Junping Du commented on YARN-41:

bq. In the case of work-preserving NM restart (or under supervision as 
YARN-2331 calls it), we can make the NM not do an unregister?
I think latest patch (-4.patch) already did this, but my concern is a little 
broader: does user (or management tools for YARN cluster, like: Ambari) expect 
the same behavior for kill -9 on NM daemon and shutdown for NM daemon? With 
current patch (assume NM work preserving is disabled), user will find RM don't 
have this NM info anymore if shutdown NM daemon while the kill -9 on NM daemon 
has the old behavior (RM still has NM info with running state and switch to 
LOST after timeout). Previously, the behavior of these two operations is the 
same. However, I don't think we care too much about consistency behavior for 
these two operations, but would like to call it out loudly to make sure we 
don't miss anything important. 

> The RM should handle the graceful shutdown of the NM.
> -----------------------------------------------------
>                 Key: YARN-41
>                 URL: https://issues.apache.org/jira/browse/YARN-41
>             Project: Hadoop YARN
>          Issue Type: New Feature
>          Components: nodemanager, resourcemanager
>            Reporter: Ravi Teja Ch N V
>            Assignee: Devaraj K
>              Labels: BB2015-05-TBR
>         Attachments: MAPREDUCE-3494.1.patch, MAPREDUCE-3494.2.patch, 
> MAPREDUCE-3494.patch, YARN-41-1.patch, YARN-41-2.patch, YARN-41-3.patch, 
> YARN-41-4.patch, YARN-41.patch
> Instead of waiting for the NM expiry, RM should remove and handle the NM, 
> which is shutdown gracefully.

This message was sent by Atlassian JIRA

Reply via email to