[
https://issues.apache.org/jira/browse/FLINK-37642?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17945217#comment-17945217
]
Renxiang Zhou edited comment on FLINK-37642 at 4/17/25 3:10 AM:
----------------------------------------------------------------
[~zakelly]
Yes, I agree with you very much, but the purpose of adding this feat is to
complete the callback generated by the asynchronous framework as soon as
possible. However, the priority of these callbacks is the same as the timer,
which causes the timer maybe triggered earlier than the asynchronous callback,
so it seems this implementation has not achieved the expected effect?
was (Author: JIRAUSER295459):
[~zakelly]
Yes, I agree with you very much, but the purpose of adding this function is to
complete the callback generated by the asynchronous framework as soon as
possible, but now the priority of these callbacks is the same as the timer,
which causes the timer to be triggered earlier than the asynchronous callback.
I feel that it has not achieved the expected effect.
> Use the mailboxExecutor for operator in Async state processing
> --------------------------------------------------------------
>
> Key: FLINK-37642
> URL: https://issues.apache.org/jira/browse/FLINK-37642
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Async State Processing
> Affects Versions: 2.0.0
> Reporter: Zakelly Lan
> Assignee: Yanfei Lei
> Priority: Major
> Labels: pull-request-available
> Fix For: 2.1.0
>
>
> Currently, the operators of async state processing use the main mailbox
> executor for callback execution, which has low priority. It's better to use
> the one for `YieldingOperator`, as:
> {code:java}
> MailboxExecutor mailboxExecutor =
> containingTask
> .getMailboxExecutorFactory()
> .createExecutor(configuration.getChainIndex());
> {code}
> It has higher priority, which could help us do efficient `yield()`.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
