Craig Welch commented on YARN-1198:

[~john.jian.fang] I look a look at implementing the change with the tweaked .7 
approach per your suggestion above and it seemed to just be trading some 
complexities for others, so I set it aside and I think the current .7 approach 
is as good as any.  I uploaded a .10 patch which is the .7 fixed to apply 
cleanly to current trunk (.7 no longer quite does for me).  I took a look at 
incorporating [YARN-1857] into this change but chose not to, as I think they 
should be committed independently.  The .10 (.7) patch factors the change for 
[YARN-1857] up into a different method, getHeadroom(), if you replace it with 
the below:

private Resource getHeadroom(User user, Resource queueMaxCap,
      Resource clusterResource, Resource userLimit) {
    Resource headroom = 
      Resources.min(resourceCalculator, clusterResource,
            Resources.min(resourceCalculator, clusterResource, 
                userLimit, queueMaxCap), 
        Resources.subtract(queueMaxCap, usedResources));
    return headroom;
then you should have the combined logic.  Note, the LeafQueue tests will then 
not all pass, I believe because results changed when that patch was applied - 
I've not before tried the two in combination, assuming we would apply one at a 
time, and then address the impact on the other.

> 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-1198.10.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, YARN-1198.8.patch, YARN-1198.9.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