Siqi Li updated YARN-1735:
    Attachment: YARN-1735.v4.patch

> For FairScheduler AvailableMB in QueueMetrics is the same as AllocateMB
> -----------------------------------------------------------------------
>                 Key: YARN-1735
>                 URL: https://issues.apache.org/jira/browse/YARN-1735
>             Project: Hadoop YARN
>          Issue Type: Bug
>          Components: fairscheduler
>            Reporter: Siqi Li
>         Attachments: YARN-1735.v1.patch, YARN-1735.v2.patch, 
> YARN-1735.v3.patch, YARN-1735.v4.patch
> in monitoring graphs the AvailableMB of each queue regularly spikes between 
> the AllocatedMB and the entire cluster capacity.
> This cannot be correct since AvailableMB should never be more than the queue 
> max allocation. The spikes are quite confusing since the availableMB is set 
> as the fair share of each queue and the fair share of each queue is bond by 
> their allowed max resource.
> Other than the spiking, the availableMB is always equal to allocatedMB. I 
> think this is not very useful, availableMB for each queue should be their 
> allowed max resource minus allocatedMB.

This message was sent by Atlassian JIRA

Reply via email to