[
https://issues.apache.org/jira/browse/FLINK-32339?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17732495#comment-17732495
]
Matthias Pohl commented on FLINK-32339:
---------------------------------------
But I'm still not 100% sure whether it will help: The event handling is,
indeed, not implemented consistently (with the leadership confirmation not
being handled in the leader event operation thread). But we need the lock
anyway to confirm that service isn't closed (by accessing the {{running}}
field) within the lock.
> Align lock usage in DefaultLeaderElectionService
> ------------------------------------------------
>
> Key: FLINK-32339
> URL: https://issues.apache.org/jira/browse/FLINK-32339
> Project: Flink
> Issue Type: Sub-task
> Components: Runtime / Coordination
> Reporter: Matthias Pohl
> Priority: Major
>
> {{DefaultLeaderElectionService}} uses sequential execution through a single
> thread executor for the leader event handling (introduced in FLINK-31838). We
> missed moving the leadership confirmation event into that queue.
> This issue came up in a [PR
> comment|https://github.com/apache/flink/pull/22769#pullrequestreview-1478763256].
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
