[
https://issues.apache.org/jira/browse/HIVE-5381?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13781637#comment-13781637
]
Navis commented on HIVE-5381:
-----------------------------
HIVE-4965 is for skipping final serialization into PTFPartition. This is for
making rows of lead/lag function per row basis instead of per partition basis.
> OOM when using many lead/lag funtions against large partition
> -------------------------------------------------------------
>
> Key: HIVE-5381
> URL: https://issues.apache.org/jira/browse/HIVE-5381
> Project: Hive
> Issue Type: Improvement
> Components: Query Processor
> Reporter: Navis
> Assignee: Navis
> Priority: Minor
> Attachments: D13161.1.patch
>
>
> Currently, columns containing window function are evaluated for all rows in a
> partition, which possibly make OOM if the partition is big (400MB in my
> case). For lead/lag functions, the value can be evaluated per row.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
