Maximilian Michels commented on FLINK-4555:

There has been some refactoring previously. However, I couldn't find that the 
feature has been removed. The cluster shuts down cleanly in the following order:

1. Client requests shutdown 
2. JobManager receives StopCluster message 
3. JobManager sends StopCluster message to all connected TaskManager
  3.1 TaskManagers shut down and start a ProcessShutDownThread which kills 
after 10 seconds. 
4. JobManager sends StopCluster message to the ResourceManager
  4.1 ResourceManager unregisters as application master
5. JobManager shuts down, informs the client, and starts a 
ProcessShutDownThread which kills after 10 seconds. 

Still, a good idea to back this with a test.

> Explicitly kill TaskManager on YARN when ApplicationMaster is shutting down
> ---------------------------------------------------------------------------
>                 Key: FLINK-4555
>                 URL: https://issues.apache.org/jira/browse/FLINK-4555
>             Project: Flink
>          Issue Type: Bug
>          Components: YARN Client
>    Affects Versions: 1.1.0
>            Reporter: Robert Metzger
>            Assignee: Maximilian Michels
>            Priority: Critical
> It seems that Flink is not explicitly destroying the TaskManager JVM when the 
> ApplicationMaster is shutting down (when the YARN application is stopping).
> Since this was once in Flink (in 1.0.x) we should add a test case to ensure 
> this feature stays in the code.

This message was sent by Atlassian JIRA

Reply via email to