You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Xintong Song (Jira)" <ji...@apache.org> on 2022/05/23 05:33:00 UTC

[jira] [Commented] (FLINK-27737) Remove legacy support for unfenced executor in FencedRpcEndpoint

    [ https://issues.apache.org/jira/browse/FLINK-27737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17540757#comment-17540757 ] 

Xintong Song commented on FLINK-27737:
--------------------------------------

Thanks for pointing this out, [~aitozi]. I think this would be a nice clean-up.

Would you like to work on this?

> Remove legacy support for unfenced executor in FencedRpcEndpoint
> ----------------------------------------------------------------
>
>                 Key: FLINK-27737
>                 URL: https://issues.apache.org/jira/browse/FLINK-27737
>             Project: Flink
>          Issue Type: Technical Debt
>          Components: Runtime / Coordination
>            Reporter: Aitozi
>            Priority: Major
>
> There are 2 main thread executors in {{{}FencedRpcEndpoint{}}}, the fenced- and unfenced-. The fenced executor was available only during a specific leader session, while the unfenced executor was used for actions out of a leader session (e.g., during the initialization). This is no longer needed because currently for all the fenced endpoints ({{{}Dispatcher{}}}, {{JobMaster}} and {{{}ResourceManager{}}}), the leader election finishes before the endpoints are initiated.
> In this ticket we propose to remove the legacy codes for supporting unfenced executor and updating fencing tokens, as well as to simplify the type hierarchy of {{{}FencedRpcEndpoint{}}}.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)