You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@beam.apache.org by "Jiayi Zhao (JIRA)" <ji...@apache.org> on 2019/04/01 21:42:00 UTC
[jira] [Comment Edited] (BEAM-6956)
--experiments=worker_threads=100 issue
[ https://issues.apache.org/jira/browse/BEAM-6956?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16807065#comment-16807065 ]
Jiayi Zhao edited comment on BEAM-6956 at 4/1/19 9:41 PM:
----------------------------------------------------------
it might due to some random stuff, I tried in thread flink again with 100 worker_threads, and this time it works(I also changed -parallelism=1 instead of -parallelism=2 before, not sure if it's related),
but when I tried separate local flink cluster, it works for some simple pipeline but got stuck for some complex tensorflow transform pipelines
was (Author: 1025kb):
it might due to some random stuff, I tried in thread flink again with 100 worker_threads, and this time it works(I also changed --parallelism=1 instead of --parallelism=2 before, not sure if it's related),
when I tried separate local flink cluster, each time pipeline stuck at different places, I will try confirm it and put more information about where it stucks
> --experiments=worker_threads=100 issue
> --------------------------------------
>
> Key: BEAM-6956
> URL: https://issues.apache.org/jira/browse/BEAM-6956
> Project: Beam
> Issue Type: New Feature
> Components: runner-flink
> Reporter: Jiayi Zhao
> Priority: Major
>
> I noticed that without this "–experiments=worker_threads=100", pipeline will stuck,
> The weird thing is, I tried some complex pipeline using the in thread flink method (./gradlew :beam-runners-flink_2.11-job-server:runShadow)
> "–experiments=worker_threads=100" doesn't work, but "–experiments=worker_threads=1000" works fine
> Then I tried the same pipeline using the separate local flink cluster (./gradlew :beam-runners-flink_2.11-job-server:runShadow -PflinkMasterUrl=localhost:8081), flink version is 1.5.6 (other version doesn't work, see BEAM-6915)
> Neither did "–experiments=worker_threads=1000" or "–experiments=worker_threads=10000" work, pipeline stuck at certain stage (shows running in flink UI but won't finish forever)
> any real fix to that? Thanks!
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)