[
https://issues.apache.org/jira/browse/KAFKA-21106?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18116685#comment-18116685
]
Apoorv Mittal commented on KAFKA-21106:
---------------------------------------
There is another PR from [~schofielaj] being discussed:
[https://github.com/apache/kafka/pull/23499.] I think this can be closed today.
> Share consumer leaks ShareCompletedFetch when two fetches for the same
> partition are merged before collection
> -------------------------------------------------------------------------------------------------------------
>
> Key: KAFKA-21106
> URL: https://issues.apache.org/jira/browse/KAFKA-21106
> Project: Kafka
> Issue Type: Bug
> Reporter: Apoorv Mittal
> Assignee: Apoorv Mittal
> Priority: Blocker
> Fix For: 4.4.0
>
>
> ShareConsumeRequestManager.fetch() fetches on every poll() regardless of
> whether the previous fetch response for a partition was collected, so two
> completed fetches for the same partition can end up queued at once.
> ShareFetchCollector then merges them via ShareInFlightBatch.merge(), which
> copied the absorbed batch's records but never cleared them. That batch's
> ShareCompletedFetch leaks forever in
> ShareFetchBuffer.pendingAcknowledgementFetches, along with its retained
> records.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
