[
https://issues.apache.org/jira/browse/FLINK-8690?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16445682#comment-16445682
]
Fabian Hueske commented on FLINK-8690:
--------------------------------------
Hi [~walterddr] and [~hequn8128],
Thanks for the good discussion!
I think there are two ways to solve this issue:
1. Separate logical optimization rule sets for batch and streaming as you did
in your branch.
2. Move the {{AggregateExpandDistinctAggregatesRule.JOIN}} rule to the existing
batch normalization rules, generalize {{FlinkLogicalAggregateConverter}} to
accept {{DISTINCT}} aggregates, and guard for {{DISTINCT}} aggregations in the
batch physical translation.
Both approaches will work, but I think the second approach is a bit cleaner
because doesn't rely on the cost pruning.
What do you think?
> Update logical rule set to generate FlinkLogicalAggregate explicitly allow
> distinct agg on DataStream
> -----------------------------------------------------------------------------------------------------
>
> Key: FLINK-8690
> URL: https://issues.apache.org/jira/browse/FLINK-8690
> Project: Flink
> Issue Type: Sub-task
> Reporter: Rong Rong
> Assignee: Rong Rong
> Priority: Major
>
> Currently, *FlinkLogicalAggregate / FlinkLogicalWindowAggregate* does not
> allow distinct aggregate.
> We are proposing to reuse distinct aggregate codegen work designed for
> *FlinkLogicalOverAggregate*, to support unbounded distinct aggregation on
> datastream as well.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
