Sunil G commented on YARN-3216:

Synced up with [~Naganarasimha Garla] offline. 
One of the concern with option2 as such, is the calculation of max-am-resource 
per partition level which will be a subset of Queue itself. Assuming a 
configuration (its a corner case scenario), where queue has 10GB resource with 
0.1 as max-am-resource-percent. Now considering 2 partition sharing this queue 
with 50%, max am resource will be 500MB per partition. Earlier, a single 
application could have got 1Gb (as Naga told, over utilizing from other 
partition). Such case will pop-up as we go with Option2. 
I feel this also can be considered as a point to decide over these 2 options. 
Also I agree that having a non-used DEFAULT-PARTITION sharing am resources may 
also lead issue in future too. So earlier we get rid of that, its more better. 
:). So my suggestion is implementing option 2, with one of partition can borrow 
AM resources from other partition. 

[~leftnoteasy], could you also share your thoughts over here and we can syncup 
offline as needed to discuss this.

> Max-AM-Resource-Percentage should respect node labels
> -----------------------------------------------------
>                 Key: YARN-3216
>                 URL: https://issues.apache.org/jira/browse/YARN-3216
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager
>            Reporter: Wangda Tan
>            Assignee: Sunil G
>            Priority: Critical
>         Attachments: 0001-YARN-3216.patch
> Currently, max-am-resource-percentage considers default_partition only. When 
> a queue can access multiple partitions, we should be able to compute 
> max-am-resource-percentage based on that.

This message was sent by Atlassian JIRA

Reply via email to