[
https://issues.apache.org/jira/browse/SAMZA-1125?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15904193#comment-15904193
]
Xinyu Liu commented on SAMZA-1125:
----------------------------------
Sorry for not being very clear. When I was reviewing the rb for the
ZkJobCoordinator, in the onNewJobModelAvailble(), it will do
containerContorller.stopContainer() to stop the current container and then in
onNewJobModelConfirmed it will invoke containerController.startContainer(). The
problem I see is that stopContainer() doesn't wait for the container being
fully shutdown. It just sets the flag and return. So it's possible that a new
container is started while the old one is still running/shutting down. I think
this is the scenario we want to wait for the old container to be shut down
before starting the new one. Is it correct?
> Container.shutdown() only stops the runloop
> -------------------------------------------
>
> Key: SAMZA-1125
> URL: https://issues.apache.org/jira/browse/SAMZA-1125
> Project: Samza
> Issue Type: New Feature
> Reporter: Xinyu Liu
> Assignee: Navina Ramesh
>
> Currently SamzaContainer.shutdown() only stops the RunLoop. Seems to me all
> the dependencies used by the container are not properly shutdown. E.g. the
> kafka system consumer/producer will not be shut down. In standalone case
> (ZkJobCoordinator.java), the next time controller.startContainer() will
> create the container with all these dependencies again. The Kafka
> producers/consumers will be created again without closing the previous ones.
> Please verify this is a problem and needs to be fixed.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
