flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-5934) Scheduler in ExecutionGraph null if failure happens in ExecutionGraph.restoreLatestCheckpointedState
Date Wed, 01 Mar 2017 13:48:45 GMT

    [ https://issues.apache.org/jira/browse/FLINK-5934?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15890186#comment-15890186
] 

ASF GitHub Bot commented on FLINK-5934:
---------------------------------------

GitHub user tillrohrmann opened a pull request:

    https://github.com/apache/flink/pull/3440

    [backport-1.2] [FLINK-5934] Set the Scheduler in the ExecutionGraph via its constructor

    This is a backport of #3437.
    
    This PR is based on #3439.
    
    Before the scheduler was set when calling ExecutionGraph.scheduleForExecution(). This
    has the disadvantage that the ExecutionGraph has not scheduler set if something else
    went wrong before the scheduleForExecution call. Consequently, the job will be stuck
    in a restart loop because the recovery will fail if there is no Scheduler set. In
    order to solve the problem, the Scheduler is not passed to the ExecutionGraph when
    it is created.

You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/tillrohrmann/flink fixExecutionGraphSchedulerBp1.2

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/flink/pull/3440.patch

To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:

    This closes #3440
    
----
commit 045e2187f6d43679115909b10aaeb78c627b758c
Author: Till Rohrmann <trohrmann@apache.org>
Date:   2017-02-28T16:27:03Z

    [FLINK-5938] Replace ExecutionContext by Executor in Scheduler
    
    In order to remove the Scheduler's dependency on Scala's ExecutionContext and
    Akka's futures, this PR replaces the ExecutionContext by an Executor which is
    used to execute the concurrent handleNewSlot call.

commit 2525fb2a0e08418f09319568f58ed7daa9284376
Author: Till Rohrmann <trohrmann@apache.org>
Date:   2017-02-28T14:20:47Z

    [FLINK-5934] Set the Scheduler in the ExecutionGraph via its constructor
    
    Before the scheduler was set when calling ExecutionGraph.scheduleForExecution(). This
    has the disadvantage that the ExecutionGraph has not scheduler set if something else
    went wrong before the scheduleForExecution call. Consequently, the job will be stuck
    in a restart loop because the recovery will fail if there is no Scheduler set. In
    order to solve the problem, the Scheduler is not passed to the ExecutionGraph when
    it is created.

----


> Scheduler in ExecutionGraph null if failure happens in ExecutionGraph.restoreLatestCheckpointedState
> ----------------------------------------------------------------------------------------------------
>
>                 Key: FLINK-5934
>                 URL: https://issues.apache.org/jira/browse/FLINK-5934
>             Project: Flink
>          Issue Type: Bug
>          Components: Distributed Coordination
>    Affects Versions: 1.2.0, 1.1.4, 1.3.0
>            Reporter: Till Rohrmann
>            Assignee: Till Rohrmann
>
> If {{ExecutionGraph.restoreLatestCheckpointedState}} fails with an exception, then all
subsequent recoveries will fail because the {{scheduler}} has not been set in the {{ExecutionGraph}}.
> I propose to set the {{scheduler}} when the {{ExecutionGraph}} is created to avoid this
problem.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message