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

Posted to issues@flink.apache.org by "Piotr Nowojski (Jira)" <ji...@apache.org> on 2023/06/27 14:25:00 UTC

[jira] [Closed] (FLINK-32316) Duplicated announceCombinedWatermark task maybe scheduled if jobmanager failover

     [ https://issues.apache.org/jira/browse/FLINK-32316?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Piotr Nowojski closed FLINK-32316.
----------------------------------
    Fix Version/s: 1.18.0
                   1.16.3
                   1.17.2
       Resolution: Fixed

merged to master as b2117bed393
merged to release 1.17 as 758fc14fc59
merged to release 1.16 as 780f675430f

Thanks!

> Duplicated announceCombinedWatermark task maybe scheduled if jobmanager failover
> --------------------------------------------------------------------------------
>
>                 Key: FLINK-32316
>                 URL: https://issues.apache.org/jira/browse/FLINK-32316
>             Project: Flink
>          Issue Type: Bug
>    Affects Versions: 1.16.0
>            Reporter: Cai Liuyang
>            Assignee: Cai Liuyang
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 1.18.0, 1.16.3, 1.17.2
>
>
> When we try SourceAlignment feature, we found there will be a duplicated announceCombinedWatermark task will be scheduled after JobManager failover and auto recover job from checkpoint.
> The reason i think is  we should schedule announceCombinedWatermark task during SourceCoordinator::start function not in SourceCoordinator construct function (see  [https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/source/coordinator/SourceCoordinator.java#L149] ), because when jobManager encounter failover and auto recover job, it will create SourceCoordinator twice:
>  * The first one is  when JobMaster is created it will create the DefaultExecutionGraph, this will init the first sourceCoordinator but will not start it.
>  * The Second one is JobMaster call restoreLatestCheckpointedStateInternal method, which will be reset old sourceCoordinator and initialize a new one, but because the first sourceCoordinator is not started(SourceCoordinator will be started before SchedulerBase::startScheduling), so the first SourceCoordinator will not be fully closed.
>  
> And we also found another problem see jira: https://issues.apache.org/jira/browse/FLINK-32362



--
This message was sent by Atlassian Jira
(v8.20.10#820010)