You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@flink.apache.org by "Zhu Zhu (Jira)" <ji...@apache.org> on 2022/12/08 09:16:00 UTC

[jira] [Commented] (FLINK-28982) Start TaskInterrupter when task switch from DEPLOYING to CANCELING

    [ https://issues.apache.org/jira/browse/FLINK-28982?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17644688#comment-17644688 ] 

Zhu Zhu commented on FLINK-28982:
---------------------------------

This is an intentional behavior because Flink does not want to interrupt actions originates from Flink framework, to avoid unexpected problems. The interruption only targets for user code. 
The task cancellation stuck in Flink code, this means a bug to fix.
For this specific case, to solve it, I think we should fix the problem of the Remote Shuffle Service.

> Start TaskInterrupter when task switch from DEPLOYING to CANCELING
> ------------------------------------------------------------------
>
>                 Key: FLINK-28982
>                 URL: https://issues.apache.org/jira/browse/FLINK-28982
>             Project: Flink
>          Issue Type: Technical Debt
>          Components: Runtime / Web Frontend
>            Reporter: LI Mingkun
>            Priority: Major
>         Attachments: image-2022-08-16-12-10-43-894.png
>
>
> Task will start TaskInterrupter only when `ExecutionState` is INITIALIZING or RUNNING in the function: org.apache.flink.runtime.taskmanager.Task#cancelOrFailAndCancelInvokableInternal
>  
> I got a dead lock in multi task which caused by Flink Remote Shuffle's sharing TCP connection bug and blocked tasks destruction when I use Flink Remote Shuffle.
> stack as following:
> !image-2022-08-16-12-10-43-894.png!
> My question: Why not start the TaskInterrupter when cancel a deploying task?
>  



--
This message was sent by Atlassian Jira
(v8.20.10#820010)