[
https://issues.apache.org/jira/browse/FLINK-3474?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15182844#comment-15182844
]
ASF GitHub Bot commented on FLINK-3474:
---------------------------------------
Github user fhueske commented on the pull request:
https://github.com/apache/flink/pull/1746#issuecomment-193194745
Thanks Chengxiang for the update. I think we can merge the PR now.
But we should open a JIRA and improve the LongAvg aggregation. The
communication between Mapper and GroupReduce for aggregation is Flink internal
and there is no need to expose this to Calcite. I am not sure if our
abstraction is well chosen, i.e., that a single DataSetNode should only
translate to a single DataSet operator. Maybe we should do another refactoring
before merging back to master?
> Partial aggregate interface design and sort-based implementation
> ----------------------------------------------------------------
>
> Key: FLINK-3474
> URL: https://issues.apache.org/jira/browse/FLINK-3474
> Project: Flink
> Issue Type: Sub-task
> Components: Table API
> Reporter: Chengxiang Li
> Assignee: Chengxiang Li
>
> The scope of this sub task includes:
> # Partial aggregate interface.
> # Simple aggregate function implementation, such as SUM/AVG/COUNT/MIN/MAX.
> # DataSetAggregateRule which translate logical calcite aggregate node to
> Flink user functions. As hash-based combiner is not available yet(see PR
> #1517), we would use sort-based combine as default.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
