futriix/src/ae.c

572 lines
20 KiB
C
Raw Normal View History

2009-03-22 10:30:00 +01:00
/* A simple event-driven programming library. Originally I wrote this code
* for the Jim's event-loop (Jim is a Tcl interpreter) but later translated
* it in form of a library for easy reuse.
*
* Copyright (c) 2006-2010, Redis Ltd.
2009-03-22 10:30:00 +01:00
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* * Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of Redis nor the names of its contributors may be used
* to endorse or promote products derived from this software without
* specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
#include "ae.h"
#include "anet.h"
#include "serverassert.h"
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
2009-03-22 10:30:00 +01:00
#include <stdio.h>
#include <sys/time.h>
#include <sys/types.h>
#include <unistd.h>
#include <stdlib.h>
#include <poll.h>
2012-02-08 22:24:59 +01:00
#include <string.h>
#include <time.h>
#include <errno.h>
2009-03-22 10:30:00 +01:00
#include "zmalloc.h"
#include "config.h"
/* Include the best multiplexing layer supported by this system.
* The following should be ordered by performances, descending. */
2012-03-26 17:58:19 -07:00
#ifdef HAVE_EVPORT
#include "ae_evport.c"
#else
#ifdef HAVE_EPOLL
#include "ae_epoll.c"
#else
#ifdef HAVE_KQUEUE
#include "ae_kqueue.c"
#else
#include "ae_select.c"
#endif
#endif
#endif
2009-03-22 10:30:00 +01:00
Io thread work offload (#763) ### IO-Threads Work Offloading This PR is the 2nd of 3 PRs intended to achieve the goal of 1M requests per second. (1st PR: https://github.com/valkey-io/valkey/pull/758) This PR offloads additional work to the I/O threads, beyond the current read-parse/write operations, to better utilize the I/O threads and reduce the load on the main thread. It contains the following 3 commits: ### Poll Offload Currently, the main thread is responsible for executing the poll-wait system call, while the IO threads wait for tasks from the main thread. The poll-wait operation is expensive and can consume up to 30% of the main thread's time. We could have let the IO threads do the poll-wait by themselves, with each thread listening to some of the clients and notifying the main thread when a client's command is ready to execute. However, the current approach, where the main thread listens for events from the network, has several benefits. The main thread remains in charge, allowing it to know the state of each client (idle/read/write/close) at any given time. Additionally, it makes the threads flexible, enabling us to drain an IO thread's job queue and stop a thread when the load is light without modifying the event loop and moving its clients to a different IO thread. Furthermore, with this approach, the IO threads don't need to wait for both messages from the network and from the main thread instead, the threads wait only for tasks from the main thread. To enjoy the benefits of both the main thread remaining in charge and the poll being offloaded, we propose offloading the poll-wait as a single-time, non-blocking job to one of the IO threads. The IO thread will perform a poll-wait non-blocking call while the main thread processes the client commands. Later, in `aeProcessEvents`, instead of sleeping on the poll, we check for the IO thread's poll-wait results. The poll-wait will be offloaded in `beforeSleep` only when there are ready events for the main thread to process. If no events are pending, the main thread will revert to the current behavior and sleep on the poll by itself. **Implementation Details** A new call back `custompoll` was added to the `aeEventLoop` when not set to `NULL` the ae will call the `custompoll` callback instead of the `aeApiPoll`. When the poll is offloaded we will set the `custompoll` to `getIOThreadPollResults` and send a poll-job to the thread. the thread will take a mutex, call a non-blocking (with timeout 0) to `aePoll` which will populate the fired events array. the IO thread will set the `server.io_fired_events` to the number of the returning `numevents`, later the main-thread in `custompoll` will return the `server.io_fired_events` and will set the `customPoll` back to `NULL`. To ensure thread safety when accessing server.el, all functions that modify the eventloop events were wrapped with a mutex to ensure mutual exclusion when modifying the events. ### Command Lookup Offload As the IO thread parses the command from the client's Querybuf, it can perform a command lookup in the commands dictionary, which can consume up to ~5% of the main-thread runtime. **Implementation details** The IO thread will store the looked-up command in the client's new field `io_parsed_cmd` field. We can't use `c->cmd` for that since we use `c->cmd `to check if a command was reprocessed or not. To ensure thread safety when accessing the command dictionary, we make sure the main thread isn't changing the dictionary while IO threads are accessing it. This is accomplished by introducing a new flag called `no_incremental_rehash` for the `dictType` commands. When performing `dictResize`, we will rehash the entire dictionary in place rather than deferring the process. ### Free Offload Since the command arguments are allocated by the I/O thread, it would be beneficial if they were also freed by the same thread. If the main thread frees objects allocated by the I/O thread, two issues arise: 1. During the freeing process, the main thread needs to access the SDS pointed to by the object to get its length. 2. With Jemalloc, each thread manages thread local pool (`tcache`) of buffers for quick reallocation without accessing the arena. If the main thread constantly frees objects allocated by other threads, those threads will have to frequently access the shared arena to obtain new memory allocations **Implementation Details** When freeing the client's argv, we will send the argv array to the thread that allocated it. The thread will be identified by the client ID. When freeing an object during `dbOverwrite`, we will offload the object free as well. We will extend this to offload the free during `dbDelete` in a future PR, as its effects on defrag/memory evictions need to be studied. --------- Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-19 05:21:45 +03:00
#define AE_LOCK(eventLoop) \
if ((eventLoop)->flags & AE_PROTECT_POLL) { \
assert(pthread_mutex_lock(&(eventLoop)->poll_mutex) == 0); \
}
#define AE_UNLOCK(eventLoop) \
if ((eventLoop)->flags & AE_PROTECT_POLL) { \
assert(pthread_mutex_unlock(&(eventLoop)->poll_mutex) == 0); \
}
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
aeEventLoop *aeCreateEventLoop(int setsize) {
2009-03-22 10:30:00 +01:00
aeEventLoop *eventLoop;
int i;
2009-03-22 10:30:00 +01:00
monotonicInit(); /* just in case the calling app didn't initialize */
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
if ((eventLoop = zmalloc(sizeof(*eventLoop))) == NULL) goto err;
eventLoop->events = zmalloc(sizeof(aeFileEvent) * setsize);
eventLoop->fired = zmalloc(sizeof(aeFiredEvent) * setsize);
if (eventLoop->events == NULL || eventLoop->fired == NULL) goto err;
eventLoop->setsize = setsize;
2009-03-22 10:30:00 +01:00
eventLoop->timeEventHead = NULL;
eventLoop->timeEventNextId = 0;
eventLoop->stop = 0;
eventLoop->maxfd = -1;
2010-01-28 10:12:04 -05:00
eventLoop->beforesleep = NULL;
eventLoop->aftersleep = NULL;
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
eventLoop->custompoll = NULL;
eventLoop->flags = 0;
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
/* Initialize the eventloop mutex with PTHREAD_MUTEX_ERRORCHECK type */
pthread_mutexattr_t attr;
pthread_mutexattr_init(&attr);
pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK);
if (pthread_mutex_init(&eventLoop->poll_mutex, &attr) != 0) goto err;
if (aeApiCreate(eventLoop) == -1) goto err;
/* Events with mask == AE_NONE are not set. So let's initialize the
* vector with it. */
for (i = 0; i < setsize; i++) eventLoop->events[i].mask = AE_NONE;
2009-03-22 10:30:00 +01:00
return eventLoop;
err:
if (eventLoop) {
zfree(eventLoop->events);
zfree(eventLoop->fired);
zfree(eventLoop);
}
return NULL;
2009-03-22 10:30:00 +01:00
}
/* Return the current set size. */
int aeGetSetSize(aeEventLoop *eventLoop) {
return eventLoop->setsize;
}
/*
* Tell the event processing to change the wait timeout as soon as possible.
*
* Note: it just means you turn on/off the global AE_DONT_WAIT.
*/
void aeSetDontWait(aeEventLoop *eventLoop, int noWait) {
if (noWait)
eventLoop->flags |= AE_DONT_WAIT;
else
eventLoop->flags &= ~AE_DONT_WAIT;
}
/* Resize the maximum set size of the event loop.
* If the requested set size is smaller than the current set size, but
* there is already a file descriptor in use that is >= the requested
* set size minus one, AE_ERR is returned and the operation is not
* performed at all.
*
* Otherwise AE_OK is returned and the operation is successful. */
int aeResizeSetSize(aeEventLoop *eventLoop, int setsize) {
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
AE_LOCK(eventLoop);
int ret = AE_OK;
int i;
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
if (setsize == eventLoop->setsize) goto done;
if (eventLoop->maxfd >= setsize) goto err;
if (aeApiResize(eventLoop, setsize) == -1) goto err;
eventLoop->events = zrealloc(eventLoop->events, sizeof(aeFileEvent) * setsize);
eventLoop->fired = zrealloc(eventLoop->fired, sizeof(aeFiredEvent) * setsize);
eventLoop->setsize = setsize;
/* Make sure that if we created new slots, they are initialized with
* an AE_NONE mask. */
for (i = eventLoop->maxfd + 1; i < setsize; i++) eventLoop->events[i].mask = AE_NONE;
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
goto done;
err:
ret = AE_ERR;
done:
AE_UNLOCK(eventLoop);
return ret;
}
2009-03-22 10:30:00 +01:00
void aeDeleteEventLoop(aeEventLoop *eventLoop) {
aeApiFree(eventLoop);
zfree(eventLoop->events);
zfree(eventLoop->fired);
/* Free the time events list. */
2019-12-31 19:53:00 +08:00
aeTimeEvent *next_te, *te = eventLoop->timeEventHead;
while (te) {
next_te = te->next;
if (te->finalizerProc) te->finalizerProc(eventLoop, te->clientData);
2019-12-31 19:53:00 +08:00
zfree(te);
te = next_te;
}
2009-03-22 10:30:00 +01:00
zfree(eventLoop);
}
void aeStop(aeEventLoop *eventLoop) {
eventLoop->stop = 1;
}
int aeCreateFileEvent(aeEventLoop *eventLoop, int fd, int mask, aeFileProc *proc, void *clientData) {
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
AE_LOCK(eventLoop);
int ret = AE_ERR;
if (fd >= eventLoop->setsize) {
errno = ERANGE;
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
goto done;
}
aeFileEvent *fe = &eventLoop->events[fd];
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
if (aeApiAddEvent(eventLoop, fd, mask) == -1) goto done;
fe->mask |= mask;
if (mask & AE_READABLE) fe->rfileProc = proc;
if (mask & AE_WRITABLE) fe->wfileProc = proc;
2009-03-22 10:30:00 +01:00
fe->clientData = clientData;
if (fd > eventLoop->maxfd) eventLoop->maxfd = fd;
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
ret = AE_OK;
done:
AE_UNLOCK(eventLoop);
return ret;
2009-03-22 10:30:00 +01:00
}
void aeDeleteFileEvent(aeEventLoop *eventLoop, int fd, int mask) {
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
AE_LOCK(eventLoop);
if (fd >= eventLoop->setsize) goto done;
aeFileEvent *fe = &eventLoop->events[fd];
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
if (fe->mask == AE_NONE) goto done;
ae.c: introduce the concept of read->write barrier. AOF fsync=always, and certain Redis Cluster bus operations, require to fsync data on disk before replying with an acknowledge. In such case, in order to implement Group Commits, we want to be sure that queries that are read in a given cycle of the event loop, are never served to clients in the same event loop iteration. This way, by using the event loop "before sleep" callback, we can fsync the information just one time before returning into the event loop for the next cycle. This is much more efficient compared to calling fsync() multiple times. Unfortunately because of a bug, this was not always guaranteed: the actual way the events are installed was the sole thing that could control. Normally this problem is hard to trigger when AOF is enabled with fsync=always, because we try to flush the output buffers to the socekt directly in the beforeSleep() function of Redis. However if the output buffers are full, we actually install a write event, and in such a case, this bug could happen. This change to ae.c modifies the event loop implementation to make this concept explicit. Write events that are registered with: AE_WRITABLE|AE_BARRIER Are guaranteed to never fire after the readable event was fired for the same file descriptor. In this way we are sure that data is persisted to disk before the client performing the operation receives an acknowledged. However note that this semantics does not provide all the guarantees that one may believe are automatically provided. Take the example of the blocking list operations in Redis. With AOF and fsync=always we could have: Client A doing: BLPOP myqueue 0 Client B doing: RPUSH myqueue a b c In this scenario, Client A will get the "a" elements immediately after the Client B RPUSH will be executed, even before the operation is persisted. However when Client B will get the acknowledge, it can be sure that "b,c" are already safe on disk inside the list. What to note here is that it cannot be assumed that Client A receiving the element is a guaranteed that the operation succeeded from the point of view of Client B. This is due to the fact that the barrier exists within the same socket, and not between different sockets. However in the case above, the element "a" was not going to be persisted regardless, so it is a pretty synthetic argument.
2018-02-23 17:42:24 +01:00
/* We want to always remove AE_BARRIER if set when AE_WRITABLE
* is removed. */
if (mask & AE_WRITABLE) mask |= AE_BARRIER;
/* Only remove attached events */
mask = mask & fe->mask;
fe->mask = fe->mask & (~mask);
if (fd == eventLoop->maxfd && fe->mask == AE_NONE) {
/* Update the max fd */
int j;
for (j = eventLoop->maxfd - 1; j >= 0; j--)
if (eventLoop->events[j].mask != AE_NONE) break;
eventLoop->maxfd = j;
2009-03-22 10:30:00 +01:00
}
/* Check whether there are events to be removed.
* Note: user may remove the AE_BARRIER without
* touching the actual events. */
if (mask & (AE_READABLE | AE_WRITABLE)) {
/* Must be invoked after the eventLoop mask is modified,
* which is required by evport and epoll */
aeApiDelEvent(eventLoop, fd, mask);
}
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
done:
AE_UNLOCK(eventLoop);
2009-03-22 10:30:00 +01:00
}
Add event loop support to the module API (#10001) Modules can now register sockets/pipe to the Redis main thread event loop and do network operations asynchronously. Previously, modules had to maintain an event loop and another thread for asynchronous network operations. Also, if a module is calling API functions after doing some network operations, it had to synchronize its event loop thread's access with Redis main thread by locking the GIL, causing contention on the lock. After this commit, no synchronization is needed as module can operate in Redis main thread context. So, this commit may improve the performance for some use cases. Added three functions to the module API: * RedisModule_EventLoopAdd(int fd, int mask, RedisModuleEventLoopFunc func, void *user_data) * RedisModule_EventLoopDel(int fd, int mask) * RedisModule_EventLoopAddOneShot(RedisModuleEventLoopOneShotFunc func, void *user_data) - This function can be called from other threads to trigger callback on Redis main thread. Callback will be triggered only once. If Redis main thread is sleeping, this call will wake up the Redis main thread. Event loop callbacks are called by Redis main thread after locking the GIL. Inside callbacks, modules can operate as if they are holding the GIL. Added REDISMODULE_EVENT_EVENTLOOP event with two subevents: * REDISMODULE_SUBEVENT_EVENTLOOP_BEFORE_SLEEP * REDISMODULE_SUBEVENT_EVENTLOOP_AFTER_SLEEP These events are for modules that want to participate in the before and after sleep action. e.g It might be useful to implement batching : Read data from the network, write all to a file in one go on BEFORE_SLEEP event.
2022-01-18 14:10:07 +03:00
void *aeGetFileClientData(aeEventLoop *eventLoop, int fd) {
if (fd >= eventLoop->setsize) return NULL;
aeFileEvent *fe = &eventLoop->events[fd];
if (fe->mask == AE_NONE) return NULL;
return fe->clientData;
}
int aeGetFileEvents(aeEventLoop *eventLoop, int fd) {
if (fd >= eventLoop->setsize) return 0;
aeFileEvent *fe = &eventLoop->events[fd];
return fe->mask;
}
long long aeCreateTimeEvent(aeEventLoop *eventLoop,
long long milliseconds,
aeTimeProc *proc,
void *clientData,
aeEventFinalizerProc *finalizerProc) {
2009-03-22 10:30:00 +01:00
long long id = eventLoop->timeEventNextId++;
aeTimeEvent *te;
te = zmalloc(sizeof(*te));
if (te == NULL) return AE_ERR;
te->id = id;
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
te->when = getMonotonicUs() + milliseconds * 1000;
2009-03-22 10:30:00 +01:00
te->timeProc = proc;
te->finalizerProc = finalizerProc;
te->clientData = clientData;
Fix ae.c when a timer finalizerProc adds an event. While this feature is not used by Redis, ae.c implements the ability for a timer to call a finalizer callback when an timer event is deleted. This feature was bugged since the start, and because it was never used we never noticed a problem. However Anthony LaTorre was using the same library in order to implement a different system: he found a bug that he describes as follows, and which he fixed with the patch in this commit, sent me by private email: --- Anthony email --- 've found one bug in the current implementation of the timed events. It's possible to lose track of a timed event if an event is added in the finalizerProc of another event. For example, suppose you start off with three timed events 1, 2, and 3. Then the linked list looks like: 3 -> 2 -> 1 Then, you run processTimeEvents and events 2 and 3 finish, so now the list looks like: -1 -> -1 -> 2 Now, on the next iteration of processTimeEvents it starts by deleting the first event, and suppose this finalizerProc creates a new event, so that the list looks like this: 4 -> -1 -> 2 On the next iteration of the while loop, when it gets to the second event, the variable prev is still set to NULL, so that the head of the event loop after the next event will be set to 2, i.e. after deleting the next event the event loop will look like: 2 and the event with id 4 will be lost. I've attached an example program to illustrate the issue. If you run it you will see that it prints: ``` foo id = 0 spam! ``` But if you uncomment line 29 and run it again it won't print "spam!". --- End of email --- Test.c source code is as follows: #include "ae.h" #include <stdio.h> aeEventLoop *el; int foo(struct aeEventLoop *el, long long id, void *data) { printf("foo id = %lld\n", id); return AE_NOMORE; } int spam(struct aeEventLoop *el, long long id, void *data) { printf("spam!\n"); return AE_NOMORE; } void bar(struct aeEventLoop *el, void *data) { aeCreateTimeEvent(el, 0, spam, NULL, NULL); } int main(int argc, char **argv) { el = aeCreateEventLoop(100); //aeCreateTimeEvent(el, 0, foo, NULL, NULL); aeCreateTimeEvent(el, 0, foo, NULL, bar); aeMain(el); return 0; } Anthony fixed the problem by using a linked list for the list of timers, and sent me back this patch after he tested the code in production for some time. The code looks sane to me, so committing it to Redis.
2018-03-28 14:06:08 +02:00
te->prev = NULL;
2009-03-22 10:30:00 +01:00
te->next = eventLoop->timeEventHead;
te->refcount = 0;
if (te->next) te->next->prev = te;
2009-03-22 10:30:00 +01:00
eventLoop->timeEventHead = te;
return id;
}
int aeDeleteTimeEvent(aeEventLoop *eventLoop, long long id) {
aeTimeEvent *te = eventLoop->timeEventHead;
while (te) {
2009-03-22 10:30:00 +01:00
if (te->id == id) {
te->id = AE_DELETED_EVENT_ID;
2009-03-22 10:30:00 +01:00
return AE_OK;
}
te = te->next;
}
return AE_ERR; /* NO event with the specified ID found */
}
/* How many microseconds until the first timer should fire.
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
* If there are no timers, -1 is returned.
2009-03-22 10:30:00 +01:00
*
* Note that's O(N) since time events are unsorted.
* Possible optimizations (not needed so far, but...):
* 1) Insert the event in order, so that the nearest is just the head.
* Much better but still insertion or deletion of timers is O(N).
* 2) Use a skiplist to have this operation as O(1) and insertion as O(log(N)).
*/
static int64_t usUntilEarliestTimer(aeEventLoop *eventLoop) {
2009-03-22 10:30:00 +01:00
aeTimeEvent *te = eventLoop->timeEventHead;
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
if (te == NULL) return -1;
2009-03-22 10:30:00 +01:00
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
aeTimeEvent *earliest = NULL;
while (te) {
if ((!earliest || te->when < earliest->when) && te->id != AE_DELETED_EVENT_ID) earliest = te;
2009-03-22 10:30:00 +01:00
te = te->next;
}
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
monotime now = getMonotonicUs();
return (now >= earliest->when) ? 0 : earliest->when - now;
2009-03-22 10:30:00 +01:00
}
/* Process time events */
static int processTimeEvents(aeEventLoop *eventLoop) {
int processed = 0;
Fix ae.c when a timer finalizerProc adds an event. While this feature is not used by Redis, ae.c implements the ability for a timer to call a finalizer callback when an timer event is deleted. This feature was bugged since the start, and because it was never used we never noticed a problem. However Anthony LaTorre was using the same library in order to implement a different system: he found a bug that he describes as follows, and which he fixed with the patch in this commit, sent me by private email: --- Anthony email --- 've found one bug in the current implementation of the timed events. It's possible to lose track of a timed event if an event is added in the finalizerProc of another event. For example, suppose you start off with three timed events 1, 2, and 3. Then the linked list looks like: 3 -> 2 -> 1 Then, you run processTimeEvents and events 2 and 3 finish, so now the list looks like: -1 -> -1 -> 2 Now, on the next iteration of processTimeEvents it starts by deleting the first event, and suppose this finalizerProc creates a new event, so that the list looks like this: 4 -> -1 -> 2 On the next iteration of the while loop, when it gets to the second event, the variable prev is still set to NULL, so that the head of the event loop after the next event will be set to 2, i.e. after deleting the next event the event loop will look like: 2 and the event with id 4 will be lost. I've attached an example program to illustrate the issue. If you run it you will see that it prints: ``` foo id = 0 spam! ``` But if you uncomment line 29 and run it again it won't print "spam!". --- End of email --- Test.c source code is as follows: #include "ae.h" #include <stdio.h> aeEventLoop *el; int foo(struct aeEventLoop *el, long long id, void *data) { printf("foo id = %lld\n", id); return AE_NOMORE; } int spam(struct aeEventLoop *el, long long id, void *data) { printf("spam!\n"); return AE_NOMORE; } void bar(struct aeEventLoop *el, void *data) { aeCreateTimeEvent(el, 0, spam, NULL, NULL); } int main(int argc, char **argv) { el = aeCreateEventLoop(100); //aeCreateTimeEvent(el, 0, foo, NULL, NULL); aeCreateTimeEvent(el, 0, foo, NULL, bar); aeMain(el); return 0; } Anthony fixed the problem by using a linked list for the list of timers, and sent me back this patch after he tested the code in production for some time. The code looks sane to me, so committing it to Redis.
2018-03-28 14:06:08 +02:00
aeTimeEvent *te;
long long maxId;
te = eventLoop->timeEventHead;
maxId = eventLoop->timeEventNextId - 1;
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
monotime now = getMonotonicUs();
while (te) {
long long id;
/* Remove events scheduled for deletion. */
if (te->id == AE_DELETED_EVENT_ID) {
aeTimeEvent *next = te->next;
/* If a reference exists for this timer event,
* don't free it. This is currently incremented
* for recursive timerProc calls */
if (te->refcount) {
te = next;
continue;
}
Fix ae.c when a timer finalizerProc adds an event. While this feature is not used by Redis, ae.c implements the ability for a timer to call a finalizer callback when an timer event is deleted. This feature was bugged since the start, and because it was never used we never noticed a problem. However Anthony LaTorre was using the same library in order to implement a different system: he found a bug that he describes as follows, and which he fixed with the patch in this commit, sent me by private email: --- Anthony email --- 've found one bug in the current implementation of the timed events. It's possible to lose track of a timed event if an event is added in the finalizerProc of another event. For example, suppose you start off with three timed events 1, 2, and 3. Then the linked list looks like: 3 -> 2 -> 1 Then, you run processTimeEvents and events 2 and 3 finish, so now the list looks like: -1 -> -1 -> 2 Now, on the next iteration of processTimeEvents it starts by deleting the first event, and suppose this finalizerProc creates a new event, so that the list looks like this: 4 -> -1 -> 2 On the next iteration of the while loop, when it gets to the second event, the variable prev is still set to NULL, so that the head of the event loop after the next event will be set to 2, i.e. after deleting the next event the event loop will look like: 2 and the event with id 4 will be lost. I've attached an example program to illustrate the issue. If you run it you will see that it prints: ``` foo id = 0 spam! ``` But if you uncomment line 29 and run it again it won't print "spam!". --- End of email --- Test.c source code is as follows: #include "ae.h" #include <stdio.h> aeEventLoop *el; int foo(struct aeEventLoop *el, long long id, void *data) { printf("foo id = %lld\n", id); return AE_NOMORE; } int spam(struct aeEventLoop *el, long long id, void *data) { printf("spam!\n"); return AE_NOMORE; } void bar(struct aeEventLoop *el, void *data) { aeCreateTimeEvent(el, 0, spam, NULL, NULL); } int main(int argc, char **argv) { el = aeCreateEventLoop(100); //aeCreateTimeEvent(el, 0, foo, NULL, NULL); aeCreateTimeEvent(el, 0, foo, NULL, bar); aeMain(el); return 0; } Anthony fixed the problem by using a linked list for the list of timers, and sent me back this patch after he tested the code in production for some time. The code looks sane to me, so committing it to Redis.
2018-03-28 14:06:08 +02:00
if (te->prev)
te->prev->next = te->next;
else
Fix ae.c when a timer finalizerProc adds an event. While this feature is not used by Redis, ae.c implements the ability for a timer to call a finalizer callback when an timer event is deleted. This feature was bugged since the start, and because it was never used we never noticed a problem. However Anthony LaTorre was using the same library in order to implement a different system: he found a bug that he describes as follows, and which he fixed with the patch in this commit, sent me by private email: --- Anthony email --- 've found one bug in the current implementation of the timed events. It's possible to lose track of a timed event if an event is added in the finalizerProc of another event. For example, suppose you start off with three timed events 1, 2, and 3. Then the linked list looks like: 3 -> 2 -> 1 Then, you run processTimeEvents and events 2 and 3 finish, so now the list looks like: -1 -> -1 -> 2 Now, on the next iteration of processTimeEvents it starts by deleting the first event, and suppose this finalizerProc creates a new event, so that the list looks like this: 4 -> -1 -> 2 On the next iteration of the while loop, when it gets to the second event, the variable prev is still set to NULL, so that the head of the event loop after the next event will be set to 2, i.e. after deleting the next event the event loop will look like: 2 and the event with id 4 will be lost. I've attached an example program to illustrate the issue. If you run it you will see that it prints: ``` foo id = 0 spam! ``` But if you uncomment line 29 and run it again it won't print "spam!". --- End of email --- Test.c source code is as follows: #include "ae.h" #include <stdio.h> aeEventLoop *el; int foo(struct aeEventLoop *el, long long id, void *data) { printf("foo id = %lld\n", id); return AE_NOMORE; } int spam(struct aeEventLoop *el, long long id, void *data) { printf("spam!\n"); return AE_NOMORE; } void bar(struct aeEventLoop *el, void *data) { aeCreateTimeEvent(el, 0, spam, NULL, NULL); } int main(int argc, char **argv) { el = aeCreateEventLoop(100); //aeCreateTimeEvent(el, 0, foo, NULL, NULL); aeCreateTimeEvent(el, 0, foo, NULL, bar); aeMain(el); return 0; } Anthony fixed the problem by using a linked list for the list of timers, and sent me back this patch after he tested the code in production for some time. The code looks sane to me, so committing it to Redis.
2018-03-28 14:06:08 +02:00
eventLoop->timeEventHead = te->next;
if (te->next) te->next->prev = te->prev;
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
if (te->finalizerProc) {
te->finalizerProc(eventLoop, te->clientData);
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
now = getMonotonicUs();
}
zfree(te);
te = next;
continue;
}
/* Make sure we don't process time events created by time events in
* this iteration. Note that this check is currently useless: we always
* add new timers on the head, however if we change the implementation
* detail, this check may be useful again: we keep it here for future
* defense. */
if (te->id > maxId) {
te = te->next;
continue;
}
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
if (te->when <= now) {
int retval;
id = te->id;
te->refcount++;
retval = te->timeProc(eventLoop, id, te->clientData);
te->refcount--;
processed++;
Use H/W Monotonic clock and updates to AE (#7644) Update adds a general source for retrieving a monotonic time. In addition, AE has been updated to utilize the new monotonic clock for timer processing. This performance improvement is **not** enabled in a default build due to various H/W compatibility concerns, see README.md for details. It does however change the default use of gettimeofday with clock_gettime and somewhat improves performance. This update provides the following 1. An interface for retrieving a monotonic clock. getMonotonicUs returns a uint64_t (aka monotime) with the number of micro-seconds from an arbitrary point. No more messing with tv_sec/tv_usec. Simple routines are provided for measuring elapsed milli-seconds or elapsed micro-seconds (the most common use case for a monotonic timer). No worries about time moving backwards. 2. High-speed assembler implementation for x86 and ARM. The standard method for retrieving the monotonic clock is POSIX.1b (1993): clock_gettime(CLOCK_MONOTONIC, timespec*). However, most modern processors provide a constant speed instruction clock which can be retrieved in a fraction of the time that it takes to call clock_gettime. For x86, this is provided by the RDTSC instruction. For ARM, this is provided by the CNTVCT_EL0 instruction. As a compile-time option, these high-speed timers can be chosen. (Default is POSIX clock_gettime.) 3. Refactor of event loop timers. The timer processing in ae.c has been refactored to use the new monotonic clock interface. This results in simpler/cleaner logic and improved performance.
2020-08-28 01:54:10 -07:00
now = getMonotonicUs();
if (retval != AE_NOMORE) {
te->when = now + (monotime)retval * 1000;
} else {
te->id = AE_DELETED_EVENT_ID;
}
}
te = te->next;
}
return processed;
}
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
/* This function provides direct access to the aeApiPoll call.
* It is intended to be called from a custom poll function.*/
int aePoll(aeEventLoop *eventLoop, struct timeval *tvp) {
AE_LOCK(eventLoop);
int ret = aeApiPoll(eventLoop, tvp);
AE_UNLOCK(eventLoop);
return ret;
}
/* Process every pending file event, then every pending time event
* (that may be registered by file event callbacks just processed).
2009-03-22 10:30:00 +01:00
* Without special flags the function sleeps until some file event
2013-01-17 01:00:20 +08:00
* fires, or when the next time event occurs (if any).
2009-03-22 10:30:00 +01:00
*
* If flags is 0, the function does nothing and returns.
* if flags has AE_ALL_EVENTS set, all the kind of events are processed.
* if flags has AE_FILE_EVENTS set, file events are processed.
* if flags has AE_TIME_EVENTS set, time events are processed.
* if flags has AE_DONT_WAIT set, the function returns ASAP once all
* the events that can be handled without a wait are processed.
2019-01-06 15:01:25 +08:00
* if flags has AE_CALL_AFTER_SLEEP set, the aftersleep callback is called.
2020-05-12 13:07:44 +02:00
* if flags has AE_CALL_BEFORE_SLEEP set, the beforesleep callback is called.
2009-03-22 10:30:00 +01:00
*
* The function returns the number of events processed. */
int aeProcessEvents(aeEventLoop *eventLoop, int flags) {
int processed = 0, numevents;
2009-03-22 10:30:00 +01:00
/* Nothing to do? return ASAP */
if (!(flags & AE_TIME_EVENTS) && !(flags & AE_FILE_EVENTS)) return 0;
/* Note that we want to call aeApiPoll() even if there are no
2009-03-22 10:30:00 +01:00
* file events to process as long as we want to process time
* events, in order to sleep until the next time event is ready
* to fire. */
if (eventLoop->maxfd != -1 || ((flags & AE_TIME_EVENTS) && !(flags & AE_DONT_WAIT))) {
int j;
struct timeval tv, *tvp = NULL; /* NULL means infinite wait. */
int64_t usUntilTimer;
2009-03-22 10:30:00 +01:00
if (eventLoop->beforesleep != NULL && (flags & AE_CALL_BEFORE_SLEEP)) eventLoop->beforesleep(eventLoop);
2009-03-22 10:30:00 +01:00
Io thread work offload (#763) ### IO-Threads Work Offloading This PR is the 2nd of 3 PRs intended to achieve the goal of 1M requests per second. (1st PR: https://github.com/valkey-io/valkey/pull/758) This PR offloads additional work to the I/O threads, beyond the current read-parse/write operations, to better utilize the I/O threads and reduce the load on the main thread. It contains the following 3 commits: ### Poll Offload Currently, the main thread is responsible for executing the poll-wait system call, while the IO threads wait for tasks from the main thread. The poll-wait operation is expensive and can consume up to 30% of the main thread's time. We could have let the IO threads do the poll-wait by themselves, with each thread listening to some of the clients and notifying the main thread when a client's command is ready to execute. However, the current approach, where the main thread listens for events from the network, has several benefits. The main thread remains in charge, allowing it to know the state of each client (idle/read/write/close) at any given time. Additionally, it makes the threads flexible, enabling us to drain an IO thread's job queue and stop a thread when the load is light without modifying the event loop and moving its clients to a different IO thread. Furthermore, with this approach, the IO threads don't need to wait for both messages from the network and from the main thread instead, the threads wait only for tasks from the main thread. To enjoy the benefits of both the main thread remaining in charge and the poll being offloaded, we propose offloading the poll-wait as a single-time, non-blocking job to one of the IO threads. The IO thread will perform a poll-wait non-blocking call while the main thread processes the client commands. Later, in `aeProcessEvents`, instead of sleeping on the poll, we check for the IO thread's poll-wait results. The poll-wait will be offloaded in `beforeSleep` only when there are ready events for the main thread to process. If no events are pending, the main thread will revert to the current behavior and sleep on the poll by itself. **Implementation Details** A new call back `custompoll` was added to the `aeEventLoop` when not set to `NULL` the ae will call the `custompoll` callback instead of the `aeApiPoll`. When the poll is offloaded we will set the `custompoll` to `getIOThreadPollResults` and send a poll-job to the thread. the thread will take a mutex, call a non-blocking (with timeout 0) to `aePoll` which will populate the fired events array. the IO thread will set the `server.io_fired_events` to the number of the returning `numevents`, later the main-thread in `custompoll` will return the `server.io_fired_events` and will set the `customPoll` back to `NULL`. To ensure thread safety when accessing server.el, all functions that modify the eventloop events were wrapped with a mutex to ensure mutual exclusion when modifying the events. ### Command Lookup Offload As the IO thread parses the command from the client's Querybuf, it can perform a command lookup in the commands dictionary, which can consume up to ~5% of the main-thread runtime. **Implementation details** The IO thread will store the looked-up command in the client's new field `io_parsed_cmd` field. We can't use `c->cmd` for that since we use `c->cmd `to check if a command was reprocessed or not. To ensure thread safety when accessing the command dictionary, we make sure the main thread isn't changing the dictionary while IO threads are accessing it. This is accomplished by introducing a new flag called `no_incremental_rehash` for the `dictType` commands. When performing `dictResize`, we will rehash the entire dictionary in place rather than deferring the process. ### Free Offload Since the command arguments are allocated by the I/O thread, it would be beneficial if they were also freed by the same thread. If the main thread frees objects allocated by the I/O thread, two issues arise: 1. During the freeing process, the main thread needs to access the SDS pointed to by the object to get its length. 2. With Jemalloc, each thread manages thread local pool (`tcache`) of buffers for quick reallocation without accessing the arena. If the main thread constantly frees objects allocated by other threads, those threads will have to frequently access the shared arena to obtain new memory allocations **Implementation Details** When freeing the client's argv, we will send the argv array to the thread that allocated it. The thread will be identified by the client ID. When freeing an object during `dbOverwrite`, we will offload the object free as well. We will extend this to offload the free during `dbDelete` in a future PR, as its effects on defrag/memory evictions need to be studied. --------- Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-19 05:21:45 +03:00
if (eventLoop->custompoll != NULL) {
numevents = eventLoop->custompoll(eventLoop);
} else {
/* The eventLoop->flags may be changed inside beforesleep.
* So we should check it after beforesleep be called. At the same time,
* the parameter flags always should have the highest priority.
* That is to say, once the parameter flag is set to AE_DONT_WAIT,
* no matter what value eventLoop->flags is set to, we should ignore it. */
if ((flags & AE_DONT_WAIT) || (eventLoop->flags & AE_DONT_WAIT)) {
tv.tv_sec = tv.tv_usec = 0;
2009-03-22 10:30:00 +01:00
tvp = &tv;
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
} else if (flags & AE_TIME_EVENTS) {
usUntilTimer = usUntilEarliestTimer(eventLoop);
if (usUntilTimer >= 0) {
tv.tv_sec = usUntilTimer / 1000000;
tv.tv_usec = usUntilTimer % 1000000;
tvp = &tv;
}
2009-03-22 10:30:00 +01:00
}
Io thread work offload (#763) ### IO-Threads Work Offloading This PR is the 2nd of 3 PRs intended to achieve the goal of 1M requests per second. (1st PR: https://github.com/valkey-io/valkey/pull/758) This PR offloads additional work to the I/O threads, beyond the current read-parse/write operations, to better utilize the I/O threads and reduce the load on the main thread. It contains the following 3 commits: ### Poll Offload Currently, the main thread is responsible for executing the poll-wait system call, while the IO threads wait for tasks from the main thread. The poll-wait operation is expensive and can consume up to 30% of the main thread's time. We could have let the IO threads do the poll-wait by themselves, with each thread listening to some of the clients and notifying the main thread when a client's command is ready to execute. However, the current approach, where the main thread listens for events from the network, has several benefits. The main thread remains in charge, allowing it to know the state of each client (idle/read/write/close) at any given time. Additionally, it makes the threads flexible, enabling us to drain an IO thread's job queue and stop a thread when the load is light without modifying the event loop and moving its clients to a different IO thread. Furthermore, with this approach, the IO threads don't need to wait for both messages from the network and from the main thread instead, the threads wait only for tasks from the main thread. To enjoy the benefits of both the main thread remaining in charge and the poll being offloaded, we propose offloading the poll-wait as a single-time, non-blocking job to one of the IO threads. The IO thread will perform a poll-wait non-blocking call while the main thread processes the client commands. Later, in `aeProcessEvents`, instead of sleeping on the poll, we check for the IO thread's poll-wait results. The poll-wait will be offloaded in `beforeSleep` only when there are ready events for the main thread to process. If no events are pending, the main thread will revert to the current behavior and sleep on the poll by itself. **Implementation Details** A new call back `custompoll` was added to the `aeEventLoop` when not set to `NULL` the ae will call the `custompoll` callback instead of the `aeApiPoll`. When the poll is offloaded we will set the `custompoll` to `getIOThreadPollResults` and send a poll-job to the thread. the thread will take a mutex, call a non-blocking (with timeout 0) to `aePoll` which will populate the fired events array. the IO thread will set the `server.io_fired_events` to the number of the returning `numevents`, later the main-thread in `custompoll` will return the `server.io_fired_events` and will set the `customPoll` back to `NULL`. To ensure thread safety when accessing server.el, all functions that modify the eventloop events were wrapped with a mutex to ensure mutual exclusion when modifying the events. ### Command Lookup Offload As the IO thread parses the command from the client's Querybuf, it can perform a command lookup in the commands dictionary, which can consume up to ~5% of the main-thread runtime. **Implementation details** The IO thread will store the looked-up command in the client's new field `io_parsed_cmd` field. We can't use `c->cmd` for that since we use `c->cmd `to check if a command was reprocessed or not. To ensure thread safety when accessing the command dictionary, we make sure the main thread isn't changing the dictionary while IO threads are accessing it. This is accomplished by introducing a new flag called `no_incremental_rehash` for the `dictType` commands. When performing `dictResize`, we will rehash the entire dictionary in place rather than deferring the process. ### Free Offload Since the command arguments are allocated by the I/O thread, it would be beneficial if they were also freed by the same thread. If the main thread frees objects allocated by the I/O thread, two issues arise: 1. During the freeing process, the main thread needs to access the SDS pointed to by the object to get its length. 2. With Jemalloc, each thread manages thread local pool (`tcache`) of buffers for quick reallocation without accessing the arena. If the main thread constantly frees objects allocated by other threads, those threads will have to frequently access the shared arena to obtain new memory allocations **Implementation Details** When freeing the client's argv, we will send the argv array to the thread that allocated it. The thread will be identified by the client ID. When freeing an object during `dbOverwrite`, we will offload the object free as well. We will extend this to offload the free during `dbDelete` in a future PR, as its effects on defrag/memory evictions need to be studied. --------- Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-19 05:21:45 +03:00
/* Call the multiplexing API, will return only on timeout or when
* some event fires. */
numevents = aeApiPoll(eventLoop, tvp);
2009-03-22 10:30:00 +01:00
}
/* Don't process file events if not requested. */
if (!(flags & AE_FILE_EVENTS)) {
numevents = 0;
}
/* After sleep callback. */
Async IO threads (#758) This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests per second, as detailed by [dan touitou](https://github.com/touitou-dan) in https://github.com/valkey-io/valkey/issues/22. This PR modifies the IO threads to be fully asynchronous, which is a first and necessary step to allow more work offloading and better utilization of the IO threads. ### Current IO threads state: Valkey IO threads were introduced in Redis 6.0 to allow better utilization of multi-core machines. Before this, Redis was single-threaded and could only use one CPU core for network and command processing. The introduction of IO threads helps in offloading the IO operations to multiple threads. **Current IO Threads flow:** 1. Initialization: When Redis starts, it initializes a specified number of IO threads. These threads are in addition to the main thread, each thread starts with an empty list, the main thread will populate that list in each event-loop with pending-read-clients or pending-write-clients. 2. Read Phase: The main thread accepts incoming connections and reads requests from clients. The reading of requests are offloaded to IO threads. The main thread puts the clients ready-to-read in a list and set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick the clients up, perform the read operation and parse the first incoming command. 3. Command Processing: After reading the requests, command processing is still single-threaded and handled by the main thread. 4. Write Phase: Similar to the read phase, the write phase is also be offloaded to IO threads. The main thread prepares the response in the clients’ output buffer then the main thread puts the client in the list, and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO threads then pick the clients up and perform the write operation to send the responses back to clients. 5. Synchronization: The main-thread communicate with the threads on how many jobs left per each thread with atomic counter. The main-thread doesn’t access the clients while being handled by the IO threads. **Issues with current implementation:** * Underutilized Cores: The current implementation of IO-threads leads to the underutilization of CPU cores. * The main thread remains responsible for a significant portion of IO-related tasks that could be offloaded to IO-threads. * When the main-thread is processing client’s commands, the IO threads are idle for a considerable amount of time. * Notably, the main thread's performance during the IO-related tasks is constrained by the speed of the slowest IO-thread. * Limited Offloading: Currently, Since the Main-threads waits synchronously for the IO threads, the Threads perform only read-parse, and write operations, with parsing done only for the first command. If the threads can do work asynchronously we may offload more work to the threads reducing the load from the main-thread. * TLS: Currently, we don't support IO threads with TLS (where offloading IO would be more beneficial) since TLS read/write operations are not thread-safe with the current implementation. ### Suggested change Non-blocking main thread - The main thread and IO threads will operate in parallel to maximize efficiency. The main thread will not be blocked by IO operations. It will continue to process commands independently of the IO thread's activities. **Implementation details** **Inter-thread communication.** * We use a static, lock-free ring buffer of fixed size (2048 jobs) for the main thread to send jobs and for the IO to receive them. If the ring buffer fills up, the main thread will handle the task itself, acting as back pressure (in case IO operations are more expensive than command processing). A static ring buffer is a better candidate than a dynamic job queue as it eliminates the need for allocation/freeing per job. * An IO job will be in the format: ` [void* function-call-back | void *data] `where data is either a client to read/write from and the function-ptr is the function to be called with the data for example readQueryFromClient using this format we can use it later to offload other types of works to the IO threads. * The Ring buffer is one way from the main-thread to the IO thread, Upon read/write event the main thread will send a read/write job then in before sleep it will iterate over the pending read/write clients to checking for each client if the IO threads has already finished handling it. The IO thread signals it has finished handling a client read/write by toggling an atomic flag read_state / write_state on the client struct. **Thread Safety** As suggested in this solution, the IO threads are reading from and writing to the clients' buffers while the main thread may access those clients. We must ensure no race conditions or unsafe access occurs while keeping the Valkey code simple and lock free. Minimal Action in the IO Threads The main change is to limit the IO thread operations to the bare minimum. The IO thread will access only the client's struct and only the necessary fields in this struct. The IO threads will be responsible for the following: * Read Operation: The IO thread will only read and parse a single command. It will not update the server stats, handle read errors, or parsing errors. These tasks will be taken care of by the main thread. * Write Operation: The IO thread will only write the available data. It will not free the client's replies, handle write errors, or update the server statistics. To achieve this without code duplication, the read/write code has been refactored into smaller, independent components: * Functions that perform only the read/parse/write calls. * Functions that handle the read/parse/write results. This refactor accounts for the majority of the modifications in this PR. **Client Struct Safe Access** As we ensure that the IO threads access memory only within the client struct, we need to ensure thread safety only for the client's struct's shared fields. * Query Buffer * Command parsing - The main thread will not try to parse a command from the query buffer when a client is offloaded to the IO thread. * Client's memory checks in client-cron - The main thread will not access the client query buffer if it is offloaded and will handle the querybuf grow/shrink when the client is back. * CLIENT LIST command - The main thread will busy-wait for the IO thread to finish handling the client, falling back to the current behavior where the main thread waits for the IO thread to finish their processing. * Output Buffer * The IO thread will not change the client's bufpos and won't free the client's reply lists. These actions will be done by the main thread on the client's return from the IO thread. * bufpos / block→used: As the main thread may change the bufpos, the reply-block→used, or add/delete blocks to the reply list while the IO thread writes, we add two fields to the client struct: io_last_bufpos and io_last_reply_block. The IO thread will write until the io_last_bufpos, which was set by the main-thread before sending the client to the IO thread. If more data has been added to the cob in between, it will be written in the next write-job. In addition, the main thread will not trim or merge reply blocks while the client is offloaded. * Parsing Fields * Client's cmd, argc, argv, reqtype, etc., are set during parsing. * The main thread will indicate to the IO thread not to parse a cmd if the client is not reset. In this case, the IO thread will only read from the network and won't attempt to parse a new command. * The main thread won't access the c→cmd/c→argv in the CLIENT LIST command as stated before it will busy wait for the IO threads. * Client Flags * c→flags, which may be changed by the main thread in multiple places, won't be accessed by the IO thread. Instead, the main thread will set the c→io_flags with the information necessary for the IO thread to know the client's state. * Client Close * On freeClient, the main thread will busy wait for the IO thread to finish processing the client's read/write before proceeding to free the client. * Client's Memory Limits * The IO thread won't handle the qb/cob limits. In case a client crosses the qb limit, the IO thread will stop reading for it, letting the main thread know that the client crossed the limit. **TLS** TLS is currently not supported with IO threads for the following reasons: 1. Pending reads - If SSL has pending data that has already been read from the socket, there is a risk of not calling the read handler again. To handle this, a list is used to hold the pending clients. With IO threads, multiple threads can access the list concurrently. 2. Event loop modification - Currently, the TLS code registers/unregisters the file descriptor from the event loop depending on the read/write results. With IO threads, multiple threads can modify the event loop struct simultaneously. 3. The same client can be sent to 2 different threads concurrently (https://github.com/redis/redis/issues/12540). Those issues were handled in the current PR: 1. The IO thread only performs the read operation. The main thread will check for pending reads after the client returns from the IO thread and will be the only one to access the pending list. 2. The registering/unregistering of events will be similarly postponed and handled by the main thread only. 3. Each client is being sent to the same dedicated thread (c→id % num_of_threads). **Sending Replies Immediately with IO threads.** Currently, after processing a command, we add the client to the pending_writes_list. Only after processing all the clients do we send all the replies. Since the IO threads are now working asynchronously, we can send the reply immediately after processing the client’s requests, reducing the command latency. However, if we are using AOF=always, we must wait for the AOF buffer to be written, in which case we revert to the current behavior. **IO threads dynamic adjustment** Currently, we use an all-or-nothing approach when activating the IO threads. The current logic is as follows: if the number of pending write clients is greater than twice the number of threads (including the main thread), we enable all threads; otherwise, we enable none. For example, if 8 IO threads are defined, we enable all 8 threads if there are 16 pending clients; else, we enable none. It makes more sense to enable partial activation of the IO threads. If we have 10 pending clients, we will enable 5 threads, and so on. This approach allows for a more granular and efficient allocation of resources based on the current workload. In addition, the user will now be able to change the number of I/O threads at runtime. For example, when decreasing the number of threads from 4 to 2, threads 3 and 4 will be closed after flushing their job queues. **Tests** Currently, we run the io-threads tests with 4 IO threads (https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353). This means that we will not activate the IO threads unless there are 8 (threads * 2) pending write clients per single loop, which is unlikely to happened in most of tests, meaning the IO threads are not currently being tested. To enforce the main thread to always offload work to the IO threads, regardless of the number of pending events, we add an events-per-io-thread configuration with a default value of 2. When set to 0, this configuration will force the main thread to always offload work to the IO threads. When we offload every single read/write operation to the IO threads, the IO-threads are running with 100% CPU when running multiple tests concurrently some tests fail as a result of larger than expected command latencies. To address this issue, we have to add some after or wait_for calls to some of the tests to ensure they pass with IO threads as well. Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
if (eventLoop->aftersleep != NULL && flags & AE_CALL_AFTER_SLEEP) eventLoop->aftersleep(eventLoop, numevents);
for (j = 0; j < numevents; j++) {
int fd = eventLoop->fired[j].fd;
aeFileEvent *fe = &eventLoop->events[fd];
int mask = eventLoop->fired[j].mask;
int fired = 0; /* Number of events fired for current fd. */
/* Normally we execute the readable event first, and the writable
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
* event later. This is useful as sometimes we may be able
* to serve the reply of a query immediately after processing the
* query.
*
* However if AE_BARRIER is set in the mask, our application is
* asking us to do the reverse: never fire the writable event
* after the readable. In such a case, we invert the calls.
* This is useful when, for instance, we want to do things
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
* in the beforeSleep() hook, like fsyncing a file to disk,
* before replying to a client. */
int invert = fe->mask & AE_BARRIER;
2018-07-04 20:04:06 +08:00
/* Note the "fe->mask & mask & ..." code: maybe an already
* processed event removed an element that fired and we still
* didn't processed, so we check if the event is still valid.
*
* Fire the readable event if the call sequence is not
* inverted. */
if (!invert && fe->mask & mask & AE_READABLE) {
fe->rfileProc(eventLoop, fd, fe->clientData, mask);
fired++;
fe = &eventLoop->events[fd]; /* Refresh in case of resize. */
}
/* Fire the writable event. */
if (fe->mask & mask & AE_WRITABLE) {
if (!fired || fe->wfileProc != fe->rfileProc) {
fe->wfileProc(eventLoop, fd, fe->clientData, mask);
fired++;
}
}
/* If we have to invert the call, fire the readable event now
* after the writable one. */
if (invert) {
fe = &eventLoop->events[fd]; /* Refresh in case of resize. */
if ((fe->mask & mask & AE_READABLE) && (!fired || fe->wfileProc != fe->rfileProc)) {
fe->rfileProc(eventLoop, fd, fe->clientData, mask);
fired++;
ae.c: introduce the concept of read->write barrier. AOF fsync=always, and certain Redis Cluster bus operations, require to fsync data on disk before replying with an acknowledge. In such case, in order to implement Group Commits, we want to be sure that queries that are read in a given cycle of the event loop, are never served to clients in the same event loop iteration. This way, by using the event loop "before sleep" callback, we can fsync the information just one time before returning into the event loop for the next cycle. This is much more efficient compared to calling fsync() multiple times. Unfortunately because of a bug, this was not always guaranteed: the actual way the events are installed was the sole thing that could control. Normally this problem is hard to trigger when AOF is enabled with fsync=always, because we try to flush the output buffers to the socekt directly in the beforeSleep() function of Redis. However if the output buffers are full, we actually install a write event, and in such a case, this bug could happen. This change to ae.c modifies the event loop implementation to make this concept explicit. Write events that are registered with: AE_WRITABLE|AE_BARRIER Are guaranteed to never fire after the readable event was fired for the same file descriptor. In this way we are sure that data is persisted to disk before the client performing the operation receives an acknowledged. However note that this semantics does not provide all the guarantees that one may believe are automatically provided. Take the example of the blocking list operations in Redis. With AOF and fsync=always we could have: Client A doing: BLPOP myqueue 0 Client B doing: RPUSH myqueue a b c In this scenario, Client A will get the "a" elements immediately after the Client B RPUSH will be executed, even before the operation is persisted. However when Client B will get the acknowledge, it can be sure that "b,c" are already safe on disk inside the list. What to note here is that it cannot be assumed that Client A receiving the element is a guaranteed that the operation succeeded from the point of view of Client B. This is due to the fact that the barrier exists within the same socket, and not between different sockets. However in the case above, the element "a" was not going to be persisted regardless, so it is a pretty synthetic argument.
2018-02-23 17:42:24 +01:00
}
}
processed++;
2009-03-22 10:30:00 +01:00
}
}
/* Check time events */
if (flags & AE_TIME_EVENTS) processed += processTimeEvents(eventLoop);
2009-03-22 10:30:00 +01:00
return processed; /* return the number of processed file/time events */
}
2013-01-17 01:00:20 +08:00
/* Wait for milliseconds until the given file descriptor becomes
2009-03-22 10:30:00 +01:00
* writable/readable/exception */
int aeWait(int fd, int mask, long long milliseconds) {
struct pollfd pfd;
2009-03-22 10:30:00 +01:00
int retmask = 0, retval;
memset(&pfd, 0, sizeof(pfd));
pfd.fd = fd;
if (mask & AE_READABLE) pfd.events |= POLLIN;
if (mask & AE_WRITABLE) pfd.events |= POLLOUT;
if ((retval = poll(&pfd, 1, milliseconds)) == 1) {
if (pfd.revents & POLLIN) retmask |= AE_READABLE;
if (pfd.revents & POLLOUT) retmask |= AE_WRITABLE;
2018-07-04 20:04:06 +08:00
if (pfd.revents & POLLERR) retmask |= AE_WRITABLE;
if (pfd.revents & POLLHUP) retmask |= AE_WRITABLE;
2009-03-22 10:30:00 +01:00
return retmask;
} else {
return retval;
}
}
void aeMain(aeEventLoop *eventLoop) {
2009-03-22 10:30:00 +01:00
eventLoop->stop = 0;
2010-01-28 10:12:04 -05:00
while (!eventLoop->stop) {
aeProcessEvents(eventLoop, AE_ALL_EVENTS | AE_CALL_BEFORE_SLEEP | AE_CALL_AFTER_SLEEP);
2010-01-28 10:12:04 -05:00
}
2009-03-22 10:30:00 +01:00
}
char *aeGetApiName(void) {
return aeApiName();
}
2010-01-28 10:12:04 -05:00
void aeSetBeforeSleepProc(aeEventLoop *eventLoop, aeBeforeSleepProc *beforesleep) {
eventLoop->beforesleep = beforesleep;
}
Async IO threads (#758) This PR is 1 of 3 PRs intended to achieve the goal of 1 million requests per second, as detailed by [dan touitou](https://github.com/touitou-dan) in https://github.com/valkey-io/valkey/issues/22. This PR modifies the IO threads to be fully asynchronous, which is a first and necessary step to allow more work offloading and better utilization of the IO threads. ### Current IO threads state: Valkey IO threads were introduced in Redis 6.0 to allow better utilization of multi-core machines. Before this, Redis was single-threaded and could only use one CPU core for network and command processing. The introduction of IO threads helps in offloading the IO operations to multiple threads. **Current IO Threads flow:** 1. Initialization: When Redis starts, it initializes a specified number of IO threads. These threads are in addition to the main thread, each thread starts with an empty list, the main thread will populate that list in each event-loop with pending-read-clients or pending-write-clients. 2. Read Phase: The main thread accepts incoming connections and reads requests from clients. The reading of requests are offloaded to IO threads. The main thread puts the clients ready-to-read in a list and set the global io_threads_op to IO_THREADS_OP_READ, the IO threads pick the clients up, perform the read operation and parse the first incoming command. 3. Command Processing: After reading the requests, command processing is still single-threaded and handled by the main thread. 4. Write Phase: Similar to the read phase, the write phase is also be offloaded to IO threads. The main thread prepares the response in the clients’ output buffer then the main thread puts the client in the list, and sets the global io_threads_op to the IO_THREADS_OP_WRITE. The IO threads then pick the clients up and perform the write operation to send the responses back to clients. 5. Synchronization: The main-thread communicate with the threads on how many jobs left per each thread with atomic counter. The main-thread doesn’t access the clients while being handled by the IO threads. **Issues with current implementation:** * Underutilized Cores: The current implementation of IO-threads leads to the underutilization of CPU cores. * The main thread remains responsible for a significant portion of IO-related tasks that could be offloaded to IO-threads. * When the main-thread is processing client’s commands, the IO threads are idle for a considerable amount of time. * Notably, the main thread's performance during the IO-related tasks is constrained by the speed of the slowest IO-thread. * Limited Offloading: Currently, Since the Main-threads waits synchronously for the IO threads, the Threads perform only read-parse, and write operations, with parsing done only for the first command. If the threads can do work asynchronously we may offload more work to the threads reducing the load from the main-thread. * TLS: Currently, we don't support IO threads with TLS (where offloading IO would be more beneficial) since TLS read/write operations are not thread-safe with the current implementation. ### Suggested change Non-blocking main thread - The main thread and IO threads will operate in parallel to maximize efficiency. The main thread will not be blocked by IO operations. It will continue to process commands independently of the IO thread's activities. **Implementation details** **Inter-thread communication.** * We use a static, lock-free ring buffer of fixed size (2048 jobs) for the main thread to send jobs and for the IO to receive them. If the ring buffer fills up, the main thread will handle the task itself, acting as back pressure (in case IO operations are more expensive than command processing). A static ring buffer is a better candidate than a dynamic job queue as it eliminates the need for allocation/freeing per job. * An IO job will be in the format: ` [void* function-call-back | void *data] `where data is either a client to read/write from and the function-ptr is the function to be called with the data for example readQueryFromClient using this format we can use it later to offload other types of works to the IO threads. * The Ring buffer is one way from the main-thread to the IO thread, Upon read/write event the main thread will send a read/write job then in before sleep it will iterate over the pending read/write clients to checking for each client if the IO threads has already finished handling it. The IO thread signals it has finished handling a client read/write by toggling an atomic flag read_state / write_state on the client struct. **Thread Safety** As suggested in this solution, the IO threads are reading from and writing to the clients' buffers while the main thread may access those clients. We must ensure no race conditions or unsafe access occurs while keeping the Valkey code simple and lock free. Minimal Action in the IO Threads The main change is to limit the IO thread operations to the bare minimum. The IO thread will access only the client's struct and only the necessary fields in this struct. The IO threads will be responsible for the following: * Read Operation: The IO thread will only read and parse a single command. It will not update the server stats, handle read errors, or parsing errors. These tasks will be taken care of by the main thread. * Write Operation: The IO thread will only write the available data. It will not free the client's replies, handle write errors, or update the server statistics. To achieve this without code duplication, the read/write code has been refactored into smaller, independent components: * Functions that perform only the read/parse/write calls. * Functions that handle the read/parse/write results. This refactor accounts for the majority of the modifications in this PR. **Client Struct Safe Access** As we ensure that the IO threads access memory only within the client struct, we need to ensure thread safety only for the client's struct's shared fields. * Query Buffer * Command parsing - The main thread will not try to parse a command from the query buffer when a client is offloaded to the IO thread. * Client's memory checks in client-cron - The main thread will not access the client query buffer if it is offloaded and will handle the querybuf grow/shrink when the client is back. * CLIENT LIST command - The main thread will busy-wait for the IO thread to finish handling the client, falling back to the current behavior where the main thread waits for the IO thread to finish their processing. * Output Buffer * The IO thread will not change the client's bufpos and won't free the client's reply lists. These actions will be done by the main thread on the client's return from the IO thread. * bufpos / block→used: As the main thread may change the bufpos, the reply-block→used, or add/delete blocks to the reply list while the IO thread writes, we add two fields to the client struct: io_last_bufpos and io_last_reply_block. The IO thread will write until the io_last_bufpos, which was set by the main-thread before sending the client to the IO thread. If more data has been added to the cob in between, it will be written in the next write-job. In addition, the main thread will not trim or merge reply blocks while the client is offloaded. * Parsing Fields * Client's cmd, argc, argv, reqtype, etc., are set during parsing. * The main thread will indicate to the IO thread not to parse a cmd if the client is not reset. In this case, the IO thread will only read from the network and won't attempt to parse a new command. * The main thread won't access the c→cmd/c→argv in the CLIENT LIST command as stated before it will busy wait for the IO threads. * Client Flags * c→flags, which may be changed by the main thread in multiple places, won't be accessed by the IO thread. Instead, the main thread will set the c→io_flags with the information necessary for the IO thread to know the client's state. * Client Close * On freeClient, the main thread will busy wait for the IO thread to finish processing the client's read/write before proceeding to free the client. * Client's Memory Limits * The IO thread won't handle the qb/cob limits. In case a client crosses the qb limit, the IO thread will stop reading for it, letting the main thread know that the client crossed the limit. **TLS** TLS is currently not supported with IO threads for the following reasons: 1. Pending reads - If SSL has pending data that has already been read from the socket, there is a risk of not calling the read handler again. To handle this, a list is used to hold the pending clients. With IO threads, multiple threads can access the list concurrently. 2. Event loop modification - Currently, the TLS code registers/unregisters the file descriptor from the event loop depending on the read/write results. With IO threads, multiple threads can modify the event loop struct simultaneously. 3. The same client can be sent to 2 different threads concurrently (https://github.com/redis/redis/issues/12540). Those issues were handled in the current PR: 1. The IO thread only performs the read operation. The main thread will check for pending reads after the client returns from the IO thread and will be the only one to access the pending list. 2. The registering/unregistering of events will be similarly postponed and handled by the main thread only. 3. Each client is being sent to the same dedicated thread (c→id % num_of_threads). **Sending Replies Immediately with IO threads.** Currently, after processing a command, we add the client to the pending_writes_list. Only after processing all the clients do we send all the replies. Since the IO threads are now working asynchronously, we can send the reply immediately after processing the client’s requests, reducing the command latency. However, if we are using AOF=always, we must wait for the AOF buffer to be written, in which case we revert to the current behavior. **IO threads dynamic adjustment** Currently, we use an all-or-nothing approach when activating the IO threads. The current logic is as follows: if the number of pending write clients is greater than twice the number of threads (including the main thread), we enable all threads; otherwise, we enable none. For example, if 8 IO threads are defined, we enable all 8 threads if there are 16 pending clients; else, we enable none. It makes more sense to enable partial activation of the IO threads. If we have 10 pending clients, we will enable 5 threads, and so on. This approach allows for a more granular and efficient allocation of resources based on the current workload. In addition, the user will now be able to change the number of I/O threads at runtime. For example, when decreasing the number of threads from 4 to 2, threads 3 and 4 will be closed after flushing their job queues. **Tests** Currently, we run the io-threads tests with 4 IO threads (https://github.com/valkey-io/valkey/blob/443d80f1686377ad42cbf92d98ecc6d240325ee1/.github/workflows/daily.yml#L353). This means that we will not activate the IO threads unless there are 8 (threads * 2) pending write clients per single loop, which is unlikely to happened in most of tests, meaning the IO threads are not currently being tested. To enforce the main thread to always offload work to the IO threads, regardless of the number of pending events, we add an events-per-io-thread configuration with a default value of 2. When set to 0, this configuration will force the main thread to always offload work to the IO threads. When we offload every single read/write operation to the IO threads, the IO-threads are running with 100% CPU when running multiple tests concurrently some tests fail as a result of larger than expected command latencies. To address this issue, we have to add some after or wait_for calls to some of the tests to ensure they pass with IO threads as well. Signed-off-by: Uri Yagelnik <uriy@amazon.com>
2024-07-09 06:01:39 +03:00
void aeSetAfterSleepProc(aeEventLoop *eventLoop, aeAfterSleepProc *aftersleep) {
eventLoop->aftersleep = aftersleep;
}
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
/* This function allows setting a custom poll procedure to be used by the event loop.
* The custom poll procedure, if set, will be called instead of the default aeApiPoll */
void aeSetCustomPollProc(aeEventLoop *eventLoop, aeCustomPollProc *custompoll) {
eventLoop->custompoll = custompoll;
}
void aeSetPollProtect(aeEventLoop *eventLoop, int protect) {
if (protect) {
eventLoop->flags |= AE_PROTECT_POLL;
} else {
eventLoop->flags &= ~AE_PROTECT_POLL;
}
}