Refactor client outbound queues #3733
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Investigating |
neilalexander
force-pushed
the
neil/clientoutbound
branch
2 times, most recently
from
7072394
to
bc0d67f
Compare
neilalexander
force-pushed
the
neil/clientoutbound
branch
2 times, most recently
from
e18c7c1
to
e260af9
Compare
|
4K+64K buffers vs 8 byte message: 48000 byte message: |
neilalexander
force-pushed
the
neil/clientoutbound
branch
2 times, most recently
from
f48254b
to
9aa4d4f
Compare
derekcollison
requested changes
Dec 29, 2022
neilalexander
force-pushed
the
neil/clientoutbound
branch
2 times, most recently
from
cf1eed9
to
2b37beb
Compare
neilalexander
force-pushed
the
neil/clientoutbound
branch
3 times, most recently
from
928f520
to
14f3abb
Compare
derekcollison
requested changes
Jan 6, 2023
neilalexander
force-pushed
the
neil/clientoutbound
branch
2 times, most recently
from
e72f9b7
to
0f7e419
Compare
|
@derekcollison Review comments addressed! |
|
Let's squash down to 1 commit.. |
neilalexander
force-pushed
the
neil/clientoutbound
branch
from
0f7e419
to
1ac2726
Compare
Also try to reduce flakiness of `TestClusterQueueSubs` and `TestCrossAccountServiceResponseTypes`
neilalexander
force-pushed
the
neil/clientoutbound
branch
from
1ac2726
to
e4b6ba2
Compare
derekcollison
added a commit
that referenced
this pull request
Apr 21, 2023
This extends the previous work in #3733 with the following: 1. Remove buffer coalescing, as this could result in a race condition during the `writev` syscall in rare circumstances 2. Add a third buffer size, to ensure that we aren't allocating more than we need to without coalescing 3. Refactor buffer handling in the WebSocket code to reduce allocations and ensure owned buffers aren't incorrectly being pooled resulting in further race conditions Fixes nats-io/nats.ws#194. Signed-off-by: Neil Twigg <neil@nats.io>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Previously, dynamically-sized buffers were used and discarded often in the client outbound queues. This PR removes the dynamic sizing and instead replaces them with fixed-size coalesced buffers, as well as a
sync.Poolto enable efficient buffer reuse. The result is that the server now allocates dramatically less memory when sending data to clients.The following results are generated by running
-bench=BenchmarkJetStreamConsume -benchtime=100000x -memprofile=.Memory allocations on
main(2.9.10) as of this morning:Memory allocations on this PR:
I don't know if this will make a huge difference to throughput in the usual case, but for servers that are pushing a lot of traffic or handling a large number of concurrent clients, this should reduce the GC burden noticeably and, with any luck, bring load averages down a little bit.
There are a couple TODOs around WebSockets but figured they would be better dealt with separately.
Signed-off-by: Neil Twigg neil@nats.io
/cc @nats-io/core