[ 
https://issues.apache.org/jira/browse/HIVE-9026?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14247383#comment-14247383
 ] 

Chao commented on HIVE-9026:
----------------------------

What I've found is:
- test result depends on the # of reducers. When the # of reducers is larger 
than 1, the result may be wrong.
- # of reducers may depends on whether a query is executed alone, or together 
with other queries.

subquery_in.q is an example for the above issues. Right now, 
ppd_multi_insert.q, subquery_in.q and join_nullsafe.q all have different 
results because of the above.

The same issue also present in the trunk. To reproduce, just add "set 
mapred.reduce.tasks=3" to subquery_in.q, and run the test. I'm still not sure 
what is the exact reason for this.

> Re-enable remaining tests after HIVE-8970 [Spark Branch]
> --------------------------------------------------------
>
>                 Key: HIVE-9026
>                 URL: https://issues.apache.org/jira/browse/HIVE-9026
>             Project: Hive
>          Issue Type: Sub-task
>          Components: spark-branch
>    Affects Versions: spark-branch
>            Reporter: Chao
>
> In HIVE-8970, we disabled several tests which seem to be related to an bug in 
> upstream. I filed HIVE-9025 to track it.
> {noformat}
> join38.q
> join_literals.q
> join_nullsafe.q
> subquery_in.q
> ppd_multi_insert.q
> {noformat}
> We need to re-enable these tests after HIVE-9025 is resolved.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to