[
https://issues.apache.org/jira/browse/HIVE-8943?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Szehon Ho updated HIVE-8943:
----------------------------
Attachment: HIVE-8943.patch
Attaching a patch to add size of connected mapjoin operator in same work
(spark-stage) to the calculation of whether to convert current join to mapjoin.
I added two unit tests to stress this case, but would rather wait for HIVE-8946
as the tests wont be using mapjoin until then.
> Fix memory limit check for combine nested mapjoins [Spark Branch]
> -----------------------------------------------------------------
>
> Key: HIVE-8943
> URL: https://issues.apache.org/jira/browse/HIVE-8943
> Project: Hive
> Issue Type: Sub-task
> Components: Spark
> Affects Versions: spark-branch
> Reporter: Szehon Ho
> Assignee: Szehon Ho
> Attachments: HIVE-8943.patch
>
>
> Its the opposite problem of what we thought in HIVE-8701.
> SparkMapJoinOptimizer actually does combine nested mapjoins into one work due
> to removal of RS for big-table. So we actually need to enhance the check to
> calculate if all the MapJoins in that work (spark-stage) will fit into the
> memory, otherwise it might overwhelm memory for that particular spark
> executor.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
