[
https://issues.apache.org/jira/browse/HIVE-1980?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12992753#comment-12992753
]
Ning Zhang commented on HIVE-1980:
----------------------------------
I remembered that this case is not supported by design and we should throw an
error in SemanticAnalyzer.
In the first dynamic partition insert patch we disabled merge completely
because in the case of using HiveInputFormat (eg. in Hadoop 0.17) the partition
columns are not passed to the reducer (part of the partition columns exist in
the HDFS directory). So the reducer will create one file that may mix data from
different partitions. In HIVE-1307 we enabled merge for CombineHiveInputFormat.
However we should disable merge for the case of dynamic partition inserts using
HiveInputFormat.
> Merging using mapreduce rather than map-only job failed in case of dynamic
> partition inserts
> --------------------------------------------------------------------------------------------
>
> Key: HIVE-1980
> URL: https://issues.apache.org/jira/browse/HIVE-1980
> Project: Hive
> Issue Type: Bug
> Reporter: Ning Zhang
> Assignee: Ning Zhang
>
> In dynamic partition insert and if merge is set to true and
> hive.mergejob.maponly=false, the merge MapReduce job will fail.
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
