[
https://issues.apache.org/jira/browse/FLINK-32687?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17747533#comment-17747533
]
Weihua Hu commented on FLINK-32687:
-----------------------------------
[~martijnvisser] Thanks for reporting this. This regression is caused by
https://github.com/apache/flink/pull/22913. This PR improves the Failover
performance, The time cost for the STREAMING scene is reduced by 80%, and
reducing the BATCH and STREAMING_EVENLY scenes by 20%.
But there is a certain performance regression in the BatchEvenly scenario, I
think this is acceptable. Because: 1) Batch Evenly is a strategy that is
unlikely to be used in production, batch tasks run for a short time, and
resources can be released when they are finished; 2) This only affects the
failover process, and part of batch tasks (with block shuffling) will not
trigger the global Failover.
> Performance regression on handleGlobalFailureAndRestartAllTasks.BATCH_EVENLY
> since 2023-07-23
> ---------------------------------------------------------------------------------------------
>
> Key: FLINK-32687
> URL: https://issues.apache.org/jira/browse/FLINK-32687
> Project: Flink
> Issue Type: Bug
> Affects Versions: 1.18.0
> Reporter: Martijn Visser
> Priority: Blocker
>
> http://codespeed.dak8s.net:8000/timeline/#/?exe=5&ben=handleGlobalFailureAndRestartAllTasks.BATCH_EVENLY&extr=on&quarts=on&equid=off&env=2&revs=200
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
