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.
# 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]
```
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>
Avoid calling unwatchAllKeys when running touchAllWatchedKeysInDb (which was unnecessary)
This can potentially lead to use-after-free and memory corruption when the next entry
pointer held by the watched keys iterator is freed when unwatching all keys of a specific client.
found with address sanitizer, added a test which will not always fail (depending on the random
dict hashing seed)
problem introduced in #9829 (Reids 7.0)
Co-authored-by: Oran Agra <oran@redislabs.com>
It look like it will generate a warning in FreeBSD:
```
./server.h:105:9: warning: 'member2struct' macro redefined [-Wmacro-redefined]
#define member2struct(struct_name, member_name, member_addr) \
^
/usr/include/sys/param.h:365:9: note: previous definition is here
#define member2struct(s, m, x) \
^
```
Add a `redis_` prefix to it, avoid the warning, introduced in #11511
In unwatchAllKeys() function, we traverse all the keys watched by the client,
and for each key we need to remove the client from the list of clients watching that key.
This is implemented by listSearchKey which traverses the list of clients.
If we can reach the node of the list of clients from watchedKey in O(1) time,
then we do not need to call listSearchKey anymore.
Changes in this PR: put the node of the list of clients of each watched key in the
db inside the watchedKey structure. In this way, for every key watched by the client,
we can get the watchedKey structure and then reach the node in the list of clients in
db->watched_keys to remove it from that list.
From the perspective of the list of clients watching the key, the list node is inside a
watchedKey structure, so we can get to the watchedKey struct from the listnode by
struct member offset math. And because of this, node->value is not used, we can point
node->value to the list itself, so that we don't need to fetch the list of clients from the dict.
## Issue
During the MULTI/EXEC flow, each command gets queued until the `EXEC`
command is received and during this phase on every command queue, a
`realloc` is being invoked. This could be expensive based on the realloc
behavior (if copy to a new memory location).
## Solution
In order to reduce the no. of syscall, couple of optimization I've used.
1. By default, reserve memory for atleast two commands. `MULTI/EXEC` for a
single command doesn't have any significance. Hence, I believe customer wouldn't use it.
2. For further reservation, increase the memory allocation in exponent growth (power of 2).
This reduces the no. of `realloc` call from `N` to `log(N)` times.
## Other changes:
* Include multi exec queued command array in client memory consumption calculation
(affects client eviction too)
This is harmless, we only restore mstate to make sure we
free the right pointer in freeClientMultiState, but it'll
be nicer to also sync that argv_len var back.
The important part is that read-only scripts (not just EVAL_RO
and FCALL_RO, but also ones with `no-writes` executed by normal EVAL or
FCALL), will now be permitted to run during CLIENT PAUSE WRITE (unlike
before where only the _RO commands would be processed).
Other than that, some errors like OOM, READONLY, MASTERDOWN are now
handled by processCommand, rather than the command itself affects the
error string (and even error code in some cases), and command stats.
Besides that, now the `may-replicate` commands, PFCOUNT and PUBLISH, will
be considered `write` commands in scripts and will be blocked in all
read-only scripts just like other write commands.
They'll also be blocked in EVAL_RO (i.e. even for scripts without the
`no-writes` shebang flag.
This commit also hides the `may_replicate` flag from the COMMAND command
output. this is a **breaking change**.
background about may_replicate:
We don't want to expose a no-may-replicate flag or alike to scripts, since we
consider the may-replicate thing an internal concern of redis, that we may
some day get rid of.
In fact, the may-replicate flag was initially introduced to flag EVAL: since
we didn't know what it's gonna do ahead of execution, before function-flags
existed). PUBLISH and PFCOUNT, both of which because they have side effects
which may some day be fixed differently.
code changes:
The changes in eval.c are mostly code re-ordering:
- evalCalcFunctionName is extracted out of evalGenericCommand
- evalExtractShebangFlags is extracted luaCreateFunction
- evalGetCommandFlags is new code
This PR fix 2 issues on Lua scripting:
* Server error reply statistics (some errors were counted twice).
* Error code and error strings returning from scripts (error code was missing / misplaced).
## Statistics
a Lua script user is considered part of the user application, a sophisticated transaction,
so we want to count an error even if handled silently by the script, but when it is
propagated outwards from the script we don't wanna count it twice. on the other hand,
if the script decides to throw an error on its own (using `redis.error_reply`), we wanna
count that too.
Besides, we do count the `calls` in command statistics for the commands the script calls,
we we should certainly also count `failed_calls`.
So when a simple `eval "return redis.call('set','x','y')" 0` fails, it should count the failed call
to both SET and EVAL, but the `errorstats` and `total_error_replies` should be counted only once.
The PR changes the error object that is raised on errors. Instead of raising a simple Lua
string, Redis will raise a Lua table in the following format:
```
{
err='<error message (including error code)>',
source='<User source file name>',
line='<line where the error happned>',
ignore_error_stats_update=true/false,
}
```
The `luaPushError` function was modified to construct the new error table as describe above.
The `luaRaiseError` was renamed to `luaError` and is now simply called `lua_error` to raise
the table on the top of the Lua stack as the error object.
The reason is that since its functionality is changed, in case some Redis branch / fork uses it,
it's better to have a compilation error than a bug.
The `source` and `line` fields are enriched by the error handler (if possible) and the
`ignore_error_stats_update` is optional and if its not present then the default value is `false`.
If `ignore_error_stats_update` is true, the error will not be counted on the error stats.
When parsing Redis call reply, each error is translated to a Lua table on the format describe
above and the `ignore_error_stats_update` field is set to `true` so we will not count errors
twice (we counted this error when we invoke the command).
The changes in this PR might have been considered as a breaking change for users that used
Lua `pcall` function. Before, the error was a string and now its a table. To keep backward
comparability the PR override the `pcall` implementation and extract the error message from
the error table and return it.
Example of the error stats update:
```
127.0.0.1:6379> lpush l 1
(integer) 2
127.0.0.1:6379> eval "return redis.call('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value. script: e471b73f1ef44774987ab00bdf51f21fd9f7974a, on @user_script:1.
127.0.0.1:6379> info Errorstats
# Errorstats
errorstat_WRONGTYPE:count=1
127.0.0.1:6379> info commandstats
# Commandstats
cmdstat_eval:calls=1,usec=341,usec_per_call=341.00,rejected_calls=0,failed_calls=1
cmdstat_info:calls=1,usec=35,usec_per_call=35.00,rejected_calls=0,failed_calls=0
cmdstat_lpush:calls=1,usec=14,usec_per_call=14.00,rejected_calls=0,failed_calls=0
cmdstat_get:calls=1,usec=10,usec_per_call=10.00,rejected_calls=0,failed_calls=1
```
## error message
We can now construct the error message (sent as a reply to the user) from the error table,
so this solves issues where the error message was malformed and the error code appeared
in the middle of the error message:
```diff
127.0.0.1:6379> eval "return redis.call('set','x','y')" 0
-(error) ERR Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479): @user_script:1: OOM command not allowed when used memory > 'maxmemory'.
+(error) OOM command not allowed when used memory > 'maxmemory' @user_script:1. Error running script (call to 71e6319f97b0fe8bdfa1c5df3ce4489946dda479)
```
```diff
127.0.0.1:6379> eval "redis.call('get', 'l')" 0
-(error) ERR Error running script (call to f_8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1): @user_script:1: WRONGTYPE Operation against a key holding the wrong kind of value
+(error) WRONGTYPE Operation against a key holding the wrong kind of value script: 8a705cfb9fb09515bfe57ca2bd84a5caee2cbbd1, on @user_script:1.
```
Notica that `redis.pcall` was not change:
```
127.0.0.1:6379> eval "return redis.pcall('get', 'l')" 0
(error) WRONGTYPE Operation against a key holding the wrong kind of value
```
## other notes
Notice that Some commands (like GEOADD) changes the cmd variable on the client stats so we
can not count on it to update the command stats. In order to be able to update those stats correctly
we needed to promote `realcmd` variable to be located on the client struct.
Tests was added and modified to verify the changes.
Related PR's: #10279, #10218, #10278, #10309
Co-authored-by: Oran Agra <oran@redislabs.com>
When WATCH is called on a key that's already logically expired, avoid discarding the
transaction when the keys is actually deleted.
When WATCH is called, a flag is stored if the key is already expired
at the time of watch. The expired key is not deleted, only checked.
When a key is "touched", if it is deleted and it was already expired
when a client watched it, the client is not marked as dirty.
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: zhaozhao.zz <zhaozhao.zz@alibaba-inc.com>
The mess:
Some parts use alsoPropagate for late propagation, others using an immediate one (propagate()),
causing edge cases, ugly/hacky code, and the tendency for bugs
The basic idea is that all commands are propagated via alsoPropagate (i.e. added to a list) and the
top-most call() is responsible for going over that list and actually propagating them (and wrapping
them in MULTI/EXEC if there's more than one command). This is done in the new function,
propagatePendingCommands.
Callers to propagatePendingCommands:
1. top-most call() (we want all nested call()s to add to the also_propagate array and just the top-most
one to propagate them) - via `afterCommand`
2. handleClientsBlockedOnKeys: it is out of call() context and it may propagate stuff - via `afterCommand`.
3. handleClientsBlockedOnKeys edge case: if the looked-up key is already expired, we will propagate the
expire but will not unblock any client so `afterCommand` isn't called. in that case, we have to propagate
the deletion explicitly.
4. cron stuff: active-expire and eviction may also propagate stuff
5. modules: the module API allows to propagate stuff from just about anywhere (timers, keyspace notifications,
threads). I could have tried to catch all the out-of-call-context places but it seemed easier to handle it in one
place: when we free the context. in the spirit of what was done in call(), only the top-most freeing of a module
context may cause propagation.
6. modules: when using a thread-safe ctx it's not clear when/if the ctx will be freed. we do know that the module
must lock the GIL before calling RM_Replicate/RM_Call so we propagate the pending commands when
releasing the GIL.
A "known limitation", which were actually a bug, was fixed because of this commit (see propagate.tcl):
When using a mix of RM_Call with `!` and RM_Replicate, the command would propagate out-of-order:
first all the commands from RM_Call, and then the ones from RM_Replicate
Another thing worth mentioning is that if, in the past, a client would issue a MULTI/EXEC with just one
write command the server would blindly propagate the MULTI/EXEC too, even though it's redundant.
not anymore.
This commit renames propagate() to propagateNow() in order to cause conflicts in pending PRs.
propagatePendingCommands is the only caller of propagateNow, which is now a static, internal helper function.
Optimizations:
1. alsoPropagate will not add stuff to also_propagate if there's no AOF and replicas
2. alsoPropagate reallocs also_propagagte exponentially, to save calls to memmove
Bugfixes:
1. CONFIG SET can create evictions, sending notifications which can cause to dirty++ with modules.
we need to prevent it from propagating to AOF/replicas
2. We need to set current_client in RM_Call. buggy scenario:
- CONFIG SET maxmemory, eviction notifications, module hook calls RM_Call
- assertion in lookupKey crashes, because current_client has CONFIG SET, which isn't CMD_WRITE
3. minor: in eviction, call propagateDeletion after notification, like active-expire and all commands
(we always send a notification before propagating the command)
Currently, the watching clients are marked as dirty when a watched
key is touched, but we continue watching the keys for no reason.
Then, when the same key is touched again, we iterate again on the
watching clients list and mark all clients as dirty again.
Only when the exec/unwatch command is issued will the client be
removed from the key->watching_clients list. The same applies when
a dirty client calls the WATCH command. The key will be added to be
watched by the client even if it has no effect.
In the field, no performance degradation was observed as a result of the
current implementation; it is merely a cleanup with possible memory and
performance gains in some situations.
Co-authored-by: Oran Agra <oran@redislabs.com>
For diskless replication in swapdb mode, considering we already spend replica memory
having a backup of current db to restore in case of failure, we can have the following benefits
by instead swapping database only in case we succeeded in transferring db from master:
- Avoid `LOADING` response during failed and successful synchronization for cases where the
replica is already up and running with data.
- Faster total time of diskless replication, because now we're moving from Transfer + Flush + Load
time to Transfer + Load only. Flushing the tempDb is done asynchronously after swapping.
- This could be implemented also for disk replication with similar benefits if consumers are willing
to spend the extra memory usage.
General notes:
- The concept of `backupDb` becomes `tempDb` for clarity.
- Async loading mode will only kick in if the replica is syncing from a master that has the same
repl-id the one it had before. i.e. the data it's getting belongs to a different time of the same timeline.
- New property in INFO: `async_loading` to differentiate from the blocking loading
- Slot to Key mapping is now a field of `redisDb` as it's more natural to access it from both server.db
and the tempDb that is passed around.
- Because this is affecting replicas only, we assume that if they are not readonly and write commands
during replication, they are lost after SYNC same way as before, but we're still denying CONFIG SET
here anyways to avoid complications.
Considerations for review:
- We have many cases where server.loading flag is used and even though I tried my best, there may
be cases where async_loading should be checked as well and cases where it shouldn't (would require
very good understanding of whole code)
- Several places that had different behavior depending on the loading flag where actually meant to just
handle commands coming from the AOF client differently than ones coming from real clients, changed
to check CLIENT_ID_AOF instead.
**Additional for Release Notes**
- Bugfix - server.dirty was not incremented for any kind of diskless replication, as effect it wouldn't
contribute on triggering next database SAVE
- New flag for RM_GetContextFlags module API: REDISMODULE_CTX_FLAGS_ASYNC_LOADING
- Deprecated RedisModuleEvent_ReplBackup. Starting from Redis 7.0, we don't fire this event.
Instead, we have the new RedisModuleEvent_ReplAsyncLoad holding 3 sub-events: STARTED,
ABORTED and COMPLETED.
- New module flag REDISMODULE_OPTIONS_HANDLE_REPL_ASYNC_LOAD for RedisModule_SetModuleOptions
to allow modules to declare they support the diskless replication with async loading (when absent, we fall
back to disk-based loading).
Co-authored-by: Eduardo Semprebon <edus@saxobank.com>
Co-authored-by: Oran Agra <oran@redislabs.com>
## Background
For redis master, one replica uses one copy of replication buffer, that is a big waste of memory,
more replicas more waste, and allocate/free memory for every reply list also cost much.
If we set client-output-buffer-limit small and write traffic is heavy, master may disconnect with
replicas and can't finish synchronization with replica. If we set client-output-buffer-limit big,
master may be OOM when there are many replicas that separately keep much memory.
Because replication buffers of different replica client are the same, one simple idea is that
all replicas only use one replication buffer, that will effectively save memory.
Since replication backlog content is the same as replicas' output buffer, now we
can discard replication backlog memory and use global shared replication buffer
to implement replication backlog mechanism.
## Implementation
I create one global "replication buffer" which contains content of replication stream.
The structure of "replication buffer" is similar to the reply list that exists in every client.
But the node of list is `replBufBlock`, which has `id, repl_offset, refcount` fields.
```c
/* Replication buffer blocks is the list of replBufBlock.
*
* +--------------+ +--------------+ +--------------+
* | refcount = 1 | ... | refcount = 0 | ... | refcount = 2 |
* +--------------+ +--------------+ +--------------+
* | / \
* | / \
* | / \
* Repl Backlog Replia_A Replia_B
*
* Each replica or replication backlog increments only the refcount of the
* 'ref_repl_buf_node' which it points to. So when replica walks to the next
* node, it should first increase the next node's refcount, and when we trim
* the replication buffer nodes, we remove node always from the head node which
* refcount is 0. If the refcount of the head node is not 0, we must stop
* trimming and never iterate the next node. */
/* Similar with 'clientReplyBlock', it is used for shared buffers between
* all replica clients and replication backlog. */
typedef struct replBufBlock {
int refcount; /* Number of replicas or repl backlog using. */
long long id; /* The unique incremental number. */
long long repl_offset; /* Start replication offset of the block. */
size_t size, used;
char buf[];
} replBufBlock;
```
So now when we feed replication stream into replication backlog and all replicas, we only need
to feed stream into replication buffer `feedReplicationBuffer`. In this function, we set some fields of
replication backlog and replicas to references of the global replication buffer blocks. And we also
need to check replicas' output buffer limit to free if exceeding `client-output-buffer-limit`, and trim
replication backlog if exceeding `repl-backlog-size`.
When sending reply to replicas, we also need to iterate replication buffer blocks and send its
content, when totally sending one block for replica, we decrease current node count and
increase the next current node count, and then free the block which reference is 0 from the
head of replication buffer blocks.
Since now we use linked list to manage replication backlog, it may cost much time for iterating
all linked list nodes to find corresponding replication buffer node. So we create a rax tree to
store some nodes for index, but to avoid rax tree occupying too much memory, i record
one per 64 nodes for index.
Currently, to make partial resynchronization as possible as much, we always let replication
backlog as the last reference of replication buffer blocks, backlog size may exceeds our setting
if slow replicas that reference vast replication buffer blocks, and this method doesn't increase
memory usage since they share replication buffer. To avoid freezing server for freeing unreferenced
replication buffer blocks when we need to trim backlog for exceeding backlog size setting,
we trim backlog incrementally (free 64 blocks per call now), and make it faster in
`beforeSleep` (free 640 blocks).
### Other changes
- `mem_total_replication_buffers`: we add this field in INFO command, it means the total
memory of replication buffers used.
- `mem_clients_slaves`: now even replica is slow to replicate, and its output buffer memory
is not 0, but it still may be 0, since replication backlog and replicas share one global replication
buffer, only if replication buffer memory is more than the repl backlog setting size, we consider
the excess as replicas' memory. Otherwise, we think replication buffer memory is the consumption
of repl backlog.
- Key eviction
Since all replicas and replication backlog share global replication buffer, we think only the
part of exceeding backlog size the extra separate consumption of replicas.
Because we trim backlog incrementally in the background, backlog size may exceeds our
setting if slow replicas that reference vast replication buffer blocks disconnect.
To avoid massive eviction loop, we don't count the delayed freed replication backlog into
used memory even if there are no replicas, i.e. we also regard this memory as replicas's memory.
- `client-output-buffer-limit` check for replica clients
It doesn't make sense to set the replica clients output buffer limit lower than the repl-backlog-size
config (partial sync will succeed and then replica will get disconnected). Such a configuration is
ignored (the size of repl-backlog-size will be used). This doesn't have memory consumption
implications since the replica client will share the backlog buffers memory.
- Drop replication backlog after loading data if needed
We always create replication backlog if server is a master, we need it because we put DELs in
it when loading expired keys in RDB, but if RDB doesn't have replication info or there is no rdb,
it is not possible to support partial resynchronization, to avoid extra memory of replication backlog,
we drop it.
- Multi IO threads
Since all replicas and replication backlog use global replication buffer, if I/O threads are enabled,
to guarantee data accessing thread safe, we must let main thread handle sending the output buffer
to all replicas. But before, other IO threads could handle sending output buffer of all replicas.
## Other optimizations
This solution resolve some other problem:
- When replicas disconnect with master since of out of output buffer limit, releasing the output
buffer of replicas may freeze server if we set big `client-output-buffer-limit` for replicas, but now,
it doesn't cause freezing.
- This implementation may mitigate reply list copy cost time(also freezes server) when one replication
has huge reply buffer and another replica can copy buffer for full synchronization. now, we just copy
reference info, it is very light.
- If we set replication backlog size big, it also may cost much time to copy replication backlog into
replica's output buffer. But this commit eliminates this problem.
- Resizing replication backlog size doesn't empty current replication backlog content.
When queuing a multi command we duplicated the argv (meaning an alloc
and a memcpy). This isn't needed since we can use the previously allocated
argv and just reset the client objects argv to NULL. This should saves some
memory and is a minor optimization in heavy MULTI/EXEC traffic, especially
if there are lots of arguments.
Changes in #9528 lead to memory leak if the command implementation
used rewriteClientCommandArgument inside MULTI-EXEC.
Adding an explicit test for that case since the test that uncovered it
didn't specifically target this scenario
### Description
A mechanism for disconnecting clients when the sum of all connected clients is above a
configured limit. This prevents eviction or OOM caused by accumulated used memory
between all clients. It's a complimentary mechanism to the `client-output-buffer-limit`
mechanism which takes into account not only a single client and not only output buffers
but rather all memory used by all clients.
#### Design
The general design is as following:
* We track memory usage of each client, taking into account all memory used by the
client (query buffer, output buffer, parsed arguments, etc...). This is kept up to date
after reading from the socket, after processing commands and after writing to the socket.
* Based on the used memory we sort all clients into buckets. Each bucket contains all
clients using up up to x2 memory of the clients in the bucket below it. For example up
to 1m clients, up to 2m clients, up to 4m clients, ...
* Before processing a command and before sleep we check if we're over the configured
limit. If we are we start disconnecting clients from larger buckets downwards until we're
under the limit.
#### Config
`maxmemory-clients` max memory all clients are allowed to consume, above this threshold
we disconnect clients.
This config can either be set to 0 (meaning no limit), a size in bytes (possibly with MB/GB
suffix), or as a percentage of `maxmemory` by using the `%` suffix (e.g. setting it to `10%`
would mean 10% of `maxmemory`).
#### Important code changes
* During the development I encountered yet more situations where our io-threads access
global vars. And needed to fix them. I also had to handle keeps the clients sorted into the
memory buckets (which are global) while their memory usage changes in the io-thread.
To achieve this I decided to simplify how we check if we're in an io-thread and make it
much more explicit. I removed the `CLIENT_PENDING_READ` flag used for checking
if the client is in an io-thread (it wasn't used for anything else) and just used the global
`io_threads_op` variable the same way to check during writes.
* I optimized the cleanup of the client from the `clients_pending_read` list on client freeing.
We now store a pointer in the `client` struct to this list so we don't need to search in it
(`pending_read_list_node`).
* Added `evicted_clients` stat to `INFO` command.
* Added `CLIENT NO-EVICT ON|OFF` sub command to exclude a specific client from the
client eviction mechanism. Added corrosponding 'e' flag in the client info string.
* Added `multi-mem` field in the client info string to show how much memory is used up
by buffered multi commands.
* Client `tot-mem` now accounts for buffered multi-commands, pubsub patterns and
channels (partially), tracking prefixes (partially).
* CLIENT_CLOSE_ASAP flag is now handled in a new `beforeNextClient()` function so
clients will be disconnected between processing different clients and not only before sleep.
This new function can be used in the future for work we want to do outside the command
processing loop but don't want to wait for all clients to be processed before we get to it.
Specifically I wanted to handle output-buffer-limit related closing before we process client
eviction in case the two race with each other.
* Added a `DEBUG CLIENT-EVICTION` command to print out info about the client eviction
buckets.
* Each client now holds a pointer to the client eviction memory usage bucket it belongs to
and listNode to itself in that bucket for quick removal.
* Global `io_threads_op` variable now can contain a `IO_THREADS_OP_IDLE` value
indicating no io-threading is currently being executed.
* In order to track memory used by each clients in real-time we can't rely on updating
these stats in `clientsCron()` alone anymore. So now I call `updateClientMemUsage()`
(used to be `clientsCronTrackClientsMemUsage()`) after command processing, after
writing data to pubsub clients, after writing the output buffer and after reading from the
socket (and maybe other places too). The function is written to be fast.
* Clients are evicted if needed (with appropriate log line) in `beforeSleep()` and before
processing a command (before performing oom-checks and key-eviction).
* All clients memory usage buckets are grouped as follows:
* All clients using less than 64k.
* 64K..128K
* 128K..256K
* ...
* 2G..4G
* All clients using 4g and up.
* Added client-eviction.tcl with a bunch of tests for the new mechanism.
* Extended maxmemory.tcl to test the interaction between maxmemory and
maxmemory-clients settings.
* Added an option to flag a numeric configuration variable as a "percent", this means that
if we encounter a '%' after the number in the config file (or config set command) we
consider it as valid. Such a number is store internally as a negative value. This way an
integer value can be interpreted as either a percent (negative) or absolute value (positive).
This is useful for example if some numeric configuration can optionally be set to a percentage
of something else.
Co-authored-by: Oran Agra <oran@redislabs.com>
This commit introduced a new flag to the RM_Call:
'C' - Check if the command can be executed according to the ACLs associated with it.
Also, three new API's added to check if a command, key, or channel can be executed or accessed
by a user, according to the ACLs associated with it.
- RM_ACLCheckCommandPerm
- RM_ACLCheckKeyPerm
- RM_ACLCheckChannelPerm
The user for these API's is a RedisModuleUser object, that for a Module user returned by the RM_CreateModuleUser API, or for a general ACL user can be retrieved by these two new API's:
- RM_GetCurrentUserName - Retrieve the user name of the client connection behind the current context.
- RM_GetModuleUserFromUserName - Get a RedisModuleUser from a user name
As a result of getting a RedisModuleUser from name, it can now also access the general ACL users (not just ones created by the module).
This mean the already existing API RM_SetModuleUserACL(), can be used to change the ACL rules for such users.
The `cmd` argument was completely unused, and all the code that bothered to pass it was unnecessary.
This is a prepartion for a future commit that treats subcommands as commands
There are two issues fixed in this commit:
1. we want to fail the EXEC command in case there is a watched key that's logically
expired but not yet deleted by active expire or lazy expire.
2. we saw that currently cache time is update in every `call()` (including nested calls),
this time is being also being use for the isKeyExpired comparison, we want to update
the cache time only in the first call (execCommand)
Co-authored-by: Oran Agra <oran@redislabs.com>
Do not queue command in an already aborted MULTI state.
We can detect an error (watched key).
So in queueMultiCommand, we also can return early.
Like we deal with `CLIENT_DIRTY_EXEC`.
Add publish channel permissions check in processCommand.
processCommand didn't check publish channel permissions, so we can
queue a publish command in a transaction. But when exec the transaction,
it will fail with -NOPERM.
We also union keys/commands/channels permissions check togegher in
ACLCheckAllPerm. Remove pubsubCheckACLPermissionsOrReply in
publishCommand/subscribeCommand/psubscribeCommand. Always
check permissions in processCommand/execCommand/
luaRedisGenericCommand.
Bug 1:
When a module ctx is freed moduleHandlePropagationAfterCommandCallback
is called and handles propagation. We want to prevent it from propagating
commands that were not replicated by the same context. Example:
1. module1.foo does: RM_Replicate(cmd1); RM_Call(cmd2); RM_Replicate(cmd3)
2. RM_Replicate(cmd1) propagates MULTI and adds cmd1 to also_propagagte
3. RM_Call(cmd2) create a new ctx, calls call() and destroys the ctx.
4. moduleHandlePropagationAfterCommandCallback is called, calling
alsoPropagates EXEC (Note: EXEC is still not written to socket),
setting server.in_trnsaction = 0
5. RM_Replicate(cmd3) is called, propagagting yet another MULTI (now
we have nested MULTI calls, which is no good) and then cmd3
We must prevent RM_Call(cmd2) from resetting server.in_transaction.
REDISMODULE_CTX_MULTI_EMITTED was revived for that purpose.
Bug 2:
Fix issues with nested RM_Call where some have '!' and some don't.
Example:
1. module1.foo does RM_Call of module2.bar without replication (i.e. no '!')
2. module2.bar internally calls RM_Call of INCR with '!'
3. at the end of module1.foo we call RM_ReplicateVerbatim
We want the replica/AOF to see only module1.foo and not the INCR from module2.bar
Introduced a global replication_allowed flag inside RM_Call to determine
whether we need to replicate or not (even if '!' was specified)
Other changes:
Split beforePropagateMultiOrExec to beforePropagateMulti afterPropagateExec
just for better readability
This PR not only fixes the problem that swapdb does not make the
transaction fail, but also optimizes the FLUSHALL and FLUSHDB command to
set the CLIENT_DIRTY_CAS flag to avoid unnecessary traversal of clients.
FLUSHDB was changed to first iterate on all watched keys, and then on the
clients watching each key.
Instead of iterating though all clients, and for each iterate on watched keys.
Co-authored-by: Oran Agra <oran@redislabs.com>
In the distant history there was only the read flag for commands, and whatever
command that didn't have the read flag was a write one.
Then we added the write flag, but some portions of the code still used !read
Also some commands that don't work on the keyspace at all, still have the read
flag.
Changes in this commit:
1. remove the read-only flag from TIME, ECHO, ROLE and LASTSAVE
2. EXEC command used to decides if it should propagate a MULTI by looking at
the command flags (!read & !admin).
When i was about to change it to look at the write flag instead, i realized
that this would cause it not to propagate a MULTI for PUBLISH, EVAL, and
SCRIPT, all 3 are not marked as either a read command or a write one (as
they should), but all 3 are calling forceCommandPropagation.
So instead of introducing a new flag to denote a command that "writes" but
not into the keyspace, and still needs propagation, i decided to rely on
the forceCommandPropagation, and just fix the code to propagate MULTI when
needed rather than depending on the command flags at all.
The implication of my change then is that now it won't decide to propagate
MULTI when it sees one of these: SELECT, PING, INFO, COMMAND, TIME and
other commands which are neither read nor write.
3. Changing getNodeByQuery and clusterRedirectBlockedClientIfNeeded in
cluster.c to look at !write rather than read flag.
This should have no implications, since these code paths are only reachable
for commands which access keys, and these are always marked as either read
or write.
This commit improve MULTI propagation tests, for modules and a bunch of
other special cases, all of which used to pass already before that commit.
the only one that test change that uncovered a change of behavior is the
one that DELs a non-existing key, it used to propagate an empty
multi-exec block, and no longer does.
One way this was happening is when a module issued an RM_Call which would inject MULTI.
If the module command that does that was itself issued by something else that already did
added MULTI (e.g. another module, or a Lua script), it would have caused nested MULTI.
In fact the MULTI state in the client or the MULTI_EMITTED flag in the context isn't
the right indication that we need to propagate MULTI or not, because on a nested calls
(possibly a module action called by a keyspace event of another module action), these
flags aren't retained / reflected.
instead there's now a global propagate_in_transaction flag for that.
in addition to that, we now have a global in_eval and in_exec flags, to serve the flags
of RM_GetContextFlags, since their dependence on the current client is wrong for the same
reasons mentioned above.
Fixes#7923.
This PR appropriates the special `&` symbol (because `@` and `*` are taken),
followed by a literal value or pattern for describing the Pub/Sub patterns that
an ACL user can interact with. It is similar to the existing key patterns
mechanism in function (additive) and implementation (copy-pasta). It also adds
the allchannels and resetchannels ACL keywords, naturally.
The default user is given allchannels permissions, whereas new users get
whatever is defined by the acl-pubsub-default configuration directive. For
backward compatibility in 6.2, the default of this directive is allchannels but
this is likely to be changed to resetchannels in the next major version for
stronger default security settings.
Unless allchannels is set for the user, channel access permissions are checked
as follows :
* Calls to both PUBLISH and SUBSCRIBE will fail unless a pattern matching the
argumentative channel name(s) exists for the user.
* Calls to PSUBSCRIBE will fail unless the pattern(s) provided as an argument
literally exist(s) in the user's list.
Such failures are logged to the ACL log.
Runtime changes to channel permissions for a user with existing subscribing
clients cause said clients to disconnect unless the new permissions permit the
connections to continue. Note, however, that PSUBSCRIBErs' patterns are matched
literally, so given the change bar:* -> b*, pattern subscribers to bar:* will be
disconnected.
Notes/questions:
* UNSUBSCRIBE, PUNSUBSCRIBE and PUBSUB remain unprotected due to lack of reasons
for touching them.
Blocking command should not be used with MULTI, LUA, and RM_Call. This is because,
the caller, who executes the command in this context, expects a reply.
Today, LUA and MULTI have a special (and different) treatment to blocking commands:
LUA - Most commands are marked with no-script flag which are checked when executing
and command from LUA, commands that are not marked (like XREAD) verify that their
blocking mode is not used inside LUA (by checking the CLIENT_LUA client flag).
MULTI - Command that is going to block, first verify that the client is not inside
multi (by checking the CLIENT_MULTI client flag). If the client is inside multi, they
return a result which is a match to the empty key with no timeout (for example blpop
inside MULTI will act as lpop)
For modules that perform RM_Call with blocking command, the returned results type is
REDISMODULE_REPLY_UNKNOWN and the caller can not really know what happened.
Disadvantages of the current state are:
No unified approach, LUA, MULTI, and RM_Call, each has a different treatment
Module can not safely execute blocking command (and get reply or error).
Though It is true that modules are not like LUA or MULTI and should be smarter not
to execute blocking commands on RM_Call, sometimes you want to execute a command base
on client input (for example if you create a module that provides a new scripting
language like javascript or python).
While modules (on modules command) can check for REDISMODULE_CTX_FLAGS_LUA or
REDISMODULE_CTX_FLAGS_MULTI to know not to block the client, there is no way to
check if the command came from another module using RM_Call. So there is no way
for a module to know not to block another module RM_Call execution.
This commit adds a way to unify the treatment for blocking clients by introducing
a new CLIENT_DENY_BLOCKING client flag. On LUA, MULTI, and RM_Call the new flag
turned on to signify that the client should not be blocked. A blocking command
verifies that the flag is turned off before blocking. If a blocking command sees
that the CLIENT_DENY_BLOCKING flag is on, it's not blocking and return results
which are matches to empty key with no timeout (as MULTI does today).
The new flag is checked on the following commands:
List blocking commands: BLPOP, BRPOP, BRPOPLPUSH, BLMOVE,
Zset blocking commands: BZPOPMIN, BZPOPMAX
Stream blocking commands: XREAD, XREADGROUP
SUBSCRIBE, PSUBSCRIBE, MONITOR
In addition, the new flag is turned on inside the AOF client, we do not want to
block the AOF client to prevent deadlocks and commands ordering issues (and there
is also an existing assert in the code that verifies it).
To keep backward compatibility on LUA, all the no-script flags on existing commands
were kept untouched. In addition, a LUA special treatment on XREAD and XREADGROUP was kept.
To keep backward compatibility on MULTI (which today allows SUBSCRIBE, and PSUBSCRIBE).
We added a special treatment on those commands to allow executing them on MULTI.
The only backward compatibility issue that this PR introduces is that now MONITOR
is not allowed inside MULTI.
Tests were added to verify blocking commands are not blocking the client on LUA, MULTI,
or RM_Call. Tests were added to verify the module can check for CLIENT_DENY_BLOCKING flag.
Co-authored-by: Oran Agra <oran@redislabs.com>
Co-authored-by: Itamar Haber <itamar@redislabs.com>
65a3307bc (released in 6.0.6) has a side effect, when processCommand
rejects a command with pre-made shared object error string, it trims the
newlines from the end of the string. if that string is later used with
addReply, the newline will be missing, breaking the protocol, and
leaving the client hung.
It seems that the only scenario which this happens is when replying with
-LOADING to some command, and later using that reply from the CONFIG
SET command (still during loading). this will result in hung client.
Refactoring the code in order to avoid trimming these newlines from
shared string objects, and do the newline trimming only in other cases
where it's needed.
Co-authored-by: Guy Benoish <guy.benoish@redislabs.com>
In order to support the use of multi-exec in pipeline, it is important that
MULTI and EXEC are never rejected and it is easy for the client to know if the
connection is still in multi state.
It was easy to make sure MULTI and DISCARD never fail (done by previous
commits) since these only change the client state and don't do any actual
change in the server, but EXEC is a different story.
Since in the past, it was possible for clients to handle some EXEC errors and
retry the EXEC, we now can't affort to return any error on EXEC other than
EXECABORT, which now carries with it the real reason for the abort too.
Other fixes in this commit:
- Some checks that where performed at the time of queuing need to be re-
validated when EXEC runs, for instance if the transaction contains writes
commands, it needs to be aborted. there was one check that was already done
in execCommand (-READONLY), but other checks where missing: -OOM, -MISCONF,
-NOREPLICAS, -MASTERDOWN
- When a command is rejected by processCommand it was rejected with addReply,
which was not recognized as an error in case the bad command came from the
master. this will enable to count or MONITOR these errors in the future.
- make it easier for tests to create additional (non deferred) clients.
- add tests for the fixes of this commit.
Random command like SPOP with count is replicated as
some SREM operations, and store them in also_propagate
array to propagate after the call, but this would break
atomicity.
To keep the command's atomicity, wrap also_propagate
array with MULTI/EXEC.
Pieter Cailliau