[
https://issues.apache.org/jira/browse/YARN-999?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16766317#comment-16766317
]
Íñigo Goiri commented on YARN-999:
----------------------------------
{quote}
If it is an opportunistic container, it will already be killed fast, so I think
you don't need a distinction between guaranteed/opportunistic (you will do
preemption only in the guaranteed after the timeout).
{quote}
Right now there is no plumbing at all so I need to build the whole preemption
from scratch.
Is there a function in the RM which I can call to adjust containers to the
resources?
Otherwise, I will need to go over the containers and selecting which ones to
kill; in this case I need to do the distinction between guaranteed and
opportunistic.
I don't think the NM is doing anything here.
> In case of long running tasks, reduce node resource should balloon out
> resource quickly by calling preemption API and suspending running task.
> -----------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-999
> URL: https://issues.apache.org/jira/browse/YARN-999
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: graceful, nodemanager, scheduler
> Reporter: Junping Du
> Priority: Major
>
> In current design and implementation, when we decrease resource on node to
> less than resource consumption of current running tasks, tasks can still be
> running until the end. But just no new task get assigned on this node
> (because AvailableResource < 0) until some tasks are finished and
> AvailableResource > 0 again. This is good for most cases but in case of long
> running task, it could be too slow for resource setting to actually work so
> preemption could be hired here.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
