[ 
https://issues.apache.org/jira/browse/YARN-7017?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16127774#comment-16127774
 ] 

Eric Payne commented on YARN-7017:
----------------------------------

Thanks [~saruntek] for raising this issue.

{quote}
As of today the only way to enable preemption at a queue level is:
- Enable cluster level preemption
- Disable preemption on the queues where not required using 
yarn.scheduler.capacity.<queue-path>.disable_preemption to true

{quote}
Actually, you don't need to {{disable_preemption}} for every queue. The 
{{disable_preemption}} property is inherited, so you can:
- Enable cluster level preemption
- Disable preemption on the root queue using 
{{yarn.scheduler.capacity.root.disable_preemption = true}}
- Enable preemption on the queues where required using 
{{yarn.scheduler.capacity.<queue-path>.disable_preemption = true}}

> Enable preemption for a single queue.
> -------------------------------------
>
>                 Key: YARN-7017
>                 URL: https://issues.apache.org/jira/browse/YARN-7017
>             Project: Hadoop YARN
>          Issue Type: New Feature
>          Components: capacity scheduler, yarn
>            Reporter: sarun
>
> *PROBLEM*
> How to enable preemption on a single queue in a cluster?
> *DESCRIPTION*
> As of today the only way to enable preemption at a queue level is:
> * Enable cluster level preemption
> * Disable preemption on the queues where not required using 
> yarn.scheduler.capacity.<queue-path>.disable_preemption to true
> Can we have some sort of a parameter like 
> *_yarn.scheduler.capacity.<queue-path>.enable_preemption_*
> which would just enable preemption per queue instead of going the other way 
> round which is more time consuming and error prone.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to