[ 
https://issues.apache.org/jira/browse/SAMZA-130?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Chris Riccomini updated SAMZA-130:
----------------------------------

    Attachment: SAMZA-130.0.patch

Attaching a patch. RB at: https://reviews.apache.org/r/17155/

Changes:

1. Move all start* methods inside try/catch block in SamzaContainer.
2. Clean up a little formatting in SamzaContainer.
3. Add a test to verify TaskInstance.close is called when a failure occurs 
during the init phase.

One thing I want to call out is that this means that stop/close methods can be 
called even though their corresponding start/init methods weren't. I think this 
is OK.

> Exception in init causes wedged SamzaContainers
> -----------------------------------------------
>
>                 Key: SAMZA-130
>                 URL: https://issues.apache.org/jira/browse/SAMZA-130
>             Project: Samza
>          Issue Type: Bug
>          Components: container
>    Affects Versions: 0.6.0
>            Reporter: Chris Riccomini
>            Assignee: Chris Riccomini
>         Attachments: SAMZA-130.0.patch
>
>
> If a failure occurs anywhere in the init phase of the SamzaContainer, the 
> exception is never caught, and shutdown is never called on the task, or any 
> of the consumers/producers.
> This causes the SamzaContainer to hang if there are non-daemon threads 
> running. We should make a best-effort attempt to catch exceptions during the 
> init phase, and call shutdown on everything before exiting. 
> We should also take care to exit with a non-zero exit code when an exception 
> is thrown and caught, so the YARN AM properly figures out that the container 
> failed, and didn't complete successfully.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Reply via email to