## Background
1. Currently Lua memory control does not pass through Redis's zmalloc.c.
Redis maxmemory cannot limit memory problems caused by users abusing lua
since these lua VM memory is not part of used_memory.
2. Since jemalloc is much better (fragmentation and speed), and also we
know it and trust it. we are
going to use jemalloc instead of libc to allocate the Lua VM code and
count it used memory.
## Process:
In this PR, we will use jemalloc in lua.
1. Create an arena for all lua vm (script and function), which is
shared, in order to avoid blocking defragger.
2. Create a bound tcache for the lua VM, since the lua VM and the main
thread are by default in the same tcache, and if there is no isolated
tcache, lua may request memory from the tcache which has just been freed
by main thread, and vice versa
On the other hand, since lua vm might be release in bio thread, but
tcache is not thread-safe, we need to recreate
the tcache every time we recreate the lua vm.
3. Remove lua memory statistics from memory fragmentation statistics to
avoid the effects of lua memory fragmentation
## Other
Add the following new fields to `INFO DEBUG` (we may promote them to
INFO MEMORY some day)
1. allocator_allocated_lua: total number of bytes allocated of lua arena
2. allocator_active_lua: total number of bytes in active pages allocated
in lua arena
3. allocator_resident_lua: maximum number of bytes in physically
resident data pages mapped in lua arena
4. allocator_frag_bytes_lua: fragment bytes in lua arena
This is oranagra's idea, and i got some help from sundb.
This solves the third point in #13102.
---------
Co-authored-by: debing.sun <debing.sun@redis.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
# Overview
Users utilize the `FLUSHDB SYNC` and `FLUSHALL SYNC` commands for a variety of
reasons. The main issue with this command is that if the database becomes
substantial in size, the server will be unresponsive for an extended period.
Other than freezing application traffic, this may also lead some clients making
incorrect judgments about the server's availability. For instance, a watchdog may
erroneously decide to terminate the process, resulting in potential adverse
outcomes. While a `FLUSH* ASYNC` can address these issues, it might not be used
for two reasons: firstly, it's not the default, and secondly, in some cases, the
client issuing the flush wants to wait for its completion before repopulating the
database.
Between the option of triggering FLUSH* asynchronously in the background without
indication for completion versus running it synchronously in the foreground by
the main thread, there is another more appealing option. We can block the
client that requested the flush, execute the flush command in the background, and
once done, unblock the client and return notification for completion. This approach
ensures the server remains responsive to other clients, and the blocked client
receives the expected response only after the flush operation has been successfully
carried out.
# Implementation details
Instead of defining yet another flavor to the flush command, we can modify
`FLUSHALL SYNC` and `FLUSHDB SYNC` always run in this new mode.
## Extending BIO Threads capabilities
Today jobs that are carried out by BIO threads don't have the capability to
indicate completion to the main thread. We can add this infrastructure by having
an additional dummy job, coined as completion-job, that eventually will be written
by BIO threads to a response-queue. The main thread will take care to consume items
from the response-queue and call the provided callback function of each
completion-job.
## FLUSH* SYNC to run as blocking ASYNC
Command `FLUSH* SYNC` will be modified to create one or more async jobs to flush
DB(s) and afterward will push additional completion-job request. By sending the
completion job request only at the end, the main thread will be called back only
after all the preceding jobs completed their task in the background. During that
time, the client of the command is suspended and marked as `BLOCKED_LAZYFREE`
whereas any other client will be able to communicate with the server without any
issue.
Users who abuse lua error_reply will generate a new error object on each
error call, which can make server.errors get bigger and bigger. This
will
cause the server to block when calling INFO (we also return errorstats
by
default).
To prevent the damage it can cause, when a misuse is detected, we will
print a warning log and disable the errorstats to avoid adding more new
errors. It can be re-enabled via CONFIG RESETSTAT.
Because server.errors may be very large (it may be better now since we
have the limit), config resetstat may block for a while. So in
resetErrorTableStats, we will try to lazyfree server.errors.
See the related discussion at the end of #8217.
In some cases, users will abuse lua eval. Each EVAL call generates
a new lua script, which is added to the lua interpreter and cached
to redis-server, consuming a large amount of memory over time.
Since EVAL is mostly the one that abuses the lua cache, and these
won't have pipeline issues (i.e. the script won't disappear
unexpectedly,
and cause errors like it would with SCRIPT LOAD and EVALSHA),
we implement a plain FIFO LRU eviction only for these (not for
scripts loaded with SCRIPT LOAD).
### Implementation notes:
When not abused we'll probably have less than 100 scripts, and when
abused we'll have many thousands. So we use a hard coded value of 500
scripts. And considering that we don't have many scripts, then unlike
keys, we don't need to worry about the memory usage of keeping a true
sorted LRU linked list. We compute the SHA of each script anyway,
and put the script in a dict, we can store a listNode there, and use
it for quick removal and re-insertion into an LRU list each time the
script is used.
### New interfaces:
At the same time, a new `evicted_scripts` field is added to
INFO, which represents the number of evicted eval scripts. Users
can check it to see if they are abusing EVAL.
### benchmark:
`./src/redis-benchmark -P 10 -n 1000000 -r 10000000000 eval "return
__rand_int__" 0`
The simple abuse of eval benchmark test that will create 1 million EVAL
scripts. The performance has been improved by 50%, and the max latency
has dropped from 500ms to 13ms (this may be caused by table expansion
inside Lua when the number of scripts is large). And in the INFO memory,
it used to consume 120MB (server cache) + 310MB (lua engine), but now
it only consumes 70KB (server cache) + 210KB (lua_engine) because of
the scripts eviction.
For non-abusive case of about 100 EVAL scripts, there's no noticeable
change in performance or memory usage.
### unlikely potentially breaking change:
in theory, a user can maybe load a
script with EVAL and then use EVALSHA to call it (by calculating the
SHA1 value on the client side), it could be that if we read the docs
carefully we'll realized it's a valid scenario, but we suppose it's
extremely rare. So it may happen that EVALSHA acts on a script created
by EVAL, and the script is evicted and EVALSHA returns a NOSCRIPT error.
that is if you have more than 500 scripts being used in the same
transaction / pipeline.
This solves the second point in #13102.
After #13013
### This PR make effort to defrag the pubsub kvstore in the following
ways:
1. Till now server.pubsub(shard)_channels only share channel name obj
with the first subscribed client, now change it so that the clients and
the pubsub kvstore share the channel name robj.
This would save a lot of memory when there are many subscribers to the
same channel.
It also means that we only need to defrag the channel name robj in the
pubsub kvstore, and then update
all client references for the current channel, avoiding the need to
iterate through all the clients to do the same things.
2. Refactor the code to defragment pubsub(shard) in the same way as
defragment of keys and EXPIRES, with the exception that we only
defragment pubsub(without shard) when slot is zero.
### Other
Fix an overlook in #11695, if defragment doesn't reach the end time, we
should wait for the current
db's keys and expires, pubsub and pubsubshard to finish before leaving,
now it's possible to exit
early when the keys are defragmented.
---------
Co-authored-by: oranagra <oran@redislabs.com>
Sometimes we need to make fast judgement about why Redis is suddenly
taking more memory. One of the reasons is main DB's dicts doing
rehashing.
We may use `MEMORY STATS` to monitor the overhead memory of each DB, but
there still lacks a total sum to show an overall trend. So this PR adds
the total overhead of all DBs to `INFO MEMORY` section, together with
the total count of rehashing DB dicts, providing some intuitive metrics
about main dicts rehashing.
This PR adds the following metrics to INFO MEMORY
* `mem_overhead_db_hashtable_rehashing` - only size of ht[0] in
dictionaries we're rehashing (i.e. the memory that's gonna get released
soon)
and a similar ones to MEMORY STATS:
* `overhead.db.hashtable.lut` (complements the existing
`overhead.hashtable.main` and `overhead.hashtable.expires` which also
counts the `dictEntry` structs too)
* `overhead.db.hashtable.rehashing` - temporary rehashing overhead.
* `db.dict.rehashing.count` - number of top level dictionaries being
rehashed.
---------
Co-authored-by: zhaozhao.zz <zhaozhao.zz@alibaba-inc.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Even if we have SCRIPT FLUSH ASYNC now, when there are a lot of
lua scripts, SCRIPT FLUSH ASYNC will still block the main thread.
This is because lua_close is executed in the main thread, and lua
heap needs to release a lot of memory.
In this PR, we take the current lua instance on lctx.lua and call
lua_close on it in a background thread, to close it in async way.
This is MeirShpilraien's idea.
Implement #12963
## Changes
1. large bins don't have external fragmentation or are at least
non-defraggable, so we should ignore the effect of
large bins when measuring fragmentation, and only measure fragmentation
of small bins. this affects both the allocator_frag* metrics and also
the active-defrag trigger
2. Adding INFO metrics for `muzzy` memory, which is memory returned to
the OS but still shows as RSS until the OS reclaims it.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
Currently aof_last_fsync is using a low resolution unixtime is really
bad,
it checks if the absolute number of (full) seconds changed by one.
depending on which side of the second barrier it falls, we can get very
different results.
This PR change the resolution to use milliseconds instead of complete
seconds.
In cases where the event loop cycle duration is short and their rapid
(e.g. running
many fast commands with short pipeline, or a high `hz` config), this
change will not
make much difference, since in anyway, we'll be quick to detect that
we're on a "new
second", and it's likely that these fsync will always be executed close
to the second
switch barrier.
But in cases of rare or slow event loops cycles (e.g. either slow
commands, or very
low rate of traffic to redis, and low `hz`), it could easily be that
with the old code,
in some cases we'll have over 1.5 seconds between fsyncs, and in others
less than 0.5.
see discussion in #8612
This PR also handle aof_flush_postponed_start as well, the damage there
is smaller
since the threshold is 2 seconds, and not 1.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
Redis has some special commands that mark the client's state, such as
`subscribe` and `blpop`, which mark the client as `CLIENT_PUBSUB` or
`CLIENT_BLOCKED`, and we have metrics for the special use cases.
However, there are also other special commands, like `WATCH`, which
although do not have a specific flags, and should also be considered
stateful client types. For stateful clients, in many scenarios, the
connections cannot be shared in "connection pool", meaning connection
pool cannot be used. For example, whenever the `WATCH` command is
executed, a new connection is required to put the client into the "watch
state" because the watched keys are stored in the client.
If different business logic requires watching different keys, separate
connections must be used; otherwise, there will be contamination. This
also means that if a user's business heavily relies on the `WATCH`
command, a large number of connections will be required.
Recently we have encountered this situation in our platform, where some
users consume a significant number of connections when using Redis
because of `WATCH`.
I hope we can have a way to observe these special use cases and special
client connections. Here I add a few monitoring metrics:
1. `watching_clients` in `INFO` reply: The number of clients currently
in the "watching" state.
2. `total_watched_keys` in `INFO` reply: The total number of keys being
watched.
3. `watch` in `CLIENT LIST` reply: The number of keys each client is
currently watching.
The test fails here and there:
```
*** [err]: expire scan should skip dictionaries with lot's of empty buckets in tests/unit/expire.tcl
scan didn't handle slot skipping logic.
```
There are two case:
1. In the case of passing the test, we use child process to avoid the
dict resize, but it can not completely limit it, since in the dictDelete
we still have chance to trigger the resize (hit the force radio). The
reason why our test passed before is because the expire dict is still
in the rehashing process, so the dictDelete, the dictShrinkIfNeeded can
not trigger the resize.
2. In the case of failing the test, the expire dict finished the
rehashing,
so the last dictDelete, the dictShrinkIfNeeded trigger the dict resize
since it hit the force radio, so the skipping logic fail.
This PR add a new DEBUG command to disbale the dict resize.
We forgot to call quicklistSetOptions after createQuicklistObject,
in the sort store scenario, we will create a quicklist with default
fill or compress options.
This PR adds fill and depth parameters to createQuicklistObject to
specify that options need to be set after creating a quicklist.
This closes#12871.
release notes:
> Fix lists created by SORT STORE to respect list compression and
packing configs.
Currently, once active defrag starts, we can not adjust
active_defrag_running
downwards. This is because active_defrag_running will be dynamically
compute
based on the fragmentation, we think we should not lower the effort when
the
fragmentation drops.
However, we need to note that active_defrag_running will also be
dynamically
computed based on configurations. In this case, we are not respecting
cycle-min
or cycle-max. Some people may realize halfway through that defrag
consumes a
lot and want to adjust it.
Previously we could only turn off activedefrag and then turn it on again
to
adjust active_defrag_running downwards. So in this PR, when a active
defrag
configuration change is made, we will re-compute it.
These configuration items are:
- active-defrag-cycle-min
- active-defrag-cycle-max
- active-defrag-threshold-upper
# Description
Gather most of the scattered `redisDb`-related code from the per-slot
dict PR (#11695) and turn it to a new data structure, `kvstore`. i.e.
it's a class that represents an array of dictionaries.
# Motivation
The main motivation is code cleanliness, the idea of using an array of
dictionaries is very well-suited to becoming a self-contained data
structure.
This allowed cleaning some ugly code, among others: loops that run twice
on the main dict and expires dict, and duplicate code for allocating and
releasing this data structure.
# Notes
1. This PR reverts the part of https://github.com/redis/redis/pull/12848
where the `rehashing` list is global (handling rehashing `dict`s is
under the responsibility of `kvstore`, and should not be managed by the
server)
2. This PR also replaces the type of `server.pubsubshard_channels` from
`dict**` to `kvstore` (original PR:
https://github.com/redis/redis/pull/12804). After that was done,
server.pubsub_channels was also chosen to be a `kvstore` (with only one
`dict`, which seems odd) just to make the code cleaner by making it the
same type as `server.pubsubshard_channels`, see
`pubsubtype.serverPubSubChannels`
3. the keys and expires kvstores are currenlty configured to allocate
the individual dicts only when the first key is added (unlike before, in
which they allocated them in advance), but they won't release them when
the last key is deleted.
Worth mentioning that due to the recent change the reply of DEBUG
HTSTATS changed, in case no keys were ever added to the db.
before:
```
127.0.0.1:6379> DEBUG htstats 9
[Dictionary HT]
Hash table 0 stats (main hash table):
No stats available for empty dictionaries
[Expires HT]
Hash table 0 stats (main hash table):
No stats available for empty dictionaries
```
after:
```
127.0.0.1:6379> DEBUG htstats 9
[Dictionary HT]
[Expires HT]
```
When we use a timer to unblock a client in module, if the timer
period and the block timeout are very close, they will unblock the
client in the same event loop, and it will trigger the assertion.
The reason is that in moduleBlockedClientTimedOut we will protect
against re-processing, so we don't actually call updateStatsOnUnblock
(see #12817), so we are not able to reset the c->duration.
The reason is unblockClientOnTimeout() didn't realize that bc had
been unblocked. We add a function to the module to determine if bc
is blocked, and then use it in unblockClientOnTimeout() to exit.
There is the stack:
```
beforeSleep
blockedBeforeSleep
handleBlockedClientsTimeout
checkBlockedClientTimeout
unblockClientOnTimeout
unblockClient
resetClient
-- assertion, crash the server
'c->duration == 0' is not true
```
In #11012, we will reprocess command when client is unblocked on keys,
in some blocking commands, for example, in the XREADGROUP BLOCK
scenario,
because of the re-processing command, we will recalculate the block
timeout,
causing the blocking time to be reset.
This commit add a new CLIENT_REPROCESSING_COMMAND clent flag, explicitly
let the command know that it is being re-processed, later in
blockForKeys
we will not reset the timeout.
Affected BLOCK cases:
- list / zset / stream, added test cases for each.
Unaffected cases:
- module (never re-process the commands).
- WAIT / WAITAOF (never re-process the commands).
Fixes#12998.
The function `tryResizeHashTables` only attempts to shrink the dicts
that has keys (change from #11695), this was a serious problem until the
change in #12850 since it meant if all keys are deleted, we won't shrink
the dick.
But still, both dictShrink and dictExpand may be blocked by a fork child
process, therefore, the cron job needs to perform both dictShrink and
dictExpand, for not just non-empty dicts, but all dicts in DBs.
What this PR does:
1. Try to resize all dicts in DBs (not just non-empty ones, as it was
since #12850)
2. handle both shrink and expand (not just shrink, as it was since
forever)
3. Refactor some APIs about dict resizing (get rid of `htNeedsShrink`
`htNeedsShrink` `dictShrinkToFit`, and expose `dictShrinkIfNeeded`
`dictExpandIfNeeded` which already contains all the code of those
functions we get rid of, to make APIs more neat)
4. In the `Don't rehash if redis has child process` test, now that cron
would do resizing, we no longer need to write to DB after the child
process got killed, and can wait for the cron to expand the hash table.
Fix#12785 and other race condition issues.
See the following isolated comments.
The following report was obtained using SANITIZER thread.
```sh
make SANITIZER=thread
./runtest-moduleapi --config io-threads 4 --config io-threads-do-reads yes --accurate
```
1. Fixed thread-safe issue in RM_UnblockClient()
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1831181220
* When blocking a client in a module using `RM_BlockClientOnKeys()` or
`RM_BlockClientOnKeysWithFlags()`
with a timeout_callback, calling RM_UnblockClient() in module threads
can lead to race conditions
in `updateStatsOnUnblock()`.
- Introduced:
Version: 6.2
PR: #7491
- Touch:
`server.stat_numcommands`, `cmd->latency_histogram`, `server.slowlog`,
and `server.latency_events`
- Harm Level: High
Potentially corrupts the memory data of `cmd->latency_histogram`,
`server.slowlog`, and `server.latency_events`
- Solution:
Differentiate whether the call to moduleBlockedClientTimedOut() comes
from the module or the main thread.
Since we can't know if RM_UnblockClient() comes from module threads, we
always assume it does and
let `updateStatsOnUnblock()` asynchronously update the unblock status.
* When error reply is called in timeout_callback(), ctx is not
thread-safe, eventually lead to race conditions in `afterErrorReply`.
- Introduced:
Version: 6.2
PR: #8217
- Touch
`server.stat_total_error_replies`, `server.errors`,
- Harm Level: High
Potentially corrupts the memory data of `server.errors`
- Solution:
Make the ctx in `timeout_callback()` with `REDISMODULE_CTX_THREAD_SAFE`,
and asynchronously reply errors to the client.
2. Made RM_Reply*() family API thread-safe
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408707239
Call chain: `RM_Reply*()` -> `_addReplyToBufferOrList()` -> touch
server.current_client
- Introduced:
Version: 7.2.0
PR: #12326
- Harm Level: None
Since the module fake client won't have the `CLIENT_PUSHING` flag, even
if we touch server.current_client,
we can still exit after `c->flags & CLIENT_PUSHING`.
- Solution
Checking `c->flags & CLIENT_PUSHING` earlier.
3. Made freeClient() thread-safe
Fix#12785
- Introduced:
Version: 4.0
Commit:
3fcf959e60
- Harm Level: Moderate
* Trigger assertion
It happens when the module thread calls freeClient while the io-thread
is in progress,
which just triggers an assertion, and doesn't make any race condiaions.
* Touch `server.current_client`, `server.stat_clients_type_memory`, and
`clientMemUsageBucket->clients`.
It happens between the main thread and the module threads, may cause
data corruption.
1. Error reset `server.current_client` to NULL, but theoretically this
won't happen,
because the module has already reset `server.current_client` to old
value before entering freeClient.
2. corrupts `clientMemUsageBucket->clients` in
updateClientMemUsageAndBucket().
3. Causes server.stat_clients_type_memory memory statistics to be
inaccurate.
- Solution:
* No longer counts memory usage on fake clients, to avoid updating
`server.stat_clients_type_memory` in freeClient.
* No longer resetting `server.current_client` in unlinkClient, because
the fake client won't be evicted or disconnected in the mid of the
process.
* Judgment assertion `io_threads_op == IO_THREADS_OP_IDLE` only if c is
not a fake client.
4. Fixed free client args without GIL
Related discussion:
https://github.com/redis/redis/pull/12817#discussion_r1408706695
When freeing retained strings in the module thread (refcount decr), or
using them in some way (refcount incr), we should do so while holding
the GIL,
otherwise, they might be simultaneously freed while the main thread is
processing the unblock client state.
- Introduced:
Version: 6.2.0
PR: #8141
- Harm Level: Low
Trigger assertion or double free or memory leak.
- Solution:
Documenting that module API users need to ensure any access to these
retained strings is done with the GIL locked
5. Fix adding fake client to server.clients_pending_write
It will incorrectly log the memory usage for the fake client.
Related discussion:
https://github.com/redis/redis/pull/12817#issuecomment-1851899163
- Introduced:
Version: 4.0
Commit:
9b01b64430
- Harm Level: None
Only result in NOP
- Solution:
* Don't add fake client into server.clients_pending_write
* Add c->conn assertion for updateClientMemUsageAndBucket() and
updateClientMemoryUsage() to avoid same
issue in the future.
So now it will be the responsibility of the caller of both of them to
avoid passing in fake client.
6. Fix calling RM_BlockedClientMeasureTimeStart() and
RM_BlockedClientMeasureTimeEnd() without GIL
- Introduced:
Version: 6.2
PR: #7491
- Harm Level: Low
Causes inaccuracies in command latency histogram and slow logs, but does
not corrupt memory.
- Solution:
Module API users, if know that non-thread-safe APIs will be used in
multi-threading, need to take responsibility for protecting them with
their own locks instead of the GIL, as using the GIL is too expensive.
### Other issue
1. RM_Yield is not thread-safe, fixed via #12905.
### Summarize
1. Fix thread-safe issues for `RM_UnblockClient()`, `freeClient()` and
`RM_Yield`, potentially preventing memory corruption, data disorder, or
assertion.
2. Updated docs and module test to clarify module API users'
responsibility for locking non-thread-safe APIs in multi-threading, such
as RM_BlockedClientMeasureTimeStart/End(), RM_FreeString(),
RM_RetainString(), and RM_HoldString().
### About backpot to 7.2
1. The implement of (1) is not too satisfying, would like to get more
eyes.
2. (2), (3) can be safely for backport
3. (4), (6) just modifying the module tests and updating the
documentation, no need for a backpot.
4. (5) is harmless, no need for a backpot.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
When we insert entries into dict, it may autonomously expand if needed.
However, when we delete entries from dict, it doesn't shrink to the
proper size. If there are few entries in a very large dict, it may cause
huge waste of memory and inefficiency when iterating.
The main keyspace dicts (keys and expires), are shrinked by cron
(`tryResizeHashTables` calls `htNeedsResize` and `dictResize`),
And some data structures such as zset and hash also do that (call
`htNeedsResize`) right after a loop of calls to `dictDelete`,
But many other dicts are completely missing that call (they can only
expand).
In this PR, we provide the ability to automatically shrink the dict when
deleting. The conditions triggering the shrinking is the same as
`htNeedsResize` used to have. i.e. we expand when we're over 100%
utilization, and shrink when we're below 10% utilization.
Additionally:
* Add `dictPauseAutoResize` so that flows that do mass deletions, will
only trigger shrinkage at the end.
* Rename `dictResize` to `dictShrinkToFit` (same logic as it used to
have, but better name describing it)
* Rename `_dictExpand` to `_dictResize` (same logic as it used to have,
but better name describing it)
related to discussion
https://github.com/redis/redis/pull/12819#discussion_r1409293878
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: zhaozhao.zz <zhaozhao.zz@alibaba-inc.com>
I'm testing the performance of Pub/Sub command recently. I find if many
clients unsubscribe or are killed simultaneously, Redis needs a long
time to deal with it.
In my experiment, I set 5000 clients and each client subscribes 100
channels. Then I call `client kill type pubsub` to simulate the
situation where clients unsubscribe all channels at the same time and
calculate the execution time. The result shows that it takes about 23s.
I use the _perf_ and find that `listSearchKey` in
`pubsubUnsubscribeChannel` costs more than 90% cpu time. I think we can
optimize this situation.
In this PR, I replace list with dict to track the clients subscribing
the channel more efficiently. It changes O(N) to O(1) in the search
phase. Then I repeat the experiment as above. The results are as
follows.
| | Execution Time(s) |used_memory(MB) |
| :---------------- | :------: | :----: |
| unstable(1bd0b54) | 23.734 | 65.41 |
| optimize-pubsub | 0.288 | 67.66 |
Thanks for #11595 , I use a no-value dict and the results shows that the
performance improves significantly but the memory usage only increases
slightly.
Notice:
- This PR will cause the performance degradation about 20% in
`[p|s]subscribe` command but won't freeze Redis.
## Issues and solutions from #12817
1. Touch ProcessingEventsWhileBlocked and calling moduleCount() without
GIL in afterSleep()
- Introduced:
Version: 7.0.0
PR: #9963
- Harm Level: Very High
If the module thread calls `RM_Yield()` before the main thread enters
afterSleep(),
and modifies `ProcessingEventsWhileBlocked`(+1), it will cause the main
thread to not wait for GIL,
which can lead to all kinds of unforeseen problems, including memory
data corruption.
- Initial / Abandoned Solution:
* Added `__thread` specifier for ProcessingEventsWhileBlocked.
`ProcessingEventsWhileBlocked` is used to protect against nested event
processing, but event processing
in the main thread and module threads should be completely independent
and unaffected, so it is safer
to use TLS.
* Adding a cached module count to keep track of the current number of
modules, to avoid having to use `dictSize()`.
- Related Warnings:
```
WARNING: ThreadSanitizer: data race (pid=1136)
Write of size 4 at 0x0001045990c0 by thread T4 (mutexes: write M0):
#0 processEventsWhileBlocked networking.c:4135 (redis-server:arm64+0x10006d124)
#1 RM_Yield module.c:2410 (redis-server:arm64+0x10018b66c)
#2 bg_call_worker <null>:83232836 (blockedclient.so:arm64+0x16a8)
Previous read of size 4 at 0x0001045990c0 by main thread:
#0 afterSleep server.c:1861 (redis-server:arm64+0x100024f98)
#1 aeProcessEvents ae.c:408 (redis-server:arm64+0x10000fd64)
#2 aeMain ae.c:496 (redis-server:arm64+0x100010f0c)
#3 main server.c:7220 (redis-server:arm64+0x10003f38c)
```
2. aeApiPoll() is not thread-safe
When using RM_Yield to handle events in a module thread, if the main
thread has not yet
entered `afterSleep()`, both the module thread and the main thread may
touch `server.el` at the same time.
- Introduced:
Version: 7.0.0
PR: #9963
- Old / Abandoned Solution:
Adding a new mutex to protect timing between after beforeSleep() and
before afterSleep().
Defect: If the main thread enters the ae loop without any IO events, it
will wait until
the next timeout or until there is any event again, and the module
thread will
always hang until the main thread leaves the event loop.
- Related Warnings:
```
SUMMARY: ThreadSanitizer: data race ae_kqueue.c:55 in addEventMask
==================
==================
WARNING: ThreadSanitizer: data race (pid=14682)
Write of size 4 at 0x000100b54000 by thread T9 (mutexes: write M0):
#0 aeApiPoll ae_kqueue.c:175 (redis-server:arm64+0x100010588)
#1 aeProcessEvents ae.c:399 (redis-server:arm64+0x10000fb84)
#2 processEventsWhileBlocked networking.c:4138 (redis-server:arm64+0x10006d3c4)
#3 RM_Yield module.c:2410 (redis-server:arm64+0x10018b66c)
#4 bg_call_worker <null>:16042052 (blockedclient.so:arm64+0x169c)
Previous write of size 4 at 0x000100b54000 by main thread:
#0 aeApiPoll ae_kqueue.c:175 (redis-server:arm64+0x100010588)
#1 aeProcessEvents ae.c:399 (redis-server:arm64+0x10000fb84)
#2 aeMain ae.c:496 (redis-server:arm64+0x100010da8)
#3 main server.c:7238 (redis-server:arm64+0x10003f51c)
```
## The final fix as the comments:
https://github.com/redis/redis/pull/12817#discussion_r1436427232
Optimized solution based on the above comment:
First, we add `module_gil_acquring` to indicate whether the main thread
is currently in the acquiring GIL state.
When the module thread starts to yield, there are two possibilities(we
assume the caller keeps the GIL):
1. The main thread is in the mid of beforeSleep() and afterSleep(), that
is, `module_gil_acquring` is not 1 now.
At this point, the module thread will wake up the main thread through
the pipe and leave the yield,
waiting for the next yield when the main thread may already in the
acquiring GIL state.
2. The main thread is in the acquiring GIL state.
The module thread release the GIL, yielding CPU to give the main thread
an opportunity to start
event processing, and then acquire the GIL again until the main thread
releases it.
This is what
https://github.com/redis/redis/pull/12817#discussion_r1436427232
mentioned direction.
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
There are situations (especially in TLS) in which the engine gets too occupied managing a large number of new connections. Existing connections may time-out while the server is processing the new connections initial TLS handshakes, which may cause cause new connections to be established, perpetuating the problem. To better manage the tradeoff between new connection rate and other workloads, this change adds a new config to manage maximum number of new connections per event loop cycle, instead of using a predetermined number (currently 1000).
This change introduces two new configurations, max-new-connections-per-cycle and max-new-tls-connections-per-cycle. The default value of the tcp connections is 10 per cycle and the default value of tls connections per cycle is 1.
---------
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
We have achieved replacing `slots_to_keys` radix tree with key->slot
linked list (#9356), and then replacing the list with slot specific
dictionaries for keys (#11695).
Shard channels behave just like keys in many ways, and we also need a
slots->channels mapping. Currently this is still done by using a radix
tree. So we should split `server.pubsubshard_channels` into 16384 dicts
and drop the radix tree, just like what we did to DBs.
Some benefits (basically the benefits of what we've done to DBs):
1. Optimize counting channels in a slot. This is currently used only in
removing channels in a slot. But this is potentially more useful:
sometimes we need to know how many channels there are in a specific slot
when doing slot migration. Counting is now implemented by traversing the
radix tree, and with this PR it will be as simple as calling `dictSize`,
from O(n) to O(1).
2. The radix tree in the cluster has been removed. The shard channel
names no longer require additional storage, which can save memory.
3. Potentially useful in slot migration, as shard channels are logically
split by slots, thus making it easier to migrate, remove or add as a
whole.
4. Avoid rehashing a big dict when there is a large number of channels.
Drawbacks:
1. Takes more memory than using radix tree when there are relatively few
shard channels.
What this PR does:
1. in cluster mode, split `server.pubsubshard_channels` into 16384
dicts, in standalone mode, still use only one dict.
2. drop the `slots_to_channels` radix tree.
3. to save memory (to solve the drawback above), all 16384 dicts are
created lazily, which means only when a channel is about to be inserted
to the dict will the dict be initialized, and when all channels are
deleted, the dict would delete itself.
5. use `server.shard_channel_count` to keep track of the number of all
shard channels.
---------
Co-authored-by: Viktor Söderqvist <viktor.soderqvist@est.tech>
Let us see which version of redis this tool is part of.
Similarly to redis-cli, redis-benchmark and redis-check-rdb
redis-rdb-check and redis-aof-check are actually symlinks to redis,
so they will directly use getVersion in server, the format became:
```
{title} v={redis_version} sha={sha}:{dirty} malloc={malloc} bits={bits} build={build}
```
Move cliVersion into cli_common, redis-cli and redis-benchmark will
use it, and the format is not change:
```
{title} {redis_version} (git:{sha})
```
After #11695, we added two functions `rehashingStarted` and
`rehashingCompleted` to the dict structure. We also registered two
handlers for the main database's dict and expire structures. This allows
the main database to record the dict in `rehashing` list when rehashing
starts. Later, in `serverCron`, the `incrementallyRehash` function is
continuously called to perform the rehashing operation. However,
currently, when rehashing is completed, `rehashingCompleted` does not
remove the dict from the `rehashing` list. This results in the
`rehashing` list containing many invalid dicts. Although subsequent cron
checks and removes dicts that don't require rehashing, it is still
inefficient.
This PR implements the functionality to remove the dict from the
`rehashing` list in `rehashingCompleted`. This is achieved by adding
`metadata` to the dict structure, which keeps track of its position in
the `rehashing` list, allowing for quick removal. This approach avoids
storing duplicate dicts in the `rehashing` list.
Additionally, there are other modifications:
1. Whether in standalone or cluster mode, the dict in database is
inserted into the rehashing linked list when rehashing starts. This
eliminates the need to distinguish between standalone and cluster mode
in `incrementallyRehash`. The function only needs to focus on the dicts
in the `rehashing` list that require rehashing.
2. `rehashing` list is moved from per-database to Redis server level.
This decouples `incrementallyRehash` from the database ID, and in
standalone mode, there is no need to iterate over all databases,
avoiding unnecessary access to databases that do not require rehashing.
In the future, even if unsharded-cluster mode supports multiple
databases, there will be no risk involved.
3. The insertion and removal operations of dict structures in the
`rehashing` list are decoupled from `activerehashing` config.
`activerehashing` only controls whether `incrementallyRehash` is
executed in serverCron. There is no need for additional steps when
modifying the `activerehashing` switch, as in #12705.
In INFO CLIENTS section, we already have blocked_clients and
tracking_clients. We should add a new metric showing the number of
pubsub connections, which helps performance monitoring and trouble
shooting.
The change in dbSwapDatabases seems harmless. Because in non-clustered
mode, dbBuckets calculations are strictly accurate and in cluster mode,
we only have one DB. Modify it for uniformity (just like resize_cursor).
The change in swapMainDbWithTempDb is needed in case we swap with the
temp db, otherwise the overhead memory usage of db can be miscalculated.
In addition we will swap all fields (including rehashing list), just for
completeness (and reduce the chance of surprises in the future).
Introduced in #12697.
Additional optimizations for the eviction logic in #11695:
To make the eviction logic clearer and decouple the number of sampled
keys from the running mode (cluster or standalone).
* When sampling in each database, we only care about the number of keys
in the current database (not the dicts we sampled from).
* If there are a insufficient number of keys in the current database
(e.g. 10 times the value of `maxmemory_samples`), we can break out
sooner (to avoid looping on a sparse database).
* We'll never try to sample the db dicts more times than the number of
non-empty dicts in the db (max 1 in non-cluster mode).
And it also ensures that each database has a sufficient amount of
sampled keys, so even if unsharded-cluster supports multiple databases,
there won't be any issues.
other changes:
1. keep track of the number of non-empty dicts in each database.
2. move key_count tracking into cumulativeKeyCountAdd rather than all
it's callers
---------
Co-authored-by: Oran Agra <oran@redislabs.com>
Introduced in #11695 .
The tryResizeHashTables function gets stuck on the last non-empty slot
while iterating through dictionaries. It does not restart from the
beginning. The reason for this issue is a problem with the usage of
dbIteratorNextDict:
/* Returns next dictionary from the iterator, or NULL if iteration is complete. */
dict *dbIteratorNextDict(dbIterator *dbit) {
if (dbit->next_slot == -1) return NULL;
dbit->slot = dbit->next_slot;
dbit->next_slot = dbGetNextNonEmptySlot(dbit->db, dbit->slot, dbit->keyType);
return dbGetDictFromIterator(dbit);
}
When iterating to the last non-empty slot, next_slot is set to -1,
causing it to loop indefinitely on that slot. We need to modify the code
to ensure that after iterating to the last non-empty slot, it returns to
the first non-empty slot.
BTW, function tryResizeHashTables is actually iterating over slots
that have keys. However, in its implementation, it leverages the
dbIterator (which is a key iterator) to obtain slot and dictionary
information. While this approach works fine, but it is not very
intuitive. This PR also improves readability by changing the iteration
to directly iterate over slots, thereby enhancing clarity.
see discussion from after https://github.com/redis/redis/pull/12453 was
merged
----
This PR replaces signals that are not considered async-signal-safe
(AS-safe) with safe calls.
#### **1. serverLog() and serverLogFromHandler()**
`serverLog` uses unsafe calls. It was decided that we will **avoid**
`serverLog` calls by the signal handlers when:
* The signal is not fatal, such as SIGALRM. In these cases, we prefer
using `serverLogFromHandler` which is the safe version of `serverLog`.
Note they have different prompts:
`serverLog`: `62220:M 26 Oct 2023 14:39:04.526 # <msg>`
`serverLogFromHandler`: `62220:signal-handler (1698331136) <msg>`
* The code was added recently. Calls to `serverLog` by the signal
handler have been there ever since Redis exists and it hasn't caused
problems so far. To avoid regression, from now we should use
`serverLogFromHandler`
#### **2. `snprintf` `fgets` and `strtoul`(base = 16) -------->
`_safe_snprintf`, `fgets_async_signal_safe`, `string_to_hex`**
The safe version of `snprintf` was taken from
[here](8cfc4ca5e7/src/mc_util.c (L754))
#### **3. fopen(), fgets(), fclose() --------> open(), read(), close()**
#### **4. opendir(), readdir(), closedir() --------> open(),
syscall(SYS_getdents64), close()**
#### **5. Threads_mngr sync mechanisms**
* waiting for the thread to generate stack trace: semaphore -------->
busy-wait
* `globals_rw_lock` was removed: as we are not using malloc and the
semaphore anymore we don't need to protect `ThreadsManager_cleanups`.
#### **6. Stacktraces buffer**
The initial problem was that we were not able to safely call malloc
within the signal handler.
To solve that we created a buffer on the stack of `writeStacktraces` and
saved it in a global pointer, assuming that under normal circumstances,
the function `writeStacktraces` would complete before any thread
attempted to write to it. However, **if threads lag behind, they might
access this global pointer after it no longer belongs to the
`writeStacktraces` stack, potentially corrupting memory.**
To address this, various solutions were discussed
[here](https://github.com/redis/redis/pull/12658#discussion_r1390442896)
Eventually, we decided to **create a pipe** at server startup that will
remain valid as long as the process is alive.
We chose this solution due to its minimal memory usage, and since
`write()` and `read()` are atomic operations. It ensures that stack
traces from different threads won't mix.
**The stacktraces collection process is now as follows:**
* Cleaning the pipe to eliminate writes of late threads from previous
runs.
* Each thread writes to the pipe its stacktrace
* Waiting for all the threads to mark completion or until a timeout (2
sec) is reached
* Reading from the pipe to print the stacktraces.
#### **7. Changes that were considered and eventually were dropped**
* replace watchdog timer with a POSIX timer:
according to [settimer man](https://linux.die.net/man/2/setitimer)
> POSIX.1-2008 marks getitimer() and setitimer() obsolete, recommending
the use of the POSIX timers API
([timer_gettime](https://linux.die.net/man/2/timer_gettime)(2),
[timer_settime](https://linux.die.net/man/2/timer_settime)(2), etc.)
instead.
However, although it is supposed to conform to POSIX std, POSIX timers
API is not supported on Mac.
You can take a look here at the Linux implementation:
[here](c7562ee135)
To avoid messing up the code, and uncertainty regarding compatibility,
it was decided to drop it for now.
* avoid using sds (uses malloc) in logConfigDebugInfo
It was considered to print config info instead of using sds, however
apparently, `logConfigDebugInfo` does more than just print the sds, so
it was decided this fix is out of this issue scope.
#### **8. fix Signal mask check**
The check `signum & sig_mask` intended to indicate whether the signal is
blocked by the thread was incorrect. Actually, the bit position in the
signal mask corresponds to the signal number. We fixed this by changing
the condition to: `sig_mask & (1L << (sig_num - 1))`
#### **9. Unrelated changes**
both `fork.tcl `and `util.tcl` implemented a function called
`count_log_message` expecting different parameters. This caused
confusion when trying to run daily tests with additional test parameters
to run a specific test.
The `count_log_message` in `fork.tcl` was removed and the calls were
replaced with calls to `count_log_message` located in `util.tcl`
---------
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
Move clusterNode into cluster_legacy.h.
In order to achieve this some accessor methods
were added and also a refactor of how debugCommand
handles cluster related subcommands.
Signed-off-by: Josh Hershberg <yehoshua@redis.com>
Move clusterState into cluster_legacy.h. In order to achieve
this some "accessor" methods needed to be added to the
cluster API and some other minor refactors.
Signed-off-by: Josh Hershberg <yehoshua@redis.com>
When using DB iterator, it will use dictInitSafeIterator to init a old safe
dict iterator. When dbIteratorNext is used, it will jump to the next slot db
dict when we are done a dict. During this process, we do not have any calls to
dictResumeRehashing, which causes the dict's pauserehash to always be > 0.
And at last, it will be returned directly in dictRehashMilliseconds, which causes
us to have slot dict in a state where rehash cannot be completed.
In the "expire scan should skip dictionaries with lot's of empty buckets" test,
adding a `keys *` can reproduce the problem stably. `keys *` will call dbIteratorNext
to trigger a traversal of all slot dicts.
Added dbReleaseIterator and dbIteratorInitNextSafeIterator methods to call dictResetIterator.
Issue was introduced in #11695.
This PR introduces a new macro, serverAssertWithInfoDebug, to do complex assertions only for debugging. The main intention is to allow running complex operations during tests without impacting runtime performance. This assertion is enabled when setting DEBUG_ASSERTIONS.
The DEBUG_ASSERTIONS flag is set for the daily and CI variants of `test-sanitizer-address`.
Optimize the performance of SCAN commands when a match pattern can only contain keys from a
single slot in cluster mode. This can happen when the pattern contains a hash tag before any
wildcard matchers or when the key contains no matchers.
In #12476 server.stat_client_qbuf_limit_disconnections was added. It is written in readQueryFromClient, which may be called by multiple threads when io-threads and io-threads-do-reads are turned on. Somehow we missed to make it an atomic variable.
As part of #11695 independent dictionaries were introduced per slot.
Time complexity to discover total no. of buckets across all dictionaries
increased to O(N) with straightforward implementation of iterating over
all dictionaries and adding dictBuckets of each.
To optimize the time complexity, we could maintain a global counter at
db level to keep track of the count of buckets and update it on the start
and end of rehashing.
---------
Co-authored-by: Roshan Khatri <rvkhatri@amazon.com>
Dictionary iterator logic in the `tryResizeHashTables` method is picking the next
(incorrect) dictionary while the cursor is at a given slot. This could lead to some
dictionary/slot getting skipped from resizing.
Also stabilize the test.
problem introduced recently in #11695
This is an implementation of https://github.com/redis/redis/issues/10589 that eliminates 16 bytes per entry in cluster mode, that are currently used to create a linked list between entries in the same slot. Main idea is splitting main dictionary into 16k smaller dictionaries (one per slot), so we can perform all slot specific operations, such as iteration, without any additional info in the `dictEntry`. For Redis cluster, the expectation is that there will be a larger number of keys, so the fixed overhead of 16k dictionaries will be The expire dictionary is also split up so that each slot is logically decoupled, so that in subsequent revisions we will be able to atomically flush a slot of data.
## Important changes
* Incremental rehashing - one big change here is that it's not one, but rather up to 16k dictionaries that can be rehashing at the same time, in order to keep track of them, we introduce a separate queue for dictionaries that are rehashing. Also instead of rehashing a single dictionary, cron job will now try to rehash as many as it can in 1ms.
* getRandomKey - now needs to not only select a random key, from the random bucket, but also needs to select a random dictionary. Fairness is a major concern here, as it's possible that keys can be unevenly distributed across the slots. In order to address this search we introduced binary index tree). With that data structure we are able to efficiently find a random slot using binary search in O(log^2(slot count)) time.
* Iteration efficiency - when iterating dictionary with a lot of empty slots, we want to skip them efficiently. We can do this using same binary index that is used for random key selection, this index allows us to find a slot for a specific key index. For example if there are 10 keys in the slot 0, then we can quickly find a slot that contains 11th key using binary search on top of the binary index tree.
* scan API - in order to perform a scan across the entire DB, the cursor now needs to not only save position within the dictionary but also the slot id. In this change we append slot id into LSB of the cursor so it can be passed around between client and the server. This has interesting side effect, now you'll be able to start scanning specific slot by simply providing slot id as a cursor value. The plan is to not document this as defined behavior, however. It's also worth nothing the SCAN API is now technically incompatible with previous versions, although practically we don't believe it's an issue.
* Checksum calculation optimizations - During command execution, we know that all of the keys are from the same slot (outside of a few notable exceptions such as cross slot scripts and modules). We don't want to compute the checksum multiple multiple times, hence we are relying on cached slot id in the client during the command executions. All operations that access random keys, either should pass in the known slot or recompute the slot.
* Slot info in RDB - in order to resize individual dictionaries correctly, while loading RDB, it's not enough to know total number of keys (of course we could approximate number of keys per slot, but it won't be precise). To address this issue, we've added additional metadata into RDB that contains number of keys in each slot, which can be used as a hint during loading.
* DB size - besides `DBSIZE` API, we need to know size of the DB in many places want, in order to avoid scanning all dictionaries and summing up their sizes in a loop, we've introduced a new field into `redisDb` that keeps track of `key_count`. This way we can keep DBSIZE operation O(1). This is also kept for O(1) expires computation as well.
## Performance
This change improves SET performance in cluster mode by ~5%, most of the gains come from us not having to maintain linked lists for keys in slot, non-cluster mode has same performance. For workloads that rely on evictions, the performance is similar because of the extra overhead for finding keys to evict.
RDB loading performance is slightly reduced, as the slot of each key needs to be computed during the load.
## Interface changes
* Removed `overhead.hashtable.slot-to-keys` to `MEMORY STATS`
* Scan API will now require 64 bits to store the cursor, even on 32 bit systems, as the slot information will be stored.
* New RDB version to support the new op code for SLOT information.
---------
Co-authored-by: Vitaly Arbuzov <arvit@amazon.com>
Co-authored-by: Harkrishn Patro <harkrisp@amazon.com>
Co-authored-by: Roshan Khatri <rvkhatri@amazon.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
We use the C standard assert() in various places in the codebase, which requires NDEBUG to be undefined. We introduced the redisassert.h file in order to allow low level files to access the assert that maps to serverPanic, but this was only applied tactically and is not available broadly.
This PR removes all usage of the standard library asserts and replaces them with an assert that maps to serverPanic. It makes us immune to accidentally setting the NDEBUG flag preventing assertions. I also marked marked the server asserts as "likely" to not execute. I spot checked various points in the code, and it didn't change the code layout on my x86 mac, but it is more consistent with redisassert.h and seems more correct overall.
In this PR we are adding the functionality to collect all the process's threads' backtraces.
## Changes made in this PR
### **introduce threads mngr API**
The **threads mngr API** which has 2 abilities:
* `ThreadsManager_init() `- register to SIGUSR2. called on the server start-up.
* ` ThreadsManager_runOnThreads()` - receives a list of a pid_t and a callback, tells every
thread in the list to invoke the callback, and returns the output collected by each invocation.
**Elaborating atomicvar API**
* `atomicIncrGet(var,newvalue_var,count) `-- Increment and get the atomic counter new value
* `atomicFlagGetSet` -- Get and set the atomic counter value to 1
### **Always set SIGALRM handler**
SIGALRM handler prints the process's stacktrace to the log file. Up until now, it was set only if the
`server.watchdog_period` > 0. This can be also useful if debugging is needed. However, in situations
where the server can't get requests, (a deadlock, for example) we weren't able to change the signal handler.
To make it available at run time we set SIGALRM handler on server startup. The signal handler name was
changed to a more general `sigalrmSignalHandler`.
### **Print all the process' threads' stacktraces**
`logStackTrace()` now calls `writeStacktraces()`, instead of logging the current thread stacktrace.
`writeStacktraces()`:
* On Linux systems we use the threads manager API to collect the backtraces of all the process' threads.
To get the `tids` list (threads ids) we read the `/proc/<redis-server-pid>/tasks` file which includes a list of directories.
Each directory name corresponds to one tid (including the main thread). For each thread, we also need to check if it
can get the signal from the threads manager (meaning it is not blocking/ignoring that signal). We send the threads
manager this tids list and `collect_stacktrace_data()` callback, which collects the thread's backtrace addresses,
its name, and tid.
* On other systems, the behavior remained as it was (writing only the current thread stacktrace to the log file).
## compatibility notes
1. **The threads mngr API is only supported in linux.**
2. glibc earlier than 2.3 We use `syscall(SYS_gettid)` and `syscall(SYS_tgkill...)` because their dedicated
alternatives (`gettid()` and `tgkill`) were added in glibc 2.3.
## Output example
Each thread backtrace will have the following format:
`<tid> <thread_name> [additional_info]`
* **tid**: as read from the `/proc/<redis-server-pid>/tasks` file
* **thread_name**: the tread name as it is registered in the os/
* **additional_info**: Sometimes we want to add specific information about one of the threads. currently.
it is only used to mark the thread that handles the backtraces collection by adding "*".
In case of crash - this also indicates which thread caused the crash. The handling thread in won't
necessarily appear first.
```
------ STACK TRACE ------
EIP:
/lib/aarch64-linux-gnu/libc.so.6(epoll_pwait+0x9c)[0xffffb9295ebc]
67089 redis-server *
linux-vdso.so.1(__kernel_rt_sigreturn+0x0)[0xffffb9437790]
/lib/aarch64-linux-gnu/libc.so.6(epoll_pwait+0x9c)[0xffffb9295ebc]
redis-server *:6379(+0x75e0c)[0xaaaac2fe5e0c]
redis-server *:6379(aeProcessEvents+0x18c)[0xaaaac2fe6c00]
redis-server *:6379(aeMain+0x24)[0xaaaac2fe7038]
redis-server *:6379(main+0xe0c)[0xaaaac3001afc]
/lib/aarch64-linux-gnu/libc.so.6(+0x273fc)[0xffffb91d73fc]
/lib/aarch64-linux-gnu/libc.so.6(__libc_start_main+0x98)[0xffffb91d74cc]
redis-server *:6379(_start+0x30)[0xaaaac2fe0370]
67093 bio_lazy_free
/lib/aarch64-linux-gnu/libc.so.6(+0x79dfc)[0xffffb9229dfc]
/lib/aarch64-linux-gnu/libc.so.6(pthread_cond_wait+0x208)[0xffffb922c8fc]
redis-server *:6379(bioProcessBackgroundJobs+0x174)[0xaaaac30976e8]
/lib/aarch64-linux-gnu/libc.so.6(+0x7d5c8)[0xffffb922d5c8]
/lib/aarch64-linux-gnu/libc.so.6(+0xe5d1c)[0xffffb9295d1c]
67091 bio_close_file
/lib/aarch64-linux-gnu/libc.so.6(+0x79dfc)[0xffffb9229dfc]
/lib/aarch64-linux-gnu/libc.so.6(pthread_cond_wait+0x208)[0xffffb922c8fc]
redis-server *:6379(bioProcessBackgroundJobs+0x174)[0xaaaac30976e8]
/lib/aarch64-linux-gnu/libc.so.6(+0x7d5c8)[0xffffb922d5c8]
/lib/aarch64-linux-gnu/libc.so.6(+0xe5d1c)[0xffffb9295d1c]
67092 bio_aof
/lib/aarch64-linux-gnu/libc.so.6(+0x79dfc)[0xffffb9229dfc]
/lib/aarch64-linux-gnu/libc.so.6(pthread_cond_wait+0x208)[0xffffb922c8fc]
redis-server *:6379(bioProcessBackgroundJobs+0x174)[0xaaaac30976e8]
/lib/aarch64-linux-gnu/libc.so.6(+0x7d5c8)[0xffffb922d5c8]
/lib/aarch64-linux-gnu/libc.so.6(+0xe5d1c)[0xffffb9295d1c]
67089:signal-handler (1693824528) --------
```
ZRANGE BYSCORE/BYLEX with [LIMIT offset count] option was
using every level in skiplist to jump to the first/last node in range,
but only use level[0] in skiplist to locate the node at offset, resulting
in sub-optimal performance using LIMIT:
```
while (ln && offset--) {
if (reverse) {
ln = ln->backward;
} else {
ln = ln->level[0].forward;
}
}
```
It could be slow when offset is very big. We can get the total rank of
the offset location and use skiplist to jump to it. It is an improvement
from O(offset) to O(log rank).
Below shows how this is implemented (if the offset is positve):
Use the skiplist to seach for the first element in the range, record its
rank `rank_0`, so we can have the rank of the target node `rank_t`.
Meanwhile we record the last node we visited which has zsl->level-1
levels and its rank `rank_1`. Then we start from the zsl->level-1 node,
use skiplist to go forward `rank_t-rank_1` nodes to reach the target node.
It is very similiar when the offset is reversed.
Note that if `rank_t` is very close to `rank_0`, we just start from the first
element in range and go node by node, this for the case when zsl->level-1
node is to far away and it is quicker to reach the target node by node.
Here is a test using a random generated zset including 10000 elements
(with different positive scores), doing a bench mark which compares how
fast the `ZRANGE` command is exucuted before and after the optimization.
The start score is set to 0 and the count is set to 1 to make sure that
most of the time is spent on locating the offset.
```
memtier_benchmark -h 127.0.0.1 -p 6379 --command="zrange test 0 +inf byscore limit <offset> 1"
```
| offset | QPS(unstable) | QPS(optimized) |
|--------|--------|--------|
| 10 | 73386.02 | 74819.82 |
| 1000 | 48084.96 | 73177.73 |
| 2000 | 31156.79 | 72805.83 |
| 5000 | 10954.83 | 71218.21 |
With the result above, we can see that the original code is greatly
slowed down when offset gets bigger, and with the optimization the
speed is almost not affected.
Similiar results are generated when testing reversed offset:
```
memtier_benchmark -h 127.0.0.1 -p 6379 --command="zrange test +inf 0 byscore rev limit <offset> 1"
```
| offset | QPS(unstable) | QPS(optimized) |
|--------|--------|--------|
| 10 | 74505.14 | 71653.67 |
| 1000 | 46829.25 | 72842.75 |
| 2000 | 28985.48 | 73669.01 |
| 5000 | 11066.22 | 73963.45 |
And the same conclusion is drawn from the tests of ZRANGE BYLEX.
This PR adds a new Module API int RM_AddACLCategory(RedisModuleCtx *ctx, const char *category_name) to add a new ACL command category.
Here, we initialize the ACLCommandCategories array by allocating space for 64 categories and duplicate the 21 default categories from the predefined array 'ACLDefaultCommandCategories' into the ACLCommandCategories array while ACL initialization. Valid ACL category names can only contain alphanumeric characters, underscores, and dashes.
The API when called, checks for the onload flag, category name validity, and for duplicate category name if present. If the conditions are satisfied, the API adds the new category to the trailing end of the ACLCommandCategories array and assigns the acl_categories flag bit according to the index at which the category is added.
If any error is encountered the errno is set accordingly by the API.
---------
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Add these INFO metrics:
* client_query_buffer_limit_disconnections
* client_output_buffer_limit_disconnections
Sometimes it is useful to monitor whether clients reaches size limit of
query buffer and output buffer, to decide whether we need to adjust the
buffer size limit or reduce client query payload.
This PR purpose is to make the crash report process thread safe.
main changes include:
1. `setupSigSegvHandler()` is introduced to initialize the signal handler.
This function first initializes the signal handler mutex (if not initialized yet)
and then registers the process to the signal handler.
2. **sigsegvHandler** flags :
SA_NODEFER - don't add the signal to the process signal mask. We use this
flag because we want to be able to handle a second call to the signal manually.
removed SA_RESETHAND: this flag resets the signal handler function upon the first
entrance to the registered function. The reason to use this flag is to protect from
recursively entering the signal handler by the same thread. But, it also means
that if a second thread crashes while handling a signal, the process will be
terminated immediately and we won't get the crash report.
In this PR we discard this flag. The signal handler guard described below purpose
is to solve the above issues.
3. Add a **signal handler lock** with ERRORCHECK attributes.
The lock's purpose is to ensure that only one thread generates a crash report.
Once a second thread enters the signal handler it will be blocked.
We use the ERRORCHECK lock in order to protect from possible deadlock in
case the thread handling the crash gets a signal. In the latest scenario, we log
what we have collected until the handler crashed.
At the end of the crash report we reset the signal handler SIG_DFL, with no flags, and
rethrow the signal to generate a core dump (if enabled) and exit the process.
During the work on this PR we wanted to understand the historical reasons for
how crash is handled.
With respect to the choice of the flag, we believe the **SA_RESETHAND** was not
added for any specific purpose.
**SA_ONSTACK** which is removed here from bugReportEnd(), was originally also
set in the initial registration to signal handler, but removed in 3ada43e73. In addition,
it was removed from another location in deee2c1ef with the following description,
which is also relevant to why it should be removed from bugReportEnd:
> it seems to be some valgrind bug with SA_ONSTACK.
> SA_ONSTACK seems unneeded since WD is not recursive (SA_NODEFER was removed),
> also, not sure if it's even valid without a call to sigaltstack()
Optimized the performance of the SCAN command in a few ways:
1. Move the key filtering (by MATCH pattern) in the scan callback,
so as to avoid collecting them for later filtering.
2. Reduce a many memory allocations and copying (use a reference
to the original sds, instead of creating an robj, an excessive 2 mallocs
and one string duplication)
3. Compare TYPE filter directly (as integers), instead of inefficient string
compare per key.
4. fixed a small bug: when scan zset and hash types, maxiterations uses
a more accurate number to avoid wrong double maxiterations.
Changes **postponed** for a later version (8.0):
1. Prepare to move the TYPE filtering to the scan callback as well. this was
put on hold since it has side effects that can be considered a breaking
change, which is that we will not attempt to do lazy expire (delete) a key
that was filtered by not matching the TYPE (changing it would mean TYPE filter
starts behaving the same as MATCH filter already does in that respect).
2. when the specified key TYPE filter is an unknown type, server will reply a error
immediately instead of doing a full scan that comes back empty handed.
Benchmark result:
For different scenarios, we obtained about 30% or more performance improvement.
Co-authored-by: Oran Agra <oran@redislabs.com>
Apart from adding the missing coverage, this PR also adds `blockedBeforeSleep`
that gathers all block-related functions from `beforeSleep`
The order inside `blockedBeforeSleep` is different: now `handleClientsBlockedOnKeys`
(which may unblock clients) is called before `processUnblockedClients` (which handles
unblocked clients).
It makes sense to have this order.
There are no visible effects of the wrong ordering, except some cleanups of the now-unblocked
client would have happen in the next `beforeSleep` (will now happen in the current one)
The reason we even got into it is because i triggers an assertion in logresreq.c (breaking
the assumption that `unblockClient` is called **before** actually flushing the reply to the socket):
`handleClientsBlockedOnKeys` is called, then it calls `moduleUnblockClientOnKey`, which calls
`moduleUnblockClient`, which adds the client to `moduleUnblockedClients` back to `beforeSleep`,
we call `handleClientsWithPendingWritesUsingThreads`, it writes the data of buf to the client, so
`client->bufpos` became 0
On the next `beforeSleep`, we call `moduleHandleBlockedClients`, which calls `unblockClient`,
which calls `reqresAppendResponse`, triggering the assert. (because the `bufpos` is 0) - see https://github.com/redis/redis/pull/12301#discussion_r1226386716
Binbin