[jira] [Commented] (FLINK-20943) AsyncCheckpointRunnable.cleanup() runs in the main task thread

2021-01-13 Thread Roman Khachatryan (Jira) 


[ 
https://issues.apache.org/jira/browse/FLINK-20943?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17263997#comment-17263997
 ] 

Roman Khachatryan commented on FLINK-20943:
---

I think only closing (cleaning up) AsyncCheckpointRunnable needs to be async. 
It is the only potentially long operation and the only one that can be async.

> AsyncCheckpointRunnable.cleanup() runs in the main task thread
> --
>
> Key: FLINK-20943
> URL: https://issues.apache.org/jira/browse/FLINK-20943
> Project: Flink
>  Issue Type: Bug
>  Components: Runtime / Checkpointing
>Affects Versions: 1.12.0, 1.11.3, 1.13.0
>Reporter: Roman Khachatryan
>Assignee: Roman Khachatryan
>Priority: Major
> Fix For: 1.13.0, 1.12.2
>
>
> In FLINK-8871 an RPC notification about checkpoint abortion was added.
> Upon receiving it, task cancels the running state-persisting futures. It does 
> so through the mailbox.
> Cancelling the futures may involve waiting for their result (if they can't be 
> cancelled).
>  
> This essentially makes checkpoint sync from async upon receiving such RPC.
>  
> cc [~yunta], [~pnowojski]



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

[jira] [Commented] (FLINK-20943) AsyncCheckpointRunnable.cleanup() runs in the main task thread

2021-01-12 Thread Yun Tang (Jira) 


[ 
https://issues.apache.org/jira/browse/FLINK-20943?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17263934#comment-17263934
 ] 

Yun Tang commented on FLINK-20943:
--

There actually has several operations in 
{{SubtaskCheckpointCoordinatorImpl#notifyCheckpointAborted}}, what's your plan 
to make several of them as async or just {{AsyncCheckpointRunnable.cleanup()}} 
as async?

> AsyncCheckpointRunnable.cleanup() runs in the main task thread
> --
>
> Key: FLINK-20943
> URL: https://issues.apache.org/jira/browse/FLINK-20943
> Project: Flink
>  Issue Type: Bug
>  Components: Runtime / Checkpointing
>Affects Versions: 1.12.0, 1.11.3, 1.13.0
>Reporter: Roman Khachatryan
>Assignee: Roman Khachatryan
>Priority: Major
> Fix For: 1.13.0, 1.12.2
>
>
> In FLINK-8871 an RPC notification about checkpoint abortion was added.
> Upon receiving it, task cancels the running state-persisting futures. It does 
> so through the mailbox.
> Cancelling the futures may involve waiting for their result (if they can't be 
> cancelled).
>  
> This essentially makes checkpoint sync from async upon receiving such RPC.
>  
> cc [~yunta], [~pnowojski]



--
This message was sent by Atlassian Jira
(v8.3.4#803005)