[
https://issues.apache.org/jira/browse/RATIS-2161?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17884082#comment-17884082
]
yuuka edited comment on RATIS-2161 at 9/24/24 3:44 AM:
-------------------------------------------------------
In our test, the server was directly stopped, for example, kill -9.
Root cause
I have described in [RATIS-2162] When closing leaderState, if the logAppender
thread sends a snapshot, a deadlock may occur - ASF JIRA (apache.org)
was (Author: JIRAUSER306226):
In our test, the server was directly stopped, for example, kill -9
> Grpc may spawn many threads
> ---------------------------
>
> Key: RATIS-2161
> URL: https://issues.apache.org/jira/browse/RATIS-2161
> Project: Ratis
> Issue Type: Wish
> Components: gRPC
> Reporter: yuuka
> Priority: Major
> Attachments: image-2024-09-23-10-18-27-800.png,
> image-2024-09-23-10-23-55-154.png, image-2024-09-23-10-28-52-065.png
>
>
> Under high load, if there are many RaftGroup leaders on one machine and there
> is a sudden crash, grpc may create many threads to handle requestVote and
> appendEntry, etc.N, frequent thread creation and context switching, which may
> lead to performance degradation.
>
> !image-2024-09-23-10-18-27-800.png!
>
> !image-2024-09-23-10-23-55-154.png!
> !image-2024-09-23-10-28-52-065.png!
> It seems that grpc will not release the thread yet.
> We can specify a fixed number of thread pools or periodically recycle threads
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
