[ 
https://issues.apache.org/jira/browse/YARN-2558?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14137757#comment-14137757
 ] 

Jason Lowe commented on YARN-2558:
----------------------------------

bq. Jason Lowe, without changing containerTokenIdentifier, we can't support 
rolling upgrades. does the current approach look good you ?

Yes, this should be fine for the short term.  I just wanted to make it clear 
that until YARN-668 is addressed we're going to continue to break backwards 
compatibility and thus rolling upgrades with seemingly simple changes like this.

Some minor comments on the patch, none of which are must-fix:
- is it necessary to call testNMToken in testContainerManagerWithEpoch?  That 
test is already covered by testContainerManager.
- the timeout on the test seems way too large
- not sure what the point of having the test catch an exception just to print a 
stacktrace and re-throw it.  Won't the stacktrace be printed also when the test 
fails due to the thrown exception?

> Updating ContainerTokenIdentifier#read/write to use ContainerId#getContainerId
> ------------------------------------------------------------------------------
>
>                 Key: YARN-2558
>                 URL: https://issues.apache.org/jira/browse/YARN-2558
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>            Reporter: Tsuyoshi OZAWA
>            Assignee: Tsuyoshi OZAWA
>            Priority: Blocker
>         Attachments: YARN-2558.1.patch, YARN-2558.2.patch
>
>
> We should update ContainerTokenIdentifier#read/write to use 
> {{getContainerId}} instead of {{getId}} to pass all container information 
> correctly.



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

Reply via email to