[
https://issues.apache.org/jira/browse/YARN-6531?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15984867#comment-15984867
]
Bibin A Chundatt commented on YARN-6531:
----------------------------------------
[[email protected]]
Since the limit is in zookeeper side my point is check should be part of
ZkStateStore. Similar to zk retry times limit validation we can have in
ZkStateStore once the event is processed. Since StateStore dispatchers are
different i think the cost for having the check after sending also should be
fine. Application can be rejected from {{NEW_SAVING}} state directly.
> Check appStateData size before saving to Zookeeper
> --------------------------------------------------
>
> Key: YARN-6531
> URL: https://issues.apache.org/jira/browse/YARN-6531
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: Bibin A Chundatt
> Assignee: Bibin A Chundatt
> Priority: Critical
>
> Application with large size Application submission context could cause store
> to Zookeeper failure due to znode size limit. Zookeeper znode limit exception
> thrown {{org.apache.zookeeper.KeeperException$ConnectionLossException}}.
> ZkStateStore will retry for configured times and will throw
> ConnectionLossException after configured limit.
> Which could cause Resource manager to switch from active To StandBy and other
> application submitted not getting save to ZK.
> Solution {{ApplicationStateData}} size to be validated before saving and
> reject application so that ResourceManager is not impacted.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
