[
https://issues.apache.org/jira/browse/YARN-6598?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16010673#comment-16010673
]
Jason Lowe commented on YARN-6598:
----------------------------------
bq. We should exhaustively check the Enum.valueOf to ensure backwards
compatibility in those cases as well.
I checked all the enums referenced for valueOf (i.e.: YarnApplicationState,
FinalApplicationStatus, YarnApplicationAttemptState, ).
YarnApplicationAttemptState had the FINAL_SAVING enum removed in YARN-1345, but
that change was committed way back in Hadoop 2.4.0 which did not support
rolling upgrades. So we should be good there. ContainerState had a QUEUED
state that was changed to the SCHEDULED state in YARN-4597. However this
change, and YARN-2883 that added the QUEUED state originally, only went into
Hadoop 2.9.0 and later, so the removed item was never in an official release.
So we should be fine there as well.
> History server getApplicationReport NPE when fetching report for pre-2.8 job
> ----------------------------------------------------------------------------
>
> Key: YARN-6598
> URL: https://issues.apache.org/jira/browse/YARN-6598
> Project: Hadoop YARN
> Issue Type: Bug
> Components: timelineserver
> Affects Versions: 2.8.0
> Reporter: Jason Lowe
> Assignee: Jason Lowe
> Priority: Blocker
> Attachments: YARN-6598.001.patch
>
>
> ApplicationHistoryManagerOnTimelineStore#convertToApplicationReport can NPE
> for a job that was run prior to the cluster upgrading to 2.8. It blindly
> assumes preemption metrics are present when CPU metrics are present, and when
> they are not it triggers the NPE.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
