[ 
https://issues.apache.org/jira/browse/YARN-2038?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14507266#comment-14507266
 ] 

sandflee commented on YARN-2038:
--------------------------------

If nm register to rm in a short time, we can add a interface to 
ApplicationMasterService to tell am container comes back.
If nm are not registered to rm after nm expire time, rm knows nothing about nm 
now. Could AM tell RM the node and container Info through 
ApplicationMasterService.registerApplicationMaster  while reregister to rm? 
with this info, RM could treat the unreigstered NM  as a lost NODE after nm 
expire time, and pass the container complete msg to am.  
In this solution , we need am to store container info.

> Revisit how AMs learn of containers from previous attempts
> ----------------------------------------------------------
>
>                 Key: YARN-2038
>                 URL: https://issues.apache.org/jira/browse/YARN-2038
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager
>    Affects Versions: 2.4.0
>            Reporter: Karthik Kambatla
>
> Based on YARN-556, we need to update the way AMs learn about containers 
> allocation previous attempts. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to