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

Posted to issues@flink.apache.org by "Anton Kalashnikov (Jira)" <ji...@apache.org> on 2022/06/24 08:15:00 UTC

[jira] [Created] (FLINK-28241) Massive regression on 20.06.2021

Anton Kalashnikov created FLINK-28241:
-----------------------------------------

             Summary: Massive regression on 20.06.2021
                 Key: FLINK-28241
                 URL: https://issues.apache.org/jira/browse/FLINK-28241
             Project: Flink
          Issue Type: Bug
            Reporter: Anton Kalashnikov


In FLINK-27921 added property `REQUIREMENTS_CHECK_DELAY` that optimizes the number of checks of resources but it also leads to a little delay during the init phase which is totally not critical for the normal scenario. But unfortunately, our benchmarks can not exclude the time of the initial phase(or shutdown phase) from the total time of benchmarks. As result, all benchmarks, that have a relatively short execution time(compare to the initial phase time), have regression(up to 50%). As a quick solution, it makes sense just to set `REQUIREMENTS_CHECK_DELAY` to zero for all benchmarks.
Not full list of affected benchmarks:
 * asyncWait
 * globalWindow
 * mapSink
 * readFileSplit
 * remoteRebalance
 * serializer*
 * sorted*
 * twoInputOneIdleMapSink



--
This message was sent by Atlassian Jira
(v8.20.7#820007)