[
https://issues.apache.org/jira/browse/TEZ-1950?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14277919#comment-14277919
]
Hitesh Shah commented on TEZ-1950:
----------------------------------
[~rajesh.balamohan] [~sseth] Does the above analysis seem correct?
> Incorrect handling of locality counter in TaskAttemptImpl
> ---------------------------------------------------------
>
> Key: TEZ-1950
> URL: https://issues.apache.org/jira/browse/TEZ-1950
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Hitesh Shah
>
> To maintain task attempt counters, we are using
> TaskAttempt.TaskAttemptStatus.counters
> Now, counters is not accessed in a thread safe manner.
> Counters are updated in either StatusUpdaterTransition or modified as part of
> TaskAttempt.TaskAttemptStatus::setLocalityCounter().
> In a scenario, where TaskAttempt::getCounters() is called before any status
> update transition comes back, the locality counter will get lost because the
> atomic boolean flag is never reset.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
