[ 
https://issues.apache.org/jira/browse/FLINK-10256?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16615610#comment-16615610
 ] 

陈梓立 commented on FLINK-10256:
-----------------------------

Thanks [~till.rohrmann]! It works for me.

However, maybe a little further than this thread, why {{JobMaster}} try to 
release resource when {{postStop}}? Since we would say that JM has failed, what 
it is responsible to do is quit itself gracefully, which means clear its state. 
But resource managed by {{ResourceManager}} and the decision how to deal with 
allocated resource should be made by the (possible) take over {{JobMaster}}, or 
timeout and released, but not the failing one. Further, such release actions 
may not success since the connection has possibly already broken.

I am afraid that with current {{postStop}} logic we nude the failing 
{{JobMaster}} act as an improper role.

> Port legacy jobmanager test to FILP-6
> -------------------------------------
>
>                 Key: FLINK-10256
>                 URL: https://issues.apache.org/jira/browse/FLINK-10256
>             Project: Flink
>          Issue Type: Improvement
>          Components: Tests
>    Affects Versions: 1.7.0
>            Reporter: 陈梓立
>            Assignee: 陈梓立
>            Priority: Major
>             Fix For: 1.7.0
>
>
> I am planning to rework JobManagerFailsITCase and JobManagerTest into 
> JobMasterITCase and JobMasterHAITCase. That is, reorganize the legacy tests, 
> make them neat and cover cases explicitly. The PR would follow before this 
> weekend.
> While reworking, I'd like to add more jm failover test cases list below, for 
> the further implement of jm failover with RECONCILING state. For "jm 
> failover", I mean a real world failover(like low power or process exit), 
> without calling Flink internal postStop logic or something like it.
> 1. Streaming task with jm failover.
> 2. Streaming task with jm failover concurrent to task fail.
> 3. Batch task with jm failover.
> 4. Batch task with jm failover concurrent to task fail.
> 5. Batch task with jm failover when some vertex has already been FINISHED.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to