[
https://issues.apache.org/jira/browse/YARN-7240?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16178482#comment-16178482
]
Arun Suresh commented on YARN-7240:
-----------------------------------
Thanks for the patch [~kartheek]. The patch looks mostly good.
Some Comments:
* When we re-queue paused containers (lines 276 - 285 of the
ContainerScheduler) see if we can push the container to the head of the queue
* Remove the "TODO: fix transition" comments from the ContainerImpl
* Since we are introducing a new state "REINITIALIZING_AWAITING_KILL", we need
to update the ContainerImpl::getCurrentState() to map it to RUNNING state.
* Existing tests in {{TestContainerManager}} should be able to verify that the
new "REINITIALIZING_AWAITING_KILL" state works fine, but we should add a test
case to {{TestContainer}}
> Add more states and transitions to stabilize the NM Container state machine
> ---------------------------------------------------------------------------
>
> Key: YARN-7240
> URL: https://issues.apache.org/jira/browse/YARN-7240
> Project: Hadoop YARN
> Issue Type: Improvement
> Reporter: Arun Suresh
> Assignee: kartheek muthyala
> Attachments: YARN-7240.001.patch
>
>
> There seem to be a few intermediate states that can be added to improve the
> stability of the NM container state machine.
> For. eg:
> * The REINITIALIZING should probably be split into REINITIALIZING and
> REINITIALIZING_AWAITING_KILL.
> * Container updates are currently handled in the ContainerScheduler, but it
> would probably be better to have it plumbed through the container state
> machine as a new state, say UPDATING and a new container event.
> The plan is to add some extra tests too to try and test every transition.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
