[
https://issues.apache.org/jira/browse/MAPREDUCE-1383?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12805754#action_12805754
]
Devaraj Das commented on MAPREDUCE-1383:
----------------------------------------
Some comments:
1) Remove the LOG.info statements from TokenCache
2) Please add a method in TokenCache to load tokens from a file specified in
the argument (rather than going indirectly through the conf). Then in
Child.java, you can call that API directly rather than setting the file in conf.
3) Does it make sense to have a
TokenCache.addPathsForGettingDelegationToken(Path[]) that's called from the
places where you currently have TokenCache.obtainTokensForNamenodes. Then just
before job submission you make one call to TokenCache.obtainTokensForNamenodes.
That way we will be sure that there is not more than one per a namenode to get
delegation tokens.
4) You define getDelegationTokens in TrackerDistributedCacheManager but don't
invoke it. Also, you just need to pass the paths to the
TokenCache.obtainTokensForNamenodes because that's already checking for
duplicate entries.
We discussed offline that you need to fix the build so that the hdfs jars are
in the classpath for the aspects compilation..
> Allow storage and caching of delegation token.
> ----------------------------------------------
>
> Key: MAPREDUCE-1383
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-1383
> Project: Hadoop Map/Reduce
> Issue Type: New Feature
> Reporter: Boris Shkolnik
> Attachments: MAPREDUCE-1383-1.patch, MAPREDUCE-1383-2.patch,
> MAPREDUCE-1383-5.patch, MAPREDUCE-1383-6.patch
>
>
> Client needs to obtain delegation tokens from all the NameNodes it is going
> to work with and pass it to the application.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
