[
https://issues.apache.org/jira/browse/YARN-4597?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16179233#comment-16179233
]
Jason Lowe commented on YARN-4597:
----------------------------------
Sorry to arrive late here, but this has backwards-compatibility issues. This
change introduces a new container state enumeration and sends that new state to
clients that may not be aware of that state since they're not running the new
version. That can cause those clients to either crash or make incorrect
assumptions, e.g.: state != RUNNING must be COMPLETED, etc.
Ran across this because YARN-6570 has that same sort of issue. That one has
the RM not expecting the new value in 2.8, but any AM asking the NM for
container status via NMClient could have the same kind of issue. The NM should
not be sending new enumeration values to old clients. Filed YARN-7248 to track
the compatibility issue.
> Introduce ContainerScheduler and a SCHEDULED state to NodeManager container
> lifecycle
> -------------------------------------------------------------------------------------
>
> Key: YARN-4597
> URL: https://issues.apache.org/jira/browse/YARN-4597
> Project: Hadoop YARN
> Issue Type: New Feature
> Components: nodemanager
> Reporter: Chris Douglas
> Assignee: Arun Suresh
> Labels: oct16-hard
> Fix For: 2.9.0, 3.0.0-alpha2
>
> Attachments: YARN-4597.001.patch, YARN-4597.002.patch,
> YARN-4597.003.patch, YARN-4597.004.patch, YARN-4597.005.patch,
> YARN-4597.006.patch, YARN-4597.007.patch, YARN-4597.008.patch,
> YARN-4597.009.patch, YARN-4597.010.patch, YARN-4597.011.patch,
> YARN-4597.012.patch, YARN-4597.013.patch
>
>
> Currently, the NM immediately launches containers after resource
> localization. Several features could be more cleanly implemented if the NM
> included a separate stage for reserving resources.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
