[
https://issues.apache.org/jira/browse/FLINK-1214?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14198339#comment-14198339
]
Stephan Ewen commented on FLINK-1214:
-------------------------------------
I am preparing a fix for this
> Prevent partitioning pushdown unless partitions fields match exactly
> --------------------------------------------------------------------
>
> Key: FLINK-1214
> URL: https://issues.apache.org/jira/browse/FLINK-1214
> Project: Flink
> Issue Type: Bug
> Components: Optimizer
> Affects Versions: 0.8-incubating
> Reporter: Stephan Ewen
> Assignee: Stephan Ewen
> Fix For: 0.8-incubating
>
>
> Consider an operation grouped on fields (A, B), followed by an operation
> grouped on field (A).
> Right now, the optimizer can push down the partitioning on (A), which serves
> both operations (the first step locally still groups by A and B). This may
> however by a bad idea for the cases where the field A has a low cardinality,
> or the value distribution is skewed.
> Since we cannot determine that robustly yet, I suggest to disable this
> optimization for now.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
