[
https://issues.apache.org/jira/browse/YARN-3575?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14548118#comment-14548118
]
Jason Lowe commented on YARN-3575:
----------------------------------
The only way to support compatibility would be to remove the epoch number field
from the container ID, but I doubt that's going to happen at this point. I
filed this mostly to document the fact that an incompatibility exists. Most
likely we'll have to recommend that users do _not_ perform a restart of the RM
where it tries to recover (and therefore starts using an epoch number in
container IDs) as long as applications are running on the grid using YARN
client jars version 2.5 or earlier. RM restart with recovery would only be
supported as long as all applications are using YARN jars >= 2.6.
> Job using 2.5 jars fails on a 2.6 cluster whose RM has been restarted
> ---------------------------------------------------------------------
>
> Key: YARN-3575
> URL: https://issues.apache.org/jira/browse/YARN-3575
> Project: Hadoop YARN
> Issue Type: Bug
> Components: client
> Affects Versions: 2.6.0
> Reporter: Jason Lowe
>
> Trying to launch a job that uses the 2.5 jars fails on a 2.6 cluster whose RM
> has been restarted (i.e.: epoch != 0) becaue the epoch number starts
> appearing in the container IDs and the 2.5 jars no longer know how to parse
> the container IDs.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
