[
https://issues.apache.org/jira/browse/RATIS-2138?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17872979#comment-17872979
]
Song Ziyang commented on RATIS-2138:
------------------------------------
It seems like the Precondition helps catch a race condition. Shall we rearrange
the order of updateMetadata and shutdownLeaderState to avoid that race?
> Remove uncessary error log in JVMPauseMonitor When GC is serious
> ----------------------------------------------------------------
>
> Key: RATIS-2138
> URL: https://issues.apache.org/jira/browse/RATIS-2138
> Project: Ratis
> Issue Type: Improvement
> Components: util
> Reporter: Xinyu Tan
> Assignee: Xinyu Tan
> Priority: Major
> Attachments: image-2024-08-12-14-13-05-825.png,
> image-2024-08-12-14-14-45-615.png, image-2024-08-12-14-17-13-030.png
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> JVMPauseMonitor may be printing unexpected Error logs when GC is heavy and
> the cluster keeps triggering re-election.
> !image-2024-08-12-14-13-05-825.png!
> After investigating the relevant code, it is found that the Preconditions
> here are unreasonable. When we update, we will first update the serverstate's
> term and then shutdown LeaderState. If any other thread calls leaderState's
> getCurrentTerm in the meantime, this will cause an error, so it's not
> necessary at this point
> !image-2024-08-12-14-14-45-615.png!
> !image-2024-08-12-14-17-13-030.png!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
