You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "sunm2n (via GitHub)" <gi...@apache.org> on 2026/10/06 07:29:09 UTC
[PR] KAFKA-21227: Speed up ColdStartStickinessIntegrationTest [kafka]
sunm2n opened a new pull request, #23720:
URL: https://github.com/apache/kafka/pull/23720
The classic protocol case of ColdStartStickinessIntegrationTest takes about
73 seconds locally, while the streams protocol case takes about 23 seconds.
After the first phase, the test closes both instances and waits for the
group to become empty before the cold restart. With the classic protocol,
the stream threads stay in the group on close, so the group only becomes
empty when both members are removed on heartbeat expiration, about 44
seconds after the close. The streams protocol case is not affected, because
dynamic members leave the group on close by default.
This closes the first-phase instances with LEAVE_GROUP, as #23600 did for
JoinGracePeriodDurabilityIntegrationTest, and keeps the original 60 second
close timeout. The test still waits for the group to become empty before
the cold restart, and the existing assertion that no records are restored
passes with both protocols.
Local runs on the same machine:
| | Before (2 runs) | After (2 runs) |
|---|---|---|
| classic | 72.8s, 72.9s | 28.8s, 29.0s |
| streams | 23.6s, 22.6s | 23.3s, 23.6s |
Both protocol cases passed in each run.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: jira-unsubscribe@kafka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [PR] KAFKA-21227: Speed up ColdStartStickinessIntegrationTest [kafka]
Posted by "sunm2n (via GitHub)" <gi...@apache.org>.
sunm2n commented on PR #23720:
URL: https://github.com/apache/kafka/pull/23720#issuecomment-6011687402
@chia7712 @m1a2st Could you take a look when you have a chance? This applies
the same LEAVE_GROUP change as #23600 to ColdStartStickinessIntegrationTest.
cc @mjsax for the cold-start stickiness coverage introduced in #22630.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: jira-unsubscribe@kafka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [PR] KAFKA-21227: Speed up ColdStartStickinessIntegrationTest [kafka]
Posted by "sunm2n (via GitHub)" <gi...@apache.org>.
sunm2n commented on PR #23720:
URL: https://github.com/apache/kafka/pull/23720#issuecomment-6394269515
@m1a2st Gentle ping. The CI failure is fixed, and all checks pass on the latest commit.
Instead of using LEAVE_GROUP, the fix now shortens the session timeout for the classic case,
bringing the local runtime down from about 73s to 34s.
Could you take another look when you have time? Thanks!
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: jira-unsubscribe@kafka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [PR] KAFKA-21227: Speed up ColdStartStickinessIntegrationTest [kafka]
Posted by "sunm2n (via GitHub)" <gi...@apache.org>.
sunm2n commented on PR #23720:
URL: https://github.com/apache/kafka/pull/23720#issuecomment-6030826724
> Thanks for the patch. Could you take a look at the failing test in CI?
Thanks for checking. The failure was caused by this change. With
LEAVE_GROUP, the first instance's LeaveGroup triggered a rebalance while
the second instance was still running, so the second instance took over
all four tasks while shutting down and created state directories for them
(6 instead of 4).
I've restored the original close and instead set a 6 second session
timeout and a 1 second heartbeat interval for the classic case. Locally,
the classic case takes about 34s (73s before) and passed in all 10 runs.
I couldn't run it on Java 25 locally, so I'll keep an eye on CI. I've
also updated the description.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: jira-unsubscribe@kafka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org