Wangda Tan commented on YARN-3265:

Thanks for your comments, addressed all of them, and:

bq. We should think more about the right place for resourceLimits, in following 
JIRAs. I'd like to avoid mixing real state of the queue with a single 
scheduling cycle state. This is a general problem with other things like 
LeafQueue.QueueResourceLimitsInfo. One simple solution is to have a 
AbstractQueue.schedulingCycleState and wrap everything in there.
Yes, I think we should refactor and put related fields together as a part of 
SchedulingCycleState to avoid using/updating them everywhere.

bq. Not caused by the patch, but the application headroom going into metrics is 
different from is different from the one pulled by apps.
Already filed YARN-3279 and several other issues when working on this JIRA 
under YARN-1198.

Uploaded new patch.

> CapacityScheduler deadlock when computing absolute max avail capacity (fix 
> for trunk/branch-2)
> ----------------------------------------------------------------------------------------------
>                 Key: YARN-3265
>                 URL: https://issues.apache.org/jira/browse/YARN-3265
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: capacityscheduler, resourcemanager
>            Reporter: Wangda Tan
>            Assignee: Wangda Tan
>            Priority: Blocker
>         Attachments: YARN-3265.1.patch, YARN-3265.2.patch, YARN-3265.3.patch, 
> YARN-3265.5.patch
> This patch is trying to solve the same problem described in YARN-3251, but 
> this is a longer term fix for trunk and branch-2.

This message was sent by Atlassian JIRA

Reply via email to