed blockedonkey.c + tcl: 1. Added test of new capabilities (FSL.BPOPGT now requires the key to exist in order to work)
2022-10-18 18:50:02 +02:00
scanDatabaseForDeletedKeys(activedb, newdb);
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
/* Swap hash tables. Note that we don't swap blocking_keys,
* ready_keys and watched_keys, since clients
* remain in the same DB they were. */
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
activedb->keys = newdb->keys;
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
activedb->expires = newdb->expires;
activedb->avg_ttl = newdb->avg_ttl;
activedb->expires_cursor = newdb->expires_cursor;
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
newdb->keys = aux.keys;
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
newdb->expires = aux.expires;
newdb->avg_ttl = aux.avg_ttl;
newdb->expires_cursor = aux.expires_cursor;
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
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
/* Now we need to handle clients blocked on lists: as an effect
* of swapping the two DBs, a client that was waiting for list
* X in a given DB, may now actually be unblocked if X happens
* to exist in the new version of the DB, after the swap.
*
* However normally we only do this check for efficiency reasons
* in dbAdd() when a list is created. So here we need to rescan
* the list of clients blocked on lists and signal lists as ready
* if needed. */
scanDatabaseForReadyKeys(activedb);
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
}
trackingInvalidateKeysOnFlush(1);
flushReplicaKeysWithExpireList();
}
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
/* SWAPDB db1 db2 */
void swapdbCommand(client *c) {
int id1, id2;
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
/* Not allowed in cluster mode: we have just DB 0 there. */
if (server.cluster_enabled) {
addReplyError(c, "SWAPDB is not allowed in cluster mode");
return;
}
/* Get the two DBs indexes. */
if (getIntFromObjectOrReply(c, c->argv[1], &id1, "invalid first DB index") != C_OK) return;
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
if (getIntFromObjectOrReply(c, c->argv[2], &id2, "invalid second DB index") != C_OK) return;
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
/* Swap... */
if (dbSwapDatabases(id1, id2) == C_ERR) {
addReplyError(c, "DB index is out of range");
return;
} else {
ValkeyModuleSwapDbInfo si = {VALKEYMODULE_SWAPDBINFO_VERSION, id1, id2};
moduleFireServerEvent(VALKEYMODULE_EVENT_SWAPDB, 0, &si);
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
server.dirty++;
addReply(c, shared.ok);
}
}
/*-----------------------------------------------------------------------------
* Expires API
*----------------------------------------------------------------------------*/
int removeExpire(serverDb *db, robj *key) {
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
return kvstoreDictDelete(db->expires, getKVStoreIndexForKey(key->ptr), key->ptr) == DICT_OK;
}
Replication: fix the infamous key leakage of writable slaves + EXPIRE. BACKGROUND AND USE CASEj Redis slaves are normally write only, however the supprot a "writable" mode which is very handy when scaling reads on slaves, that actually need write operations in order to access data. For instance imagine having slaves replicating certain Sets keys from the master. When accessing the data on the slave, we want to peform intersections between such Sets values. However we don't want to intersect each time: to cache the intersection for some time often is a good idea. To do so, it is possible to setup a slave as a writable slave, and perform the intersection on the slave side, perhaps setting a TTL on the resulting key so that it will expire after some time. THE BUG Problem: in order to have a consistent replication, expiring of keys in Redis replication is up to the master, that synthesize DEL operations to send in the replication stream. However slaves logically expire keys by hiding them from read attempts from clients so that if the master did not promptly sent a DEL, the client still see logically expired keys as non existing. Because slaves don't actively expire keys by actually evicting them but just masking from the POV of read operations, if a key is created in a writable slave, and an expire is set, the key will be leaked forever: 1. No DEL will be received from the master, which does not know about such a key at all. 2. No eviction will be performed by the slave, since it needs to disable eviction because it's up to masters, otherwise consistency of data is lost. THE FIX In order to fix the problem, the slave should be able to tag keys that were created in the slave side and have an expire set in some way. My solution involved using an unique additional dictionary created by the writable slave only if needed. The dictionary is obviously keyed by the key name that we need to track: all the keys that are set with an expire directly by a client writing to the slave are tracked. The value in the dictionary is a bitmap of all the DBs where such a key name need to be tracked, so that we can use a single dictionary to track keys in all the DBs used by the slave (actually this limits the solution to the first 64 DBs, but the default with Redis is to use 16 DBs). This solution allows to pay both a small complexity and CPU penalty, which is zero when the feature is not used, actually. The slave-side eviction is encapsulated in code which is not coupled with the rest of the Redis core, if not for the hook to track the keys. TODO I'm doing the first smoke tests to see if the feature works as expected: so far so good. Unit tests should be added before merging into the 4.0 branch.
2016-12-13 10:20:06 +01:00
/* Set an expire to the specified key. If the expire is set in the context
* of an user calling a command 'c' is the client, otherwise 'c' is set
* to NULL. The 'when' parameter is the absolute unix time in milliseconds
* after which the key will no longer be considered valid. */
void setExpire(client *c, serverDb *db, robj *key, long long when) {
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
dictEntry *kde, *de, *existing;
/* Reuse the sds from the main dict in the expire dict */
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
int dict_index = getKVStoreIndexForKey(key->ptr);
kde = kvstoreDictFind(db->keys, dict_index, key->ptr);
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
serverAssertWithInfo(NULL, key, kde != NULL);
Optimize the per slot dictionary by checking for cluster mode earlier (#995) While doing some profiling, I noticed that getKeySlot() was a fairly large part (~0.7%) of samples doing perf with high pipeline during standalone. I think this is because we do a very late check for server.cluster_mode, we first call getKeySlot() and then call calculateKeySlot(). (calculateKeySlot was surprisingly not automatically inlined, we were doing a jump into it and then immediately returning zero). We then also do useless work in the form of caching zero in client->slot, which will further mess with cache lines. So, this PR tries to accomplish a few things things. 1) The usage of the `slot` name made a lot more sense before the introduction of the kvstore. Now with kvstore, we call this the database index, so all the references to slot in standalone are no longer really accurate. 2) Pull the cluster mode check all the way out of getKeySlot(), so hopefully a bit more performant. 3) Remove calculateKeySlot() as independent from getKeySlot(). calculateKeySlot used to have 3 call sites outside of db.c, which warranted it's own function. It's now only called in two places, pubsub.c and networking.c. I ran some profiling, and saw about ~0.3% improvement, but don't really trust it because you'll see a much higher (~2%) variance in test runs just by how the branch predictions will get changed with a new memory layout. Running perf again showed no samples in getKeySlot() and a reduction in samples in lookupKey(), so maybe this will help a little bit. --------- Signed-off-by: Madelyn Olson <madelyneolson@gmail.com> Signed-off-by: Ping Xie <pingxie@google.com>
2024-09-11 09:53:42 -07:00
de = kvstoreDictAddRaw(db->expires, dict_index, dictGetKey(kde), &existing);
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
if (existing) {
dictSetSignedIntegerVal(existing, when);
} else {
dictSetSignedIntegerVal(de, when);
}
Replication: fix the infamous key leakage of writable slaves + EXPIRE. BACKGROUND AND USE CASEj Redis slaves are normally write only, however the supprot a "writable" mode which is very handy when scaling reads on slaves, that actually need write operations in order to access data. For instance imagine having slaves replicating certain Sets keys from the master. When accessing the data on the slave, we want to peform intersections between such Sets values. However we don't want to intersect each time: to cache the intersection for some time often is a good idea. To do so, it is possible to setup a slave as a writable slave, and perform the intersection on the slave side, perhaps setting a TTL on the resulting key so that it will expire after some time. THE BUG Problem: in order to have a consistent replication, expiring of keys in Redis replication is up to the master, that synthesize DEL operations to send in the replication stream. However slaves logically expire keys by hiding them from read attempts from clients so that if the master did not promptly sent a DEL, the client still see logically expired keys as non existing. Because slaves don't actively expire keys by actually evicting them but just masking from the POV of read operations, if a key is created in a writable slave, and an expire is set, the key will be leaked forever: 1. No DEL will be received from the master, which does not know about such a key at all. 2. No eviction will be performed by the slave, since it needs to disable eviction because it's up to masters, otherwise consistency of data is lost. THE FIX In order to fix the problem, the slave should be able to tag keys that were created in the slave side and have an expire set in some way. My solution involved using an unique additional dictionary created by the writable slave only if needed. The dictionary is obviously keyed by the key name that we need to track: all the keys that are set with an expire directly by a client writing to the slave are tracked. The value in the dictionary is a bitmap of all the DBs where such a key name need to be tracked, so that we can use a single dictionary to track keys in all the DBs used by the slave (actually this limits the solution to the first 64 DBs, but the default with Redis is to use 16 DBs). This solution allows to pay both a small complexity and CPU penalty, which is zero when the feature is not used, actually. The slave-side eviction is encapsulated in code which is not coupled with the rest of the Redis core, if not for the hook to track the keys. TODO I'm doing the first smoke tests to see if the feature works as expected: so far so good. Unit tests should be added before merging into the 4.0 branch.
2016-12-13 10:20:06 +01:00
int writable_replica = server.primary_host && server.repl_replica_ro == 0;
if (c && writable_replica && !c->flag.primary) rememberReplicaKeyWithExpire(db, key);
}
/* Return the expire time of the specified key, or -1 if no expire
* is associated with this key (i.e. the key is non volatile) */
long long getExpireWithDictIndex(serverDb *db, robj *key, int dict_index) {
dictEntry *de;
if ((de = dbFindExpiresWithDictIndex(db, key->ptr, dict_index)) == NULL) return -1;
return dictGetSignedIntegerVal(de);
}
/* Return the expire time of the specified key, or -1 if no expire
* is associated with this key (i.e. the key is non volatile) */
long long getExpire(serverDb *db, robj *key) {
int dict_index = getKVStoreIndexForKey(key->ptr);
return getExpireWithDictIndex(db, key, dict_index);
}
void deleteExpiredKeyAndPropagateWithDictIndex(serverDb *db, robj *keyobj, int dict_index) {
mstime_t expire_latency;
latencyStartMonitor(expire_latency);
dbGenericDeleteWithDictIndex(db, keyobj, server.lazyfree_lazy_expire, DB_FLAG_KEY_EXPIRED, dict_index);
latencyEndMonitor(expire_latency);
latencyAddSampleIfNeeded("expire-del", expire_latency);
Reverts most of the changes of #10969 (#11178) The PR reverts the changes made on #10969. The reason for revert was trigger because of occasional test failure that started after the PR was merged. The issue is that if there is a lazy expire during the command invocation, the `del` command is added to the replication stream after the command placeholder. So the logical order on the primary is: * Delete the key (lazy expiration) * Command invocation But the replication stream gets it the other way around: * Command invocation (because the command is written into the placeholder) * Delete the key (lazy expiration) So if the command write to the key that was just lazy expired we will get inconsistency between primary and replica. One solution we considered is to add another lazy expire replication stream and write all the lazy expire there. Then when replicating, we will replicate the lazy expire replication stream first. This will solve this specific test failure but we realize that the issues does not ends here and the more we dig the more problems we find.One of the example we thought about (that can actually crashes Redis) is as follow: * User perform SINTERSTORE * When Redis tries to fetch the second input key it triggers lazy expire * The lazy expire trigger a module logic that deletes the first input key * Now Redis hold the robj of the first input key that was actually freed We believe we took the wrong approach and we will come up with another PR that solve the problem differently, for now we revert the changes so we will not have the tests failure. Notice that not the entire code was revert, some parts of the PR are changes that we would like to keep. The changes that **was** reverted are: * Saving a placeholder for replication at the beginning of the command (`call` function) * Order of the replication stream on active expire and eviction (we will decide how to handle it correctly on follow up PR) * `Spop` changes are no longer needed (because we reverted the placeholder code) Changes that **was not** reverted: * On expire/eviction, wrap the `del` and the notification effect in a multi exec. * `PropagateNow` function can still accept a special dbid, -1, indicating not to replicate select. * Keep optimisation for reusing the `alsoPropagate` array instead of allocating it each time. Tests: * All tests was kept and only few tests was modify to work correctly with the changes * Test was added to verify that the revert fixes the issues.
2022-08-24 12:51:36 +03:00
notifyKeyspaceEvent(NOTIFY_EXPIRED, "expired", keyobj, db->id);
signalModifiedKey(NULL, db, keyobj);
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
propagateDeletion(db, keyobj, server.lazyfree_lazy_expire);
server.stat_expiredkeys++;
}
/* Delete the specified expired key and propagate expire. */
void deleteExpiredKeyAndPropagate(serverDb *db, robj *keyobj) {
int dict_index = getKVStoreIndexForKey(keyobj->ptr);
deleteExpiredKeyAndPropagateWithDictIndex(db, keyobj, dict_index);
}
/* Delete the specified expired key from overwriting and propagate the DEL or UNLINK. */
void deleteExpiredKeyFromOverwriteAndPropagate(client *c, robj *keyobj) {
int deleted = dbGenericDelete(c->db, keyobj, server.lazyfree_lazy_expire, DB_FLAG_KEY_EXPIRED);
serverAssertWithInfo(c, keyobj, deleted);
server.dirty++;
/* Replicate/AOF this as an explicit DEL or UNLINK. */
robj *aux = server.lazyfree_lazy_expire ? shared.unlink : shared.del;
rewriteClientCommandVector(c, 2, aux, keyobj);
signalModifiedKey(c, c->db, keyobj);
notifyKeyspaceEvent(NOTIFY_GENERIC, "del", keyobj, c->db->id);
}
/* Propagate an implicit key deletion into replicas and the AOF file.
* When a key was deleted in the primary by eviction, expiration or a similar
* mechanism a DEL/UNLINK operation for this key is sent
* to all the replicas and the AOF file if enabled.
*
* This way the key deletion is centralized in one place, and since both
* AOF and the replication link guarantee operation ordering, everything
* will be consistent even if we allow write operations against deleted
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
* keys.
*
* This function may be called from:
* 1. Within call(): Example: Lazy-expire on key access.
* In this case the caller doesn't have to do anything
* because call() handles server.also_propagate(); or
* 2. Outside of call(): Example: Active-expire, eviction, slot ownership changed.
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
* In this the caller must remember to call
* postExecutionUnitOperations, preferably just after a
* single deletion batch, so that DEL/UNLINK will NOT be wrapped
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
* in MULTI/EXEC */
void propagateDeletion(serverDb *db, robj *key, int lazy) {
robj *argv[2];
argv[0] = lazy ? shared.unlink : shared.del;
argv[1] = key;
incrRefCount(argv[0]);
incrRefCount(argv[1]);
/* If the primary decided to delete a key we must propagate it to replicas no matter what.
* Even if module executed a command without asking for propagation. */
int prev_replication_allowed = server.replication_allowed;
server.replication_allowed = 1;
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
alsoPropagate(db->id, argv, 2, PROPAGATE_AOF | PROPAGATE_REPL);
server.replication_allowed = prev_replication_allowed;
decrRefCount(argv[0]);
decrRefCount(argv[1]);
}
int keyIsExpiredWithDictIndex(serverDb *db, robj *key, int dict_index) {
/* Don't expire anything while loading. It will be done later. */
if (server.loading) return 0;
mstime_t when = getExpireWithDictIndex(db, key, dict_index);
mstime_t now;
if (when < 0) return 0; /* No expire for this key */
Freeze time sampling during command execution, and scripts (#10300) Freeze time during execution of scripts and all other commands. This means that a key is either expired or not, and doesn't change state during a script execution. resolves #10182 This PR try to add a new `commandTimeSnapshot` function. The function logic is extracted from `keyIsExpired`, but the related calls to `fixed_time_expire` and `mstime()` are removed, see below. In commands, we will avoid calling `mstime()` multiple times and just use the one that sampled in call. The background is, e.g. using `PEXPIRE 1` with valgrind sometimes result in the key being deleted rather than expired. The reason is that both `PEXPIRE` command and `checkAlreadyExpired` call `mstime()` separately. There are other more important changes in this PR: 1. Eliminate `fixed_time_expire`, it is no longer needed. When we want to sample time we should always use a time snapshot. We will use `in_nested_call` instead to update the cached time in `call`. 2. Move the call for `updateCachedTime` from `serverCron` to `afterSleep`. Now `commandTimeSnapshot` will always return the sample time, the `lookupKeyReadWithFlags` call in `getNodeByQuery` will get a outdated cached time (because `processCommand` is out of the `call` context). We put the call to `updateCachedTime` in `aftersleep`. 3. Cache the time each time the module lock Redis. Call `updateCachedTime` in `moduleGILAfterLock`, affecting `RM_ThreadSafeContextLock` and `RM_ThreadSafeContextTryLock` Currently the commandTimeSnapshot change affects the following TTL commands: - SET EX / SET PX - EXPIRE / PEXPIRE - SETEX / PSETEX - GETEX EX / GETEX PX - TTL / PTTL - EXPIRETIME / PEXPIRETIME - RESTORE key TTL And other commands just use the cached mstime (including TIME). This is considered to be a breaking change since it can break a script that uses a loop to wait for a key to expire.
2022-10-09 13:18:34 +08:00
now = commandTimeSnapshot();
/* The key expired if the current (virtual or real) time is greater
* than the expire time of the key. */
return now > when;
}
/* Check if the key is expired. */
int keyIsExpired(serverDb *db, robj *key) {
int dict_index = getKVStoreIndexForKey(key->ptr);
return keyIsExpiredWithDictIndex(db, key, dict_index);
}
keyStatus expireIfNeededWithDictIndex(serverDb *db, robj *key, int flags, int dict_index) {
if (server.lazy_expire_disabled) return KEY_VALID;
if (!keyIsExpiredWithDictIndex(db, key, dict_index)) return KEY_VALID;
2018-10-24 12:26:27 +02:00
/* If we are running in the context of a replica, instead of
2018-10-24 12:26:27 +02:00
* evicting the expired key from the database, we return ASAP:
* the replica key expiration is controlled by the primary that will
* send us synthesized DEL operations for expired keys. The
* exception is when write operations are performed on writable
* replicas.
2018-10-24 12:26:27 +02:00
*
* Still we try to return the right information to the caller,
* that is, KEY_VALID if we think the key should still be valid,
* KEY_EXPIRED if we think the key is expired but don't want to delete it at this time.
*
* When replicating commands from the primary, keys are never considered
* expired. */
if (server.primary_host != NULL) {
if (server.current_client && (server.current_client->flag.primary)) return KEY_VALID;
if (!(flags & EXPIRE_FORCE_DELETE_EXPIRED)) return KEY_EXPIRED;
}
/* In some cases we're explicitly instructed to return an indication of a
* missing key without actually deleting it, even on primaries. */
if (flags & EXPIRE_AVOID_DELETE_EXPIRED) return KEY_EXPIRED;
/* If 'expire' action is paused, for whatever reason, then don't expire any key.
* Typically, at the end of the pause we will properly expire the key OR we
* will have failed over and the new primary will send us the expire. */
if (isPausedActionsWithUpdate(PAUSE_ACTION_EXPIRE)) return KEY_EXPIRED;
/* The key needs to be converted from static to heap before deleted */
int static_key = key->refcount == OBJ_STATIC_REFCOUNT;
if (static_key) {
key = createStringObject(key->ptr, sdslen(key->ptr));
}
/* Delete the key */
deleteExpiredKeyAndPropagateWithDictIndex(db, key, dict_index);
if (static_key) {
decrRefCount(key);
}
return KEY_DELETED;
}
/* This function is called when we are going to perform some operation
* in a given key, but such key may be already logically expired even if
* it still exists in the database. The main way this function is called
* is via lookupKey*() family of functions.
*
* The behavior of the function depends on the replication role of the
* instance, because by default replicas do not delete expired keys. They
* wait for DELs from the primary for consistency matters. However even
* replicas will try to have a coherent return value for the function,
* so that read commands executed in the replica side will be able to
* behave like if the key is expired even if still present (because the
* primary has yet to propagate the DEL).
*
* In primary as a side effect of finding a key which is expired, such
* key will be evicted from the database. Also this may trigger the
* propagation of a DEL/UNLINK command in AOF / replication stream.
*
* On replicas, this function does not delete expired keys by default, but
* it still returns KEY_EXPIRED if the key is logically expired. To force deletion
* of logically expired keys even on replicas, use the EXPIRE_FORCE_DELETE_EXPIRED
* flag. Note though that if the current client is executing
* replicated commands from the primary, keys are never considered expired.
*
* On the other hand, if you just want expiration check, but need to avoid
* the actual key deletion and propagation of the deletion, use the
* EXPIRE_AVOID_DELETE_EXPIRED flag.
*
* The return value of the function is KEY_VALID if the key is still valid.
* The function returns KEY_EXPIRED if the key is expired BUT not deleted,
* or returns KEY_DELETED if the key is expired and deleted. */
keyStatus expireIfNeeded(serverDb *db, robj *key, int flags) {
int dict_index = getKVStoreIndexForKey(key->ptr);
return expireIfNeededWithDictIndex(db, key, flags, dict_index);
}
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
/* CB passed to kvstoreExpand.
* The purpose is to skip expansion of unused dicts in cluster mode (all
* dicts not mapped to *my* slots) */
static int dbExpandSkipSlot(int slot) {
return !clusterNodeCoversSlot(getMyClusterNode(), slot);
}
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
/*
* This functions increases size of the main/expires db to match desired number.
* In cluster mode resizes all individual dictionaries for slots that this node owns.
*
* Based on the parameter `try_expand`, appropriate dict expand API is invoked.
* if try_expand is set to 1, `dictTryExpand` is used else `dictExpand`.
* The return code is either `DICT_OK`/`DICT_ERR` for both the API(s).
* `DICT_OK` response is for successful expansion. However ,`DICT_ERR` response signifies failure in allocation in
* `dictTryExpand` call and in case of `dictExpand` call it signifies no expansion was performed.
*/
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
static int dbExpandGeneric(kvstore *kvs, uint64_t db_size, int try_expand) {
int ret;
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
if (server.cluster_enabled) {
/* We don't know exact number of keys that would fall into each slot, but we can
* approximate it, assuming even distribution, divide it by the number of slots. */
int slots = getMyShardSlotCount();
if (slots == 0) return C_OK;
db_size = db_size / slots;
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
ret = kvstoreExpand(kvs, db_size, try_expand, dbExpandSkipSlot);
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
} else {
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
ret = kvstoreExpand(kvs, db_size, try_expand, NULL);
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
}
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
return ret ? C_OK : C_ERR;
}
int dbExpand(serverDb *db, uint64_t db_size, int try_expand) {
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
return dbExpandGeneric(db->keys, db_size, try_expand);
}
int dbExpandExpires(serverDb *db, uint64_t db_size, int try_expand) {
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
return dbExpandGeneric(db->expires, db_size, try_expand);
}
dictEntry *dbFindWithDictIndex(serverDb *db, void *key, int dict_index) {
return kvstoreDictFind(db->keys, dict_index, 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
}
dictEntry *dbFind(serverDb *db, void *key) {
int dict_index = getKVStoreIndexForKey(key);
return dbFindWithDictIndex(db, key, dict_index);
}
dictEntry *dbFindExpiresWithDictIndex(serverDb *db, void *key, int dict_index) {
return kvstoreDictFind(db->expires, dict_index, 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
}
dictEntry *dbFindExpires(serverDb *db, void *key) {
int dict_index = getKVStoreIndexForKey(key);
return dbFindExpiresWithDictIndex(db, key, dict_index);
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
}
unsigned long long dbSize(serverDb *db) {
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
return kvstoreSize(db->keys);
}
unsigned long long dbScan(serverDb *db, unsigned long long cursor, dictScanFunction *scan_cb, void *privdata) {
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
return kvstoreScan(db->keys, cursor, -1, scan_cb, NULL, privdata);
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
}
/* -----------------------------------------------------------------------------
* API to get key arguments from commands
* ---------------------------------------------------------------------------*/
/* Prepare the getKeysResult struct to hold numkeys, either by using the
* pre-allocated keysbuf or by allocating a new array on the heap.
*
* This function must be called at least once before starting to populate
* the result, and can be called repeatedly to enlarge the result array.
*/
keyReference *getKeysPrepareResult(getKeysResult *result, int numkeys) {
/* initGetKeysResult initializes keys to NULL, point it to the pre-allocated stack
* buffer here. */
if (!result->keys) {
serverAssert(!result->numkeys);
result->keys = result->keysbuf;
}
/* Resize if necessary */
if (numkeys > result->size) {
if (result->keys != result->keysbuf) {
/* We're not using a static buffer, just (re)alloc */
result->keys = zrealloc(result->keys, numkeys * sizeof(keyReference));
} else {
/* We are using a static buffer, copy its contents */
result->keys = zmalloc(numkeys * sizeof(keyReference));
if (result->numkeys) memcpy(result->keys, result->keysbuf, result->numkeys * sizeof(keyReference));
}
result->size = numkeys;
}
return result->keys;
}
/* Returns a bitmask with all the flags found in any of the key specs of the command.
* The 'inv' argument means we'll return a mask with all flags that are missing in at least one spec. */
int64_t getAllKeySpecsFlags(struct serverCommand *cmd, int inv) {
int64_t flags = 0;
for (int j = 0; j < cmd->key_specs_num; j++) {
keySpec *spec = cmd->key_specs + j;
flags |= inv ? ~spec->flags : spec->flags;
}
return flags;
}
/* Fetch the keys based of the provided key specs. Returns the number of keys found, or -1 on error.
* There are several flags that can be used to modify how this function finds keys in a command.
*
* GET_KEYSPEC_INCLUDE_NOT_KEYS: Return 'fake' keys as if they were keys.
* GET_KEYSPEC_RETURN_PARTIAL: Skips invalid and incomplete keyspecs but returns the keys
* found in other valid keyspecs.
*/
int getKeysUsingKeySpecs(struct serverCommand *cmd, robj **argv, int argc, int search_flags, getKeysResult *result) {
int j, i, last, first, step;
keyReference *keys;
serverAssert(result->numkeys == 0); /* caller should initialize or reset it */
for (j = 0; j < cmd->key_specs_num; j++) {
keySpec *spec = cmd->key_specs + j;
serverAssert(spec->begin_search_type != KSPEC_BS_INVALID);
/* Skip specs that represent 'fake' keys */
if ((spec->flags & CMD_KEY_NOT_KEY) && !(search_flags & GET_KEYSPEC_INCLUDE_NOT_KEYS)) {
continue;
}
first = 0;
if (spec->begin_search_type == KSPEC_BS_INDEX) {
first = spec->bs.index.pos;
} else if (spec->begin_search_type == KSPEC_BS_KEYWORD) {
int start_index =
spec->bs.keyword.startfrom > 0 ? spec->bs.keyword.startfrom : argc + spec->bs.keyword.startfrom;
int end_index = spec->bs.keyword.startfrom > 0 ? argc - 1 : 1;
for (i = start_index; i != end_index; i = start_index <= end_index ? i + 1 : i - 1) {
if (i >= argc || i < 1) break;
if (!strcasecmp((char *)argv[i]->ptr, spec->bs.keyword.keyword)) {
first = i + 1;
break;
}
}
/* keyword not found */
if (!first) {
continue;
}
} else {
/* unknown spec */
goto invalid_spec;
}
if (spec->find_keys_type == KSPEC_FK_RANGE) {
step = spec->fk.range.keystep;
if (spec->fk.range.lastkey >= 0) {
last = first + spec->fk.range.lastkey;
} else {
if (!spec->fk.range.limit) {
last = argc + spec->fk.range.lastkey;
} else {
serverAssert(spec->fk.range.lastkey == -1);
last = first + ((argc - first) / spec->fk.range.limit + spec->fk.range.lastkey);
}
}
} else if (spec->find_keys_type == KSPEC_FK_KEYNUM) {
step = spec->fk.keynum.keystep;
long long numkeys;
if (spec->fk.keynum.keynumidx >= argc) goto invalid_spec;
sds keynum_str = argv[first + spec->fk.keynum.keynumidx]->ptr;
if (!string2ll(keynum_str, sdslen(keynum_str), &numkeys) || numkeys < 0) {
/* Unable to parse the numkeys argument or it was invalid */
goto invalid_spec;
}
first += spec->fk.keynum.firstkey;
last = first + (int)numkeys - 1;
} else {
/* unknown spec */
goto invalid_spec;
}
int count = ((last - first) + 1);
keys = getKeysPrepareResult(result, result->numkeys + count);
/* First or last is out of bounds, which indicates a syntax error */
if (last >= argc || last < first || first >= argc) {
goto invalid_spec;
}
for (i = first; i <= last; i += step) {
if (i >= argc || i < first) {
/* Modules commands, and standard commands with a not fixed number
* of arguments (negative arity parameter) do not have dispatch
* time arity checks, so we need to handle the case where the user
* passed an invalid number of arguments here. In this case we
* return no keys and expect the command implementation to report
* an arity or syntax error. */
if (cmd->flags & CMD_MODULE || cmd->arity < 0) {
continue;
} else {
serverPanic("%s built-in command declared keys positions"
" not matching the arity requirements.",
server.extended_redis_compat ? "Redis" : "Valkey");
}
}
keys[result->numkeys].pos = i;
keys[result->numkeys].flags = spec->flags;
result->numkeys++;
}
/* Handle incomplete specs (only after we added the current spec
* to `keys`, just in case GET_KEYSPEC_RETURN_PARTIAL was given) */
if (spec->flags & CMD_KEY_INCOMPLETE) {
goto invalid_spec;
}
/* Done with this spec */
continue;
invalid_spec:
if (search_flags & GET_KEYSPEC_RETURN_PARTIAL) {
continue;
} else {
result->numkeys = 0;
return -1;
}
}
return result->numkeys;
}
/* Return all the arguments that are keys in the command passed via argc / argv.
* This function will eventually replace getKeysFromCommand.
*
* The command returns the positions of all the key arguments inside the array,
* so the actual return value is a heap allocated array of integers. The
* length of the array is returned by reference into *numkeys.
*
* Along with the position, this command also returns the flags that are
* associated with how the server will access the key.
*
* 'cmd' must be point to the corresponding entry into the serverCommand
* table, according to the command name in argv[0]. */
int getKeysFromCommandWithSpecs(struct serverCommand *cmd,
robj **argv,
int argc,
int search_flags,
getKeysResult *result) {
/* The command has at least one key-spec not marked as NOT_KEY */
int has_keyspec = (getAllKeySpecsFlags(cmd, 1) & CMD_KEY_NOT_KEY);
/* The command has at least one key-spec marked as VARIABLE_FLAGS */
int has_varflags = (getAllKeySpecsFlags(cmd, 0) & CMD_KEY_VARIABLE_FLAGS);
/* We prefer key-specs if there are any, and their flags are reliable. */
if (has_keyspec && !has_varflags) {
int ret = getKeysUsingKeySpecs(cmd, argv, argc, search_flags, result);
if (ret >= 0) return ret;
/* If the specs returned with an error (probably an INVALID or INCOMPLETE spec),
* fallback to the callback method. */
}
/* Resort to getkeys callback methods. */
if (cmd->flags & CMD_MODULE_GETKEYS) return moduleGetCommandKeysViaAPI(cmd, argv, argc, result);
/* We use native getkeys as a last resort, since not all these native getkeys provide
* flags properly (only the ones that correspond to INVALID, INCOMPLETE or VARIABLE_FLAGS do.*/
if (cmd->getkeys_proc) return cmd->getkeys_proc(cmd, argv, argc, result);
return 0;
}
/* This function returns a sanity check if the command may have keys. */
int doesCommandHaveKeys(struct serverCommand *cmd) {
return cmd->getkeys_proc || /* has getkeys_proc (non modules) */
(cmd->flags & CMD_MODULE_GETKEYS) || /* module with GETKEYS */
(getAllKeySpecsFlags(cmd, 1) & CMD_KEY_NOT_KEY); /* has at least one key-spec not marked as NOT_KEY */
}
/* A simplified channel spec table that contains all of the commands
* and which channels they have and how they are accessed. */
typedef struct ChannelSpecs {
serverCommandProc *proc; /* Command procedure to match against */
uint64_t flags; /* CMD_CHANNEL_* flags for this command */
int start; /* The initial position of the first channel */
int count; /* The number of channels, or -1 if all remaining
* arguments are channels. */
} ChannelSpecs;
ChannelSpecs commands_with_channels[] = {
{subscribeCommand, CMD_CHANNEL_SUBSCRIBE, 1, -1},
{ssubscribeCommand, CMD_CHANNEL_SUBSCRIBE, 1, -1},
{unsubscribeCommand, CMD_CHANNEL_UNSUBSCRIBE, 1, -1},
{sunsubscribeCommand, CMD_CHANNEL_UNSUBSCRIBE, 1, -1},
{psubscribeCommand, CMD_CHANNEL_PATTERN | CMD_CHANNEL_SUBSCRIBE, 1, -1},
{punsubscribeCommand, CMD_CHANNEL_PATTERN | CMD_CHANNEL_UNSUBSCRIBE, 1, -1},
{publishCommand, CMD_CHANNEL_PUBLISH, 1, 1},
{spublishCommand, CMD_CHANNEL_PUBLISH, 1, 1},
{NULL, 0} /* Terminator. */
};
/* Returns 1 if the command may access any channels matched by the flags
* argument. */
int doesCommandHaveChannelsWithFlags(struct serverCommand *cmd, int flags) {
/* If a module declares get channels, we are just going to assume
* has channels. This API is allowed to return false positives. */
if (cmd->flags & CMD_MODULE_GETCHANNELS) {
return 1;
}
for (ChannelSpecs *spec = commands_with_channels; spec->proc != NULL; spec += 1) {
if (cmd->proc == spec->proc) {
return !!(spec->flags & flags);
}
}
return 0;
}
/* Return all the arguments that are channels in the command passed via argc / argv.
* This function behaves similar to getKeysFromCommandWithSpecs, but with channels
* instead of keys.
*
* The command returns the positions of all the channel arguments inside the array,
* so the actual return value is a heap allocated array of integers. The
* length of the array is returned by reference into *numkeys.
*
* Along with the position, this command also returns the flags that are
* associated with how the server will access the channel.
*
* 'cmd' must be point to the corresponding entry into the serverCommand
* table, according to the command name in argv[0]. */
int getChannelsFromCommand(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
keyReference *keys;
/* If a module declares get channels, use that. */
if (cmd->flags & CMD_MODULE_GETCHANNELS) {
return moduleGetCommandChannelsViaAPI(cmd, argv, argc, result);
}
/* Otherwise check the channel spec table */
for (ChannelSpecs *spec = commands_with_channels; spec != NULL; spec += 1) {
if (cmd->proc == spec->proc) {
int start = spec->start;
int stop = (spec->count == -1) ? argc : start + spec->count;
if (stop > argc) stop = argc;
int count = 0;
keys = getKeysPrepareResult(result, stop - start);
for (int i = start; i < stop; i++) {
keys[count].pos = i;
keys[count++].flags = spec->flags;
}
result->numkeys = count;
return count;
}
}
return 0;
}
/* The base case is to use the keys position as given in the command table
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
* (firstkey, lastkey, step).
* This function works only on command with the legacy_range_key_spec,
* all other commands should be handled by getkeys_proc.
*
* If the commands keyspec is incomplete, no keys will be returned, and the provided
* keys function should be called instead.
*
* NOTE: This function does not guarantee populating the flags for
* the keys, in order to get flags you should use getKeysUsingKeySpecs. */
int getKeysUsingLegacyRangeSpec(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
int j, i = 0, last, first, step;
keyReference *keys;
2015-07-27 09:41:48 +02:00
UNUSED(argv);
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
if (cmd->legacy_range_key_spec.begin_search_type == KSPEC_BS_INVALID) {
result->numkeys = 0;
return 0;
}
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
first = cmd->legacy_range_key_spec.bs.index.pos;
last = cmd->legacy_range_key_spec.fk.range.lastkey;
if (last >= 0) last += first;
step = cmd->legacy_range_key_spec.fk.range.keystep;
if (last < 0) last = argc + last;
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 count = ((last - first) + 1);
keys = getKeysPrepareResult(result, count);
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
for (j = first; j <= last; j += step) {
if (j >= argc || j < first) {
/* Modules commands, and standard commands with a not fixed number
2018-07-01 13:24:50 +08:00
* of arguments (negative arity parameter) do not have dispatch
* time arity checks, so we need to handle the case where the user
* passed an invalid number of arguments here. In this case we
* return no keys and expect the command implementation to report
* an arity or syntax error. */
if (cmd->flags & CMD_MODULE || cmd->arity < 0) {
result->numkeys = 0;
return 0;
} else {
serverPanic("%s built-in command declared keys positions"
" not matching the arity requirements.",
server.extended_redis_compat ? "Redis" : "Valkey");
}
}
keys[i].pos = j;
/* Flags are omitted from legacy key specs */
keys[i++].flags = 0;
}
result->numkeys = i;
return i;
}
/* Return all the arguments that are keys in the command passed via argc / argv.
*
* The command returns the positions of all the key arguments inside the array,
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
* so the actual return value is a heap allocated array of integers. The
* length of the array is returned by reference into *numkeys.
*
* 'cmd' must be point to the corresponding entry into the serverCommand
* table, according to the command name in argv[0].
*
* This function uses the command table if a command-specific helper function
* is not required, otherwise it calls the command-specific function. */
int getKeysFromCommand(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
if (cmd->flags & CMD_MODULE_GETKEYS) {
return moduleGetCommandKeysViaAPI(cmd, argv, argc, result);
} else if (cmd->getkeys_proc) {
return cmd->getkeys_proc(cmd, argv, argc, result);
} else {
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
return getKeysUsingLegacyRangeSpec(cmd, argv, argc, result);
}
}
/* Free the result of getKeysFromCommand. */
void getKeysFreeResult(getKeysResult *result) {
if (result && result->keys != result->keysbuf) zfree(result->keys);
}
/* Helper function to extract keys from following commands:
* COMMAND [destkey] <num-keys> <key> [...] <key> [...] ... <options>
*
* eg:
* ZUNION <num-keys> <key> <key> ... <key> <options>
* ZUNIONSTORE <destkey> <num-keys> <key> <key> ... <key> <options>
*
* 'storeKeyOfs': destkey index, 0 means destkey not exists.
* 'keyCountOfs': num-keys index.
* 'firstKeyOfs': firstkey index.
* 'keyStep': the interval of each key, usually this value is 1.
*
* The commands using this function have a fully defined keyspec, so returning flags isn't needed. */
int genericGetKeys(int storeKeyOfs,
int keyCountOfs,
int firstKeyOfs,
int keyStep,
robj **argv,
int argc,
getKeysResult *result) {
int i, num;
keyReference *keys;
num = atoi(argv[keyCountOfs]->ptr);
/* Sanity check. Don't return any key if the command is going to
* reply with syntax error. (no input keys). */
if (num < 1 || num > (argc - firstKeyOfs) / keyStep) {
result->numkeys = 0;
return 0;
}
int numkeys = storeKeyOfs ? num + 1 : num;
keys = getKeysPrepareResult(result, numkeys);
result->numkeys = numkeys;
/* Add all key positions for argv[firstKeyOfs...n] to keys[] */
for (i = 0; i < num; i++) {
keys[i].pos = firstKeyOfs + (i * keyStep);
keys[i].flags = 0;
}
if (storeKeyOfs) {
keys[num].pos = storeKeyOfs;
keys[num].flags = 0;
}
return result->numkeys;
}
int sintercardGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
UNUSED(cmd);
return genericGetKeys(0, 1, 2, 1, argv, argc, result);
}
int zunionInterDiffStoreGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
UNUSED(cmd);
return genericGetKeys(1, 2, 3, 1, argv, argc, result);
}
int zunionInterDiffGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
UNUSED(cmd);
return genericGetKeys(0, 1, 2, 1, argv, argc, result);
}
int evalGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
2015-07-27 09:41:48 +02:00
UNUSED(cmd);
return genericGetKeys(0, 2, 3, 1, argv, argc, result);
}
int functionGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
UNUSED(cmd);
return genericGetKeys(0, 2, 3, 1, argv, argc, result);
}
int lmpopGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
UNUSED(cmd);
return genericGetKeys(0, 1, 2, 1, argv, argc, result);
}
int blmpopGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
UNUSED(cmd);
return genericGetKeys(0, 2, 3, 1, argv, argc, result);
}
int zmpopGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
UNUSED(cmd);
return genericGetKeys(0, 1, 2, 1, argv, argc, result);
}
int bzmpopGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
UNUSED(cmd);
return genericGetKeys(0, 2, 3, 1, argv, argc, result);
}
/* Helper function to extract keys from the SORT RO command.
*
* SORT <sort-key>
*
* The second argument of SORT is always a key, however an arbitrary number of
* keys may be accessed while doing the sort (the BY and GET args), so the
* key-spec declares incomplete keys which is why we have to provide a concrete
* implementation to fetch the keys.
*
* This command declares incomplete keys, so the flags are correctly set for this function */
int sortROGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
keyReference *keys;
UNUSED(cmd);
UNUSED(argv);
UNUSED(argc);
keys = getKeysPrepareResult(result, 1);
keys[0].pos = 1; /* <sort-key> is always present. */
keys[0].flags = CMD_KEY_RO | CMD_KEY_ACCESS;
result->numkeys = 1;
return result->numkeys;
}
/* Helper function to extract keys from the SORT command.
*
* SORT <sort-key> ... STORE <store-key> ...
*
* The first argument of SORT is always a key, however a list of options
* follow in SQL-alike style. Here we parse just the minimum in order to
* correctly identify keys in the "STORE" option.
*
* This command declares incomplete keys, so the flags are correctly set for this function */
int sortGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
int i, j, num, found_store = 0;
keyReference *keys;
2015-07-27 09:41:48 +02:00
UNUSED(cmd);
num = 0;
keys = getKeysPrepareResult(result, 2); /* Alloc 2 places for the worst case. */
keys[num].pos = 1; /* <sort-key> is always present. */
keys[num++].flags = CMD_KEY_RO | CMD_KEY_ACCESS;
/* Search for STORE option. By default we consider options to don't
* have arguments, so if we find an unknown option name we scan the
* next. However there are options with 1 or 2 arguments, so we
* provide a list here in order to skip the right number of args. */
struct {
char *name;
int skip;
} skiplist[] = {
{"limit", 2}, {"get", 1}, {"by", 1}, {NULL, 0} /* End of elements. */
};
for (i = 2; i < argc; i++) {
for (j = 0; skiplist[j].name != NULL; j++) {
if (!strcasecmp(argv[i]->ptr, skiplist[j].name)) {
i += skiplist[j].skip;
break;
} else if (!strcasecmp(argv[i]->ptr, "store") && i + 1 < argc) {
/* Note: we don't increment "num" here and continue the loop
* to be sure to process the *last* "STORE" option if multiple
* ones are provided. This is same behavior as SORT. */
found_store = 1;
keys[num].pos = i + 1; /* <store-key> */
keys[num].flags = CMD_KEY_OW | CMD_KEY_UPDATE;
break;
}
}
}
result->numkeys = num + found_store;
return result->numkeys;
}
/* This command declares incomplete keys, so the flags are correctly set for this function */
int migrateGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
int i, j, num, first;
keyReference *keys;
UNUSED(cmd);
/* Assume the obvious form. */
first = 3;
num = 1;
/* But check for the extended one with the KEYS option. */
struct {
char *name;
int skip;
} skip_keywords[] = {{"copy", 0}, {"replace", 0}, {"auth", 1}, {"auth2", 2}, {NULL, 0}};
if (argc > 6) {
for (i = 6; i < argc; i++) {
if (!strcasecmp(argv[i]->ptr, "keys")) {
if (sdslen(argv[3]->ptr) > 0) {
/* This is a syntax error. So ignore the keys and leave
* the syntax error to be handled by migrateCommand. */
num = 0;
} else {
first = i + 1;
num = argc - first;
}
break;
}
for (j = 0; skip_keywords[j].name != NULL; j++) {
if (!strcasecmp(argv[i]->ptr, skip_keywords[j].name)) {
i += skip_keywords[j].skip;
break;
}
}
}
}
keys = getKeysPrepareResult(result, num);
for (i = 0; i < num; i++) {
keys[i].pos = first + i;
keys[i].flags = CMD_KEY_RW | CMD_KEY_ACCESS | CMD_KEY_DELETE;
}
result->numkeys = num;
return num;
}
/* Helper function to extract keys from following commands:
* GEORADIUS key x y radius unit [WITHDIST] [WITHHASH] [WITHCOORD] [ASC|DESC]
* [COUNT count] [STORE key|STOREDIST key]
* GEORADIUSBYMEMBER key member radius unit ... options ...
*
* This command has a fully defined keyspec, so returning flags isn't needed. */
int georadiusGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
int i, num;
keyReference *keys;
UNUSED(cmd);
/* Check for the presence of the stored key in the command */
int stored_key = -1;
for (i = 5; i < argc; i++) {
char *arg = argv[i]->ptr;
/* For the case when user specifies both "store" and "storedist" options, the
2018-02-11 21:02:07 +08:00
* second key specified would override the first key. This behavior is kept
* the same as in georadiusCommand method.
*/
if ((!strcasecmp(arg, "store") || !strcasecmp(arg, "storedist")) && ((i + 1) < argc)) {
stored_key = i + 1;
i++;
}
}
num = 1 + (stored_key == -1 ? 0 : 1);
/* Keys in the command come from two places:
* argv[1] = key,
* argv[5...n] = stored key if present
*/
keys = getKeysPrepareResult(result, num);
/* Add all key positions to keys[] */
keys[0].pos = 1;
keys[0].flags = 0;
if (num > 1) {
keys[1].pos = stored_key;
keys[1].flags = 0;
}
result->numkeys = num;
return num;
}
2017-09-08 11:40:16 +02:00
/* XREAD [BLOCK <milliseconds>] [COUNT <count>] [GROUP <groupname> <ttl>]
* STREAMS key_1 key_2 ... key_N ID_1 ID_2 ... ID_N
*
* This command has a fully defined keyspec, so returning flags isn't needed. */
int xreadGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
int i, num = 0;
keyReference *keys;
2017-09-08 11:40:16 +02:00
UNUSED(cmd);
/* We need to parse the options of the command in order to seek the first
* "STREAMS" string which is actually the option. This is needed because
* "STREAMS" could also be the name of the consumer group and even the
* name of the stream key. */
2017-09-08 11:40:16 +02:00
int streams_pos = -1;
for (i = 1; i < argc; i++) {
char *arg = argv[i]->ptr;
if (!strcasecmp(arg, "block")) {
i++; /* Skip option argument. */
} else if (!strcasecmp(arg, "count")) {
i++; /* Skip option argument. */
} else if (!strcasecmp(arg, "group")) {
i += 2; /* Skip option argument. */
} else if (!strcasecmp(arg, "noack")) {
/* Nothing to do. */
} else if (!strcasecmp(arg, "streams")) {
streams_pos = i;
break;
} else {
break; /* Syntax error. */
}
2017-09-08 11:40:16 +02:00
}
2017-09-08 11:51:53 +02:00
if (streams_pos != -1) num = argc - streams_pos - 1;
2017-09-08 11:40:16 +02:00
/* Syntax error. */
if (streams_pos == -1 || num == 0 || num % 2 != 0) {
result->numkeys = 0;
return 0;
2017-09-08 11:40:16 +02:00
}
2017-09-08 11:51:53 +02:00
num /= 2; /* We have half the keys as there are arguments because
there are also the IDs, one per key. */
2017-09-08 11:40:16 +02:00
keys = getKeysPrepareResult(result, num);
for (i = streams_pos + 1; i < argc - num; i++) {
keys[i - streams_pos - 1].pos = i;
keys[i - streams_pos - 1].flags = 0;
}
result->numkeys = num;
return num;
2017-09-08 11:40:16 +02:00
}
/* Helper function to extract keys from the SET command, which may have
* a read flag if the GET argument is passed in. */
int setGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
keyReference *keys;
UNUSED(cmd);
keys = getKeysPrepareResult(result, 1);
keys[0].pos = 1; /* We always know the position */
result->numkeys = 1;
for (int i = 3; i < argc; i++) {
char *arg = argv[i]->ptr;
if ((arg[0] == 'g' || arg[0] == 'G') && (arg[1] == 'e' || arg[1] == 'E') && (arg[2] == 't' || arg[2] == 'T') &&
arg[3] == '\0') {
keys[0].flags = CMD_KEY_RW | CMD_KEY_ACCESS | CMD_KEY_UPDATE;
return 1;
}
}
keys[0].flags = CMD_KEY_OW | CMD_KEY_UPDATE;
return 1;
}
/* Helper function to extract keys from the BITFIELD command, which may be
* read-only if the BITFIELD GET subcommand is used. */
int bitfieldGetKeys(struct serverCommand *cmd, robj **argv, int argc, getKeysResult *result) {
keyReference *keys;
int readonly = 1;
UNUSED(cmd);
keys = getKeysPrepareResult(result, 1);
keys[0].pos = 1; /* We always know the position */
result->numkeys = 1;
for (int i = 2; i < argc; i++) {
int remargs = argc - i - 1; /* Remaining args other than current. */
char *arg = argv[i]->ptr;
if (!strcasecmp(arg, "get") && remargs >= 2) {
i += 2;
} else if ((!strcasecmp(arg, "set") || !strcasecmp(arg, "incrby")) && remargs >= 3) {
readonly = 0;
i += 3;
break;
} else if (!strcasecmp(arg, "overflow") && remargs >= 1) {
i += 1;
} else {
readonly = 0; /* Syntax error. safer to assume non-RO. */
break;
}
}
if (readonly) {
keys[0].flags = CMD_KEY_RO | CMD_KEY_ACCESS;
} else {
keys[0].flags = CMD_KEY_RW | CMD_KEY_ACCESS | CMD_KEY_UPDATE;
}
return 1;
}