[
https://issues.apache.org/jira/browse/YARN-2001?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13990241#comment-13990241
]
Karthik Kambatla commented on YARN-2001:
----------------------------------------
bq. we may run into condition like the resource usage, capacity limit (e.g.
headroom, queue capacity etc. ) in scheduler is not yet correct until all the
nodes sync back all the running containers belong to the app,
applications/queues can potentially go beyond its limit.
My understanding has been that the RM's scheduler starts from scratch on
restart/failover and rebuilds its state as nodes heartbeat. At any point in
time, the cluster's resources correspond only to the NMs that have registered
with the "new" RM. IOW, this should be no different from a new cluster. Given
this, I am not sure how the scheduler can have incorrect information.
> Persist NMs info for RM restart
> -------------------------------
>
> Key: YARN-2001
> URL: https://issues.apache.org/jira/browse/YARN-2001
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: resourcemanager
> Reporter: Jian He
> Assignee: Jian He
>
> RM should not accept allocate requests from AMs until all the NMs have
> registered with RM. For that, RM needs to remember the previous NMs and wait
> for all the NMs to register.
> This is also useful for remembering decommissioned nodes across restarts.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
