[
https://issues.apache.org/jira/browse/HIVE-7731?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14097311#comment-14097311
]
Chao commented on HIVE-7731:
----------------------------
I think the issue is in {{SparkMapRecordHandler::map}}, in which we channel
BOTH ReduceSinkOperator to the same mapper, which then get directed to the
first ReduceWork.
And since the second ReduceWork is ignored, it is empty at end.
> Incorrect result returned when a map work has multiple downstream reduce
> works [Spark Branch]
> ---------------------------------------------------------------------------------------------
>
> Key: HIVE-7731
> URL: https://issues.apache.org/jira/browse/HIVE-7731
> Project: Hive
> Issue Type: Sub-task
> Components: Spark
> Reporter: Rui Li
>
> Encountered when running on spark. Suppose we have three tables:
> {noformat}
> table1(x int, y int);
> table2(x int);
> table3(x int);
> {noformat}
> I run the following query:
> {noformat}
> from table1
> insert overwrite table table2 select x group by x
> insert overwrite table table3 select y group by y;
> {noformat}
> The query generates 1 map and 2 reduces. The map operator has 2 RS, so I
> suppose it has output for both reduces.
> The problem is all (incorrect) results go to table2 and table3 is empty.
> I tried the same query on MR and it gives correct results.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
