[ 
https://issues.apache.org/jira/browse/YARN-5892?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15881614#comment-15881614
 ] 

Wangda Tan commented on YARN-5892:
----------------------------------

Hi [~eepayne],

Apologize for my late responses, finally get some time to look at this.

I'm a little concerned about semantics of this feature. (I've to admit that 
existing user-limit semantics is not perfect first) For existing MULP, to me it 
has two guarantees:
#1, if there're N (N <= 100 / MULP) users are consuming resource in a queue, 
each of them can get at least MULP / 100 * queue-configured-capacity.
#2, when doing preemption, if an user uses <= MULP / 100 * 
queue-configured-capacity, we will not preempt resource from such users.

If we introduced this feature, per my understanding, #1 will be break. For 
example, a queue with MULP = 33, and user=admin has MULP = 75. Assume we have 3 
users: Jack, Alice, Admin are running jobs in the queue, there's no guarantee 
that any user can use at much as their MULP.

To make a more clear semantics, in my mind there're some alternative solutions:
a. Create queue just for such vip users, and we can plan resources, (for 
example root.users.admin = 30%, and root.users.others = 50%).
b. In addition to that, special queue mapping rules can be implemented based on 
YARN-3635, which can do more flexible queue mappings for use cases like vip 
users. 

Please let me know you thoughts.

> Capacity Scheduler: Support user-specific minimum user limit percent
> --------------------------------------------------------------------
>
>                 Key: YARN-5892
>                 URL: https://issues.apache.org/jira/browse/YARN-5892
>             Project: Hadoop YARN
>          Issue Type: Improvement
>          Components: capacityscheduler
>            Reporter: Eric Payne
>            Assignee: Eric Payne
>         Attachments: YARN-5892.001.patch, YARN-5892.002.patch
>
>
> Currently, in the capacity scheduler, the {{minimum-user-limit-percent}} 
> property is per queue. A cluster admin should be able to set the minimum user 
> limit percent on a per-user basis within the queue.
> This functionality is needed so that when intra-queue preemption is enabled 
> (YARN-4945 / YARN-2113), some users can be deemed as more important than 
> other users, and resources from VIP users won't be as likely to be preempted.
> For example, if the {{getstuffdone}} queue has a MULP of 25 percent, but user 
> {{jane}} is a power user of queue {{getstuffdone}} and needs to be guaranteed 
> 75 percent, the properties for {{getstuffdone}} and {{jane}} would look like 
> this:
> {code}
>   <property>
>     
> <name>yarn.scheduler.capacity.root.getstuffdone.minimum-user-limit-percent</name>
>     <value>25</value>
>   </property>
>   <property>
>     
> <name>yarn.scheduler.capacity.root.getstuffdone.jane.minimum-user-limit-percent</name>
>     <value>75</value>
>   </property>
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org

Reply via email to