[jira] [Commented] (HIVE-3302) Race condition in query plan for merging at the end of a query

["Kevin Wilfong (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-3302?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13422782#comment-13422782 ] 

Kevin Wilfong commented on HIVE-3302:
-------------------------------------

JIRA screwed up the formatting of my diagrams, the first is supposed to show MoveTask(1) and MergeTask stemming from ConditionalTask and MoveTask(2) dependent upon MoveTask(1) and MergeTask.

The second is supposed to show MergeTask stemming from ConditionalTask, MoveTask(1) and MoveTask(2) dependent on MergeTask, and most importantly, no dependency relationship between MoveTask(1) and MoveTask(2).
                
> Race condition in query plan for merging at the end of a query
> --------------------------------------------------------------
>
>                 Key: HIVE-3302
>                 URL: https://issues.apache.org/jira/browse/HIVE-3302
>             Project: Hive
>          Issue Type: Bug
>          Components: Query Processor
>    Affects Versions: 0.10.0
>            Reporter: Kevin Wilfong
>            Assignee: Kevin Wilfong
>            Priority: Critical
>
> In the query plan that's used to merge files at the end of a query, the dependency tree looks something like:
>                    MoveTask(1)
>                   /           \
> ...ConditionalTask             MoveTask(2)...
>                   \           /
>                    MergeTask
> Here MoveTask(1) moves the partition data to a temporary location, and MoveTask(2) moves it to the final location.
> However if there are dynamic partitions generated and some of these partitions are merged and others are moved, the dependency tree is changed at runtime to:
> ...ConditionalTask           MoveTask(2)...
>                   \         /
>                    MergeTask
>                             \
>                              MoveTask(1)
> This produces a race condition between the two MoveTasks where if MoveTask(2) runs before MoveTask(1) the partitions moved by MoveTask(1) will get moved to an intermediate location and never moved to the final location.  In this case those partitions are quietly lost.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira