Wangda Tan commented on YARN-3223:

Thanks for working on this JIRA, I took a look at the patch, some comments: 

I think the general approach is fine, SchedulerNode keep updating total 
resource to used resource if it is decommissioning state, so available resource 
will be 0.

But I think there're some other places need to take care:
- Suggest to use {{CapacityScheduler#updateNodeAndQueueResource}} to update 
resources, we need to update queue's resource, cluster metrics as well.
- When async scheduling enabled, we need to make sure decommissioing node's 
total resource is updated so no new container will be allocated on these nodes.

And after this patch, I think we need to add total decommisioning nodes 
resources to cluster metrics (equals to sum(decommisioning-node.used-resource)).

> Resource update during NM graceful decommission
> -----------------------------------------------
>                 Key: YARN-3223
>                 URL: https://issues.apache.org/jira/browse/YARN-3223
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: nodemanager, resourcemanager
>    Affects Versions: 2.7.1
>            Reporter: Junping Du
>            Assignee: Brook Zhou
>         Attachments: YARN-3223-v0.patch, YARN-3223-v1.patch
> During NM graceful decommission, we should handle resource update properly, 
> include: make RMNode keep track of old resource for possible rollback, keep 
> available resource to 0 and used resource get updated when
> container finished.

This message was sent by Atlassian JIRA

Reply via email to