You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Kostas Kloudas (Jira)" <ji...@apache.org> on 2020/10/10 14:09:00 UTC
[jira] [Comment Edited] (FLINK-19358) when submit job on
application mode with HA,the jobid will be 0000000000
[ https://issues.apache.org/jira/browse/FLINK-19358?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17211710#comment-17211710 ]
Kostas Kloudas edited comment on FLINK-19358 at 10/10/20, 2:08 PM:
-------------------------------------------------------------------
[~fly_in_gis] I agree with you that for single-execute applications, relying on the {{JobGraphStore}} and not having a fixed job id seems to be a solution and I had the same thought when working on the Application Mode. The pre-existing per-job implementations used to do it so I went with it. Maybe [~trohrmann] has the reason why this was not done before.
For multi-execute applications (if we want to support them in the future with HA) I am not sure if this can work because we may have some job graphs in the application that were finished for example before the job manager failure. But I need to think this a bit more.
For adding the cluster-id always to the {{state.checkpoints.dir}}, I think this can work.
[~trohrmann] For the {{execute()}} with JobId solution, we can have a "verification step" in the {{executeAsync()}} where we check if the user has set the JobId against the specified {{executor}}.
was (Author: kkl0u):
[~fly_in_gis] I agree with you that for single-execute applications, relying on the {{JobGraphStore}} and not having a fixed job id seems to be a solution and I had the same thought when working on the Application Mode. The pre-existing per-job implementations used to do it so I went with it. Maybe [~trohrmann] has the reason why this was not done before.
For multi-execute applications (if we want to support them in the future with HA) I am not sure if this can work because we need to further map retrieved JobIds to job graphs and we may have job graphs that were finished for example. But I think I need to think this a bit more.
For adding the cluster-id always to the {{state.checkpoints.dir}}, I think this can work.
[~trohrmann] For the {{execute()}} with JobId solution, we can have a "verification step" in the {{executeAsync()}} where we check if the user has set the JobId against the specified {{executor}}.
> when submit job on application mode with HA,the jobid will be 0000000000
> ------------------------------------------------------------------------
>
> Key: FLINK-19358
> URL: https://issues.apache.org/jira/browse/FLINK-19358
> Project: Flink
> Issue Type: Bug
> Components: Deployment / YARN
> Affects Versions: 1.11.0
> Reporter: Jun Zhang
> Priority: Major
> Labels: usability
> Fix For: 1.12.0
>
>
> when submit a flink job on application mode with HA ,the flink job id will be 00000000000000000000000000000000, when I have many jobs ,they have the same job id , it will be lead to a checkpoint error
--
This message was sent by Atlassian Jira
(v8.3.4#803005)