Carlo Curino commented on YARN-4003:

The way I think about {{RM scheduling bandwidth}} is like yet another 
_orthogonal_ resource (such as CPU/memory) that you control per-queue. So yes, 
you can define a max-capacity for all resources, including {{RM scheduling 
bandwidth}}. But to be clear, you can have lots of CPU/memory and little RM 
scheduling bandwidth (e.g., for services), or viceversa (e.g., for queues 
running many small jobs with lots of short-lived tasks). So the limits on 
CPU/memory should not in principle affect RM scheduling bandwith. Practically, 
you will need to devote at least some of your CPU/memory to run AMs . Again  
this is less true as we move to federation (YARN-2915) as there can be one RM 
running the AM, but many other RMs will see this as unmanaged AMs. 

> ReservationQueue inherit getAMResourceLimit() from LeafQueue, but behavior is 
> not consistent
> --------------------------------------------------------------------------------------------
>                 Key: YARN-4003
>                 URL: https://issues.apache.org/jira/browse/YARN-4003
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager
>            Reporter: Carlo Curino
>            Assignee: Carlo Curino
>         Attachments: YARN-4003.patch
> The inherited behavior from LeafQueue (limit AM % based on capacity) is not a 
> good fit for ReservationQueue (that have highly dynamic capacity). 

This message was sent by Atlassian JIRA

Reply via email to