[ 
https://issues.apache.org/jira/browse/MAPREDUCE-3398?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13150245#comment-13150245
 ] 

Siddharth Seth commented on MAPREDUCE-3398:
-------------------------------------------

Some options to fix this.
1. Change the NM and RM interaction - to have the RM wait for all NMs to notify 
the RM about contianer / app completion (or timeout). This would pretty much be 
a reversal of the current mechanism where the RM is responsible for app 
completion. Would be risky - the RM should not have to wait for NMs.
2. Have the NMs run as a proxy user - has a lot of operational issues, and 
possibly security concerns.
3. Aggregate logs as the NM user - this breaks the CLI.
4. The NMs asks for application token renewal while log aggregation is still to 
complete. This seems like the best option. Credit to Vinod for suggesting it.
                
> Log Aggregation broken in Secure Mode
> -------------------------------------
>
>                 Key: MAPREDUCE-3398
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-3398
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: mrv2, nodemanager
>    Affects Versions: 0.23.0
>            Reporter: Siddharth Seth
>            Assignee: Siddharth Seth
>            Priority: Critical
>
> Log aggregation in secure mode does not work with MAPREDUCE-2977. The 
> nodemanager relies on the users credentials to write out logs to HDFS. These 
> credentials are currently cancelled once a job completes, before the NM can 
> write out the logs.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Reply via email to