- Add a new 'EXPERIMENTAL' command flag, which causes the command
generator to skip over it and make the command to be unavailable for
execution
- Skip experimental tests by default
- Move the SFLUSH tests from the old framework to the new one
---------
Co-authored-by: YaacovHazan <yaacov.hazan@redislabs.com>
1. `dbRandomKey`: excessive call to `dbFindExpires` (will always return
1 if `allvolatile` + anyway called inside `expireIfNeeded`
2. Add `deleteKeyAndPropagate` that is used by both expiry/eviction
3. Change the order of calls in `expireIfNeeded` to save redundant calls
to `keyIsExpired`
4. `expireIfNeeded`: move `OBJ_STATIC_REFCOUNT` to
`deleteKeyAndPropagate`
5. `performEvictions` now uses `deleteEvictedKeyAndPropagate`
6. active-expire: moved `postExecutionUnitOperations` inside
`activeExpireCycleTryExpire`
7. `activeExpireCycleTryExpire`: less indentation + expire a key if `now
== t`
8. rename `lazy_expire_disabled` to `allow_access_expired`
This PR introduces a new `SFLUSH` command to cluster mode that allows
partial flushing of nodes based on specified slot ranges. Current
implementation is designed to flush all slots of a shard, but future
extensions could allow for more granular flushing.
**Command Usage:**
`SFLUSH <start-slot> <end-slot> [<start-slot> <end-slot>]* [SYNC|ASYNC]`
This command removes all data from the specified slots, either
synchronously or asynchronously depending on the optional SYNC/ASYNC
argument.
**Functionality:**
Current imp of `SFLUSH` command verifies that the provided slot ranges
are valid and cover all of the node's slots before proceeding. If slots
are partially or incorrectly specified, the command will fail and return
an error, ensuring that all slots of a node must be fully covered for
the flush to proceed.
The function supports both synchronous (default) and asynchronous
flushing. In addition, if possible, SFLUSH SYNC will be run as blocking
ASYNC as an optimization.
This PR is based on valkey-io/valkey#829
Previously, ZUNION and ZUNIONSTORE commands used a temporary accumulator dict
and at the end copied it as-is to dstzset->dict. This PR removes accumulator and directly
stores into dstzset->dict, eliminating the extra copy.
Co-authored-by: Rayacoo zisong.cw@alibaba-inc.com
Test 1 - give more time for expiration
Test 2 - Evaluate expiration time boundaries [+1,+2] before setting expiration [+1]
Test 3 - Avoid race on test HFEs propagated to replica
The PR extends `RedisModule_OpenKey`'s flags to include
`REDISMODULE_OPEN_KEY_ACCESS_EXPIRED`, which allows to access expired
keys.
It also allows to access expired subkeys. Currently relevant only for
hash fields
and has its impact on `RM_HashGet` and `RM_Scan`.
This PR introduces the installation of the `musl`-based version of Rust,
in order to support alpine-based runtime environments (Rust is used by
[RedisJSON](https://github.com/RedisJSON/RedisJSON)).
Instead of adding runtime logic to decide which prefix/shared object to
use when doing the reply we can simply use an inline method to avoid runtime
overhead of condition checks, and also keep the code change small.
Preliminary data show improvements on commands that heavily rely on
bulk/mbulk replies (example of LRANGE).
---------
Co-authored-by: debing.sun <debing.sun@redis.com>
Fixes#8825
We're using the fast_float library[1] in our (compiled-in)
floating-point fast_float_strtod implementation for faster and more
portable parsing of 64 decimal strings.
The single file fast_float.h is an amalgamation of the entire library,
which can be (re)generated with the amalgamate.py script (from the
fast_float repository) via the command:
```
python3 ./script/amalgamate.py --license=MIT > $REDIS_SRC/deps/fast_float/fast_float.h
```
[1]: https://github.com/fastfloat/fast_float
The used commit from fast_float library was the one from
https://github.com/fastfloat/fast_float/releases/tag/v3.10.1
---------
Co-authored-by: fcostaoliveira <filipe@redis.com>
Similar to #13530 , applied to HSCAN and ZSCAN in case of listpack
encoding.
**Preliminary benchmark results showcase an improvement of 108% on the
achievable ops/sec for ZSCAN and 65% for HSCAN**.
---------
Co-authored-by: debing.sun <debing.sun@redis.com>
in https://github.com/redis/redis/pull/13519, when `eb` is empty,
`isRax` is not correctly initialized to 0, which can lead to `ebStop()`
potentially entering the wrong rax branch.
On SSCAN, in case of listpack and intset encoding we actually reply the
entire set, and always reply with the cursor 0.
For those cases, we don't need to accumulate the replies in a list and
can completely avoid the overhead of list appending and then iterating
over the list again -- meaning we do N iterations instead of 2N
iterations over the SET and save intermediate memory as well.
Preliminary benchmarks, `SSCAN set:100 0`, showcased an improvement of
60% as visible bellow on a SET with 100 string elements (listpack
encoded).
If the hash previously had HFEs (hash-fields with expiration) but later no longer
does, the key ref in the hash might become outdated after a MOVE, COPY,
RENAME or RESTORE operation. These commands maintain the key ref only
if HFEs are present. That is, we can only be sure that key ref is valid as long as the
hash has HFEs.
When a client in no-touch mode issues a TOUCH command on a key, the
key’s access time should be updated, but in scripts, and module's
RM_Call, it isn’t updated.
Command proc should be matched to the executing client, not the current
client.
Co-authored-by: Udi Ron <udi@speedb.io>
PR #13428 doesn't fully resolve an issue where corruption errors can
still occur on loading of cluster.nodes file - seen on upgrade where
there were no shard_ids (from old Redis), 7.2.5 loading generated new
random ones, and persisted them to the file before gossip/handshake
could propagate the correct ones (or some other nodes unreachable).
This results in a primary/replica having differing shard_id in the
cluster.nodes and then the server cannot startup - reports corruption.
This PR builds on #13428 by simply ignoring the replica's shard_id in
cluster.nodes (if it exists), and uses the replica's primary's shard_id.
Additional handling was necessary to cover the case where the replica
appears before the primary in cluster.nodes, where it will first use a
generated shard_id for the primary, and then correct after it loads the
primary cluster.nodes entry.
---------
Co-authored-by: debing.sun <debing.sun@redis.com>
Found by @oranagra
Currently, when the size of dict becomes 1, we do not check whether
`delta` is positive or negative.
As a result, `non_empty_dicts` is still incremented when the size of
dict changes from 2 to 1.
We should only increment `non_empty_dicts` when `delta` is positive, as
this indicates the first time an element is inserted into the dict.
---------
Co-authored-by: oranagra <oran@redislabs.com>
This is a very easy optimization, that avoids duplicate computation of
the object length for LREM, LPOS, LINSERT na LINDEX.
We can see that sdslen takes 7.7% of the total CPU cycles of the
benchmarks.
Function Stack | CPU Time: Total | CPU Time: Self | Module | Function
(Full) | Source File | Start Address
-- | -- | -- | -- | -- | -- | --
listTypeEqual | 15.50% | 2.346s | redis-server | listTypeEqual |
t_list.c | 0x845dd
sdslen | 7.70% | 2.300s | redis-server | sdslen | sds.h | 0x845e4
Preliminary data showcases 4% improvement in the achievable ops/sec of
LPOS in string elements, and 2% in int elements.
A new BUILD_WITH_MODULES flag was added to the Makefile to control
building the module directory.
The new module directory includes a general Makefile that iterates
over each module, fetch a specific version, and build it.
Co-authored-by: YaacovHazan <yaacov.hazan@redislabs.com>
#13495 introduced a change to reply -LOADING while flushing existing db on a replica. Some of our tests are
sensitive to this change and do no expect -LOADING reply.
Fixing a couple of tests that fail time to time.
This PR is based on the commits from PR
https://github.com/valkey-io/valkey/pull/258,
https://github.com/valkey-io/valkey/pull/593,
https://github.com/valkey-io/valkey/pull/639
This PR optimizes client query buffer handling in Redis by introducing
a reusable query buffer that is used by default for client reads. This
reduces memory usage by ~20KB per client by avoiding allocations for
most clients using short (<16KB) complete commands. For larger or
partial commands, the client still gets its own private buffer.
The primary changes are:
* Adding a reusable query buffer `thread_shared_qb` that clients use by
default.
* Modifying client querybuf initialization and reset logic.
* Freeing idle client query buffers when empty to allow reuse of the
reusable query buffer.
* Master client query buffers are kept private as their contents need to
be preserved for replication stream.
* When nested commands is executed, only the first user uses the reuse
buffer, and subsequent users will still use the private buffer.
In addition to the memory savings, this change shows a 3% improvement in
latency and throughput when running with 1000 active clients.
The memory reduction may also help reduce the need to evict clients when
reaching max memory limit, as the query buffer is the main memory
consumer per client.
This PR is different from https://github.com/valkey-io/valkey/pull/258
1. When a client is in the mid of requiring a reused buffer and
returning it, regardless of whether the query buffer has changed
(expanded), we do not update the reused query buffer in the middle, but
return the reused query buffer (expanded or with data remaining) or
reset it at the end.
2. Adding a new thread variable `thread_shared_qb_used` to avoid
multiple clients requiring the reusable query buffer at the same time.
---------
Signed-off-by: Uri Yagelnik <uriy@amazon.com>
Signed-off-by: Madelyn Olson <matolson@amazon.com>
Co-authored-by: Uri Yagelnik <uriy@amazon.com>
Co-authored-by: Madelyn Olson <madelyneolson@gmail.com>
Co-authored-by: oranagra <oran@redislabs.com>
After https://github.com/redis/redis/pull/13499, If the length set by
`addReplySetLen()` does not match the actual number of elements in the
reply, it will cause protocol broken and result in the client hanging.
RM_RdbLoad() disables AOF temporarily while loading RDB.
Later, it does not enable it back as it checks AOF state (disabled by then)
rather than AOF config parameter.
Added a change to restart AOF according to config parameter.
- Avoid addReplyLongLong (which converts back to string) the value we
already have as a robj, by using addReplyProto + addReply
- Avoid doing dbFind Twice for the same dictEntry on
INCR*/DECR*/SETRANGE/APPEND commands.
- Avoid multiple sdslen calls with the same input on setrangeCommand and
appendCommand
- Introduce setKeyWithDictEntry, which is like setKey(), but accepts an
optional dictEntry input: Avoids the second dictFind in SET command
---------
Co-authored-by: debing.sun <debing.sun@redis.com>
In #13279 (found by @filipecosta90), for custom lookups, we introduce a
comparison function for `lpFind()` to compare entry, but it also
introduces some overhead.
To avoid the overhead of function pointer calls:
1. Extract the lpFindCb() method into a lpFindCbInternal() method that
is easier to inline.
2. Use unlikely to annotate the comparison method, as can only success
once.
---------
Co-authored-by: Ozan Tezcan <ozantezcan@gmail.com>
All the defrag allocations API expects to get a value and replace it, leaving the old value untouchable.
In some cases a value might be shared between multiple keys, in such cases we can not simply replace
it when the defrag callback is called.
To allow support such use cases, the PR adds two new API's to the defrag API:
1. `RM_DefragAllocRaw` - allocate memory base on a given size.
2. `RM_DefragFreeRaw` - Free the given pointer.
Those API's avoid using tcache so they operate just like `RM_DefragAlloc` but allows the user to split
the allocation and the memory free operations into two stages and control when those happen.
In addition the PR adds new API to allow the module to receive notifications when defrag start and end: `RM_RegisterDefragCallbacks`
Those callbacks are the same as `RM_RegisterDefragFunc` but promised to be called and the start
and the end of the defrag process.
This PR introduces a dedicated implementation for the SMEMBERS command
that avoids using the more generalized sinterGenericCommand function.
By tailoring the logic specifically for SMEMBERS, we reduce unnecessary
processing and memory overheads that were previously incurred by
handling more complex cases like set intersections.
---------
Co-authored-by: debing.sun <debing.sun@redis.com>
# Overall improvement
TBD ( current is approximately 6% on the achievable ops/sec), coming
from:
- In case of no module we can skip 1.3% CPU cycles on dict Iterator
creation/deletion
- Use addReplyBulkCBuffer instead of addReplyBulkCString to avoid
runtime strlen overhead within HELLO reply on string constants.
## Optimization 1: In case of no module we can skip 1.3% CPU cycles on
dict Iterator creation/deletion.
## Optimization 2: Use addReplyBulkCBuffer instead of
addReplyBulkCString to avoid runtime strlen overhead within HELLO reply
on string constants.
Currently aeApiPoll panic does not record error code information. Added
variable parameter formatting to _serverPanic to fix the issue
---------
Co-authored-by: yingyin.chen <15816602944@163.com>
On a full sync, replica starts discarding existing db. If the existing
db is huge and flush is happening synchronously, replica may become
unresponsive.
Adding a change to yield back to event loop while flushing db on
a replica. Replica will reply -LOADING in this case. Note that while
replica is loading the new rdb, it may get an error and start flushing
the partial db. This step may take a long time as well. Similarly,
replica will reply -LOADING in this case.
To call processEventsWhileBlocked() and reply -LOADING, we need to do:
- Set connSetReadHandler() null not to process further data from the master
- Set server.loading flag
- Call blockingOperationStarts()
rdbload() already does these steps and calls processEventsWhileBlocked()
while loading the rdb. Added a new call rdbLoadWithEmptyFunc() which
accepts callback to flush db before loading rdb or when an error
happens while loading.
For diskless replication, doing something similar and calling emptyData()
after setting required flags.
Additional changes:
- Allow `appendonly` config change during loading.
Config can be changed while loading data on startup or on replication
when slave is loading RDB. We allow config change command to update
`server.aof_enabled` and then lazily apply config change after loading
operation is completed.
- Added a test for `replica-lazy-flush` config
so far ./runtest --dump-logs used work for servers started within the
test proc.
now it'll also work on servers started outside the test proc scope.
the downside is that these logs can be huge if they served many tests
and not just the failing one.
but for some rare failures, we rather have that than nothing.
this feature isn't enabled y default, but is used by our GH actions.
Currently, module commands are not returned for the `ACL CAT <category>`
command, but skipped instead. Since now modules can add ACL categories
they should no longer be skipped.
Fixed the issue about GETRANGE and SUBSTR command
return unexpected result caused by the `start` and `end` out of
definition range of string.
---
## break change
Before this PR, when negative `end` was out of range (i.e., end <
-strlen), we would fix it to 0 to get the substring, which also resulted
in the first character still being returned for this kind of out of
range.
After this PR, we ensure that `GETRANGE` returns an empty bulk when the
negative end index is out of range.
Closes#11738
---------
Co-authored-by: debing.sun <debing.sun@redis.com>
Move the TYPE filtering to the scan callback so that avoided the
`lookupKey` operation. This is the follow-up to #12209 . In this thread
we introduced two breaking changes：
1. we will not attempt to do lazy expire (delete) a key that was
filtered by not matching the TYPE (like we already do for MATCH
pattern).
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.
YaacovHazan