Wangda Tan commented on YARN-1198:

Hi [~cwelch],
Thanks for updating, I went through your patch just now.

I think the current approach makes more sense to me comparing to patch#4, it 
avoids iterating all apps when computing headroom. But currently, 
CapacityHeadroomProvider#getHeadroom will recompute headroom for each 
application heartbeat. Assume we have #application >> #user in a queue (the 
most possible case), it's still a little costly.

I agree with the method which mentioned by Jason more: Specifically, we can 
create a map of <user, headroom> for each queue, when we need update headroom, 
we can update the all headroom in the map. And each SchedulerApplicationAttempt 
will hold a reference to headroom. The "headroom" in the map maybe as same as 
the {{HeadroomProvider}} in your patch. I would suggest to rename the 
{{HeadroomProvider}} to {{HeadroomReference}}, because we don't need do any 
computation in it anymore.

Another benefit is, we don't need write HeadroomProvider for each scheduler. A 
simple HeadroomReference with getter/setter should be enough.

Two more things we should take care with previous method:
1) As mentioned by Jason, currently, fair/capacity scheduler all support moving 
app between queues, we should recompute and change the reference after finished 
moving app. 
2) In LeafQueue#assignContainers, we don't need call 
  Resource userLimit = 
      computeUserLimitAndSetHeadroom(application, clusterResource, 
For each application, and in LeafQueue#updateClusterResource iterate and update 
the map of <user, headroom> should be enough


> Capacity Scheduler headroom calculation does not work as expected
> -----------------------------------------------------------------
>                 Key: YARN-1198
>                 URL: https://issues.apache.org/jira/browse/YARN-1198
>             Project: Hadoop YARN
>          Issue Type: Bug
>            Reporter: Omkar Vinit Joshi
>            Assignee: Craig Welch
>         Attachments: YARN-1198.1.patch, YARN-1198.2.patch, YARN-1198.3.patch, 
> YARN-1198.4.patch, YARN-1198.5.patch, YARN-1198.6.patch, YARN-1198.7.patch
> Today headroom calculation (for the app) takes place only when
> * New node is added/removed from the cluster
> * New container is getting assigned to the application.
> However there are potentially lot of situations which are not considered for 
> this calculation
> * If a container finishes then headroom for that application will change and 
> should be notified to the AM accordingly.
> * If a single user has submitted multiple applications (app1 and app2) to the 
> same queue then
> ** If app1's container finishes then not only app1's but also app2's AM 
> should be notified about the change in headroom.
> ** Similarly if a container is assigned to any applications app1/app2 then 
> both AM should be notified about their headroom.
> ** To simplify the whole communication process it is ideal to keep headroom 
> per User per LeafQueue so that everyone gets the same picture (apps belonging 
> to same user and submitted in same queue).
> * If a new user submits an application to the queue then all applications 
> submitted by all users in that queue should be notified of the headroom 
> change.
> * Also today headroom is an absolute number ( I think it should be normalized 
> but then this is going to be not backward compatible..)
> * Also  when admin user refreshes queue headroom has to be updated.
> These all are the potential bugs in headroom calculations

This message was sent by Atlassian JIRA

Reply via email to