MENG DING commented on YARN-1651:

bq. I think killing a container in this case is too harsh. Nothing is wrong 
with the container itself. The container can run fine. I think we should revert 
the increased resource and have some way to signal back that the 
increaseRequest failed, maybe in the allocate response ?

Hi, [~jianhe], a sub-task (YARN-4138) has been logged for this issue. Based on 
our current discussion, when an increase token expires, RM will revert the 
increased resource back and also send a decrease event to NM (NM can ignore the 
decrease event if target size of container is equal to existing container 
size). As a bonus, the decrease can be pulled by AM in 
AllocateResponse.decreased_containers which acts as a signal for the expired 
increase action. 

> CapacityScheduler side changes to support increase/decrease container 
> resource.
> -------------------------------------------------------------------------------
>                 Key: YARN-1651
>                 URL: https://issues.apache.org/jira/browse/YARN-1651
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager, scheduler
>            Reporter: Wangda Tan
>            Assignee: Wangda Tan
>         Attachments: YARN-1651-1.YARN-1197.patch, 
> YARN-1651-2.YARN-1197.patch, YARN-1651-3.YARN-1197.patch, 
> YARN-1651-4.YARN-1197.patch

This message was sent by Atlassian JIRA

Reply via email to