[
https://issues.apache.org/jira/browse/TEZ-4062?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16827256#comment-16827256
]
Jonathan Eagles commented on TEZ-4062:
--------------------------------------
+1. Committing this to master and branch-0.9. [~yingdachen], as to the origin
of build failure, there is some details in TEZ-4065 on the proper fix.
Essentially, Yetus is having some difficulty with determining the root of the
patch and improperly assumes tez-dag as the top level directory. Since
dependent modules in Tez were not built, 0.10.1-SNAPSHOT dependencies were
downloaded, but were outdated. To bypass the issue, I built new 0.10.1-SNAPSHOT
snapshot artifacts and published them to allow this build to pass until Yetus
patch issue is fixed.
> Speculative attempt scheduling should be aborted when Task has completed
> ------------------------------------------------------------------------
>
> Key: TEZ-4062
> URL: https://issues.apache.org/jira/browse/TEZ-4062
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Yingda Chen
> Assignee: Ying Han
> Priority: Major
> Attachments: TEZ-4062.001.patch
>
> Time Spent: 1h 10m
> Remaining Estimate: 0h
>
> In RedundantScheduleTransition (inside TaskImpl), we try to find the oldest
> running attempt and use it as the causual attempt when doing
> "addAndScheduleAttempt".
>
> However, the task may have completed at this moment, i.e., the task attempt
> that was considered running and long-tailed by speculator is now completed.
> In this case, we may not be able to find any unfinished attempt, which will
> lead to NPE in following logic (even without NPE, it still makes no sense to
> proceed with scheduling speculative attempt anyway)
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
