[
https://issues.apache.org/jira/browse/YARN-1378?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13822030#comment-13822030
]
Bikas Saha commented on YARN-1378:
----------------------------------
Do you think loading application attempts will have cost proportional to apps
since there will be at least 1 app attempt per app? Would that double the
numbers we are seeing in the experiment? (which would be true in case the cost
is proportional to the number of znodes) What about all the delegations tokens
that we will be recovering? ZK on a real cluster will be on a different remote
machine which might be slower to access. How about other state store impls? ZK
is probably the fastest for such retrievals. We recommend using ZK store but
folks may use HDFS store.
> Implement a RMStateStore cleaner for deleting application/attempt info
> ----------------------------------------------------------------------
>
> Key: YARN-1378
> URL: https://issues.apache.org/jira/browse/YARN-1378
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: resourcemanager
> Reporter: Jian He
> Assignee: Jian He
> Attachments: YARN-1378.1.patch, YARN-1378.2.patch, YARN-1378.3.patch
>
>
> Now that we are storing the final state of application/attempt instead of
> removing application/attempt info on application/attempt
> completion(YARN-891), we need a separate RMStateStore cleaner for cleaning
> the application/attempt state.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
