[
https://issues.apache.org/jira/browse/FLINK-17726?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17124808#comment-17124808
]
Till Rohrmann commented on FLINK-17726:
---------------------------------------
Sorry for my late reply. I think we should handle all state transition to
{{CANCELED}} from a different state than {{CANCELLING}} on the {{JobMaster}} as
a failure. This effectively means that the {{JobMaster}} must initiate the
cancellation in one form or another. If we wanted the {{Task}} to be smart and
to initiate the cancellation, then it would have to be sure that there is
another {{Task}} which reported a failure back to the {{JobMaster}}. I think in
the general case this is very hard to guarantee (only if the other task sends a
message that it successfully transmitted this state transition to the
{{JobMaster}} it would be ok).
Consequently, if there is the situation of {{A1 -> B1}} and {{A1}} fails and
{{B1}} realizes it, then {{B1}} cannot be sure that {{A1}} could update the
{{JM}} and has to fail. Only if the {{JM}} sent the cancellation request it
knows that the failure of {{A1}} has been successfully reported and it can
cancel.
One way to ensure this contract on the {{Task}} side could be to only allow
state transition from {{CANCELLING}} to {{CANCELED}}. Concretely, this means
that we transition into {{FAILED}} if we see a {{CancelTaskException}} if the
current {{Task}} state is not {{CANCELLING}}.
Does this make sense?
> Scheduler should take care of tasks directly canceled by TaskManager
> --------------------------------------------------------------------
>
> Key: FLINK-17726
> URL: https://issues.apache.org/jira/browse/FLINK-17726
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination, Runtime / Task
> Affects Versions: 1.11.0, 1.12.0
> Reporter: Zhu Zhu
> Assignee: Nicholas Jiang
> Priority: Critical
> Fix For: 1.11.0, 1.12.0
>
>
> JobManager will not trigger failure handling when receiving CANCELED task
> update.
> This is because CANCELED tasks are usually caused by another FAILED task.
> These CANCELED tasks will be restarted by the failover process triggered
> FAILED task.
> However, if a task is directly CANCELED by TaskManager due to its own runtime
> issue, the task will not be recovered by JM and thus the job would hang.
> This is a potential issue and we should avoid it.
> A possible solution is to let JobManager treat tasks transitioning to
> CANCELED from all states except from CANCELING as failed tasks.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
