[
https://issues.apache.org/jira/browse/FLINK-30990?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17686485#comment-17686485
]
Zhu Zhu edited comment on FLINK-30990 at 2/9/23 1:21 PM:
---------------------------------------------------------
Thanks for reporting!
This regression should have been caused by some recent improvements to the
scheduler.
This is not a blocker to take 80-90ms to deploy 4000 tasks. As a reference,
before the improvement to the scheduling performance(FLINK-21110), the number
of this benchmark is tens of seconds.
was (Author: zhuzh):
This regression should have been caused by some recent improvements to the
scheduler.
This is not a blocker to take 80-90ms to deploy 4000 tasks. As a reference,
before the improvement to the scheduling performance(FLINK-21110), the number
of this benchmark is tens of seconds. So I think it is acceptable.
> Performance of deployDownstreamTasks.BATCH has worsened
> -------------------------------------------------------
>
> Key: FLINK-30990
> URL: https://issues.apache.org/jira/browse/FLINK-30990
> Project: Flink
> Issue Type: Bug
> Affects Versions: 1.17.0
> Reporter: Martijn Visser
> Priority: Blocker
>
> See
> http://codespeed.dak8s.net:8000/timeline/#/?exe=8&ben=deployDownstreamTasks.BATCH&extr=on&quarts=on&equid=off&env=2&revs=200
> It appears that the trend has now stabilized upwards, from 60-70 ms/op to
> 80-90 ms/op, while less is better.
> Not sure if this has meaningful impact in a production setup, but I would
> like to get this verified.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
