[ https://issues.apache.org/jira/browse/YARN-1428?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13827345#comment-13827345 ]
Zhijie Shen commented on YARN-1428: ----------------------------------- The problem described here is irrelevant to RM Restart (in particular, YARN-891). Let me clarify the problem more: Previously, I invoke RMApplicationHistoryWriter#applicationFinished/applicationAttemptFinished in the transition to the final state of RMApp/RMAppAttempt. However, during this transition, #getState() is not getting the final state, but the one before RMApp/RMAppAttempt enters the final state. Therefore, RMApp/RMAppAttempt is going to write non-final state into application history store, which is not expected. I'm targeting to fix this problem by telling RMApplicationHistoryWriter what the final state is going to be, and RMApplicationHistoryWriter writes this state instead. By the way, before touching the RM integration, we've already merge the latest changes (including YARN-891) into YARN-321 branch. Therefore, YARN-953's patch and this one are already on top of newest RMApp/RMAppAttempt's transitions. > RM cannot write the final state or RMApp/RMAppAttempt in the transition to > the final state > ------------------------------------------------------------------------------------------ > > Key: YARN-1428 > URL: https://issues.apache.org/jira/browse/YARN-1428 > Project: Hadoop YARN > Issue Type: Sub-task > Reporter: Zhijie Shen > Assignee: Zhijie Shen > Attachments: YARN-1428.1.patch > > > ApplicationFinishData and ApplicationAttemptFinishData are written in the > final transitions of RMApp/RMAppAttempt respectively. However, in the > transitions, getState() is not getting the state that RMApp/RMAppAttempt is > going to enter, but prior one. -- This message was sent by Atlassian JIRA (v6.1#6144)