[
https://issues.apache.org/jira/browse/HIVE-8298?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14152687#comment-14152687
]
Ashutosh Chauhan commented on HIVE-8298:
----------------------------------------
Issue is that while merging this n-way join we copied expressions from one join
tree into another without paying attention to fact that they could be in
different order. There are two workarounds for it:
* Introduce any udf on join condition like cast column to string.
* Rewrite the query so that expressions are in same order in two join.
First one disables the join merge altogether, thereby avoiding buggy code-path.
So, it will avoid the bug but will result in 2 jobs and thereby suboptimal
performance. Second one is a better workaround since join merging will still
happen, but since original order of expressions is maintained, bug doesn't kick
in.
> Incorrect results for n-way join when join expressions are not in same order
> across joins
> -----------------------------------------------------------------------------------------
>
> Key: HIVE-8298
> URL: https://issues.apache.org/jira/browse/HIVE-8298
> Project: Hive
> Issue Type: Bug
> Components: Logical Optimizer
> Affects Versions: 0.13.0, 0.13.1
> Reporter: Ashutosh Chauhan
> Assignee: Ashutosh Chauhan
> Priority: Blocker
>
> select count( * ) from srcpart a join srcpart b on a.key = b.key and a.hr =
> b.hr join srcpart c on a.hr = c.hr and a.key = c.key;
> is minimal query which reproduces it
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
