[
https://issues.apache.org/jira/browse/STORM-1434?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15422731#comment-15422731
]
Jungtaek Lim commented on STORM-1434:
-------------------------------------
Currently Storm SQL is relying on Trident so I'm thinking about doing it within
micro-batch for now, but would like to check that we want global aggregation
with state.
I'm not familiar with Trident optimization, but Trident API doc explained this,
"The benefits of CombinerAggregators are seen when you use them with the
aggregate method instead of partitionAggregate. In that case, Trident
automatically optimizes the computation by doing partial aggregations before
transferring tuples over the network."
Based on this explanation, we could make aggregate function based on
CombinerAggregator and call aggregate with this, and Trident does two pass
(partition -> global) aggregations.
> Support the GROUP BY clause in StormSQL
> ---------------------------------------
>
> Key: STORM-1434
> URL: https://issues.apache.org/jira/browse/STORM-1434
> Project: Apache Storm
> Issue Type: New Feature
> Components: storm-sql
> Reporter: Haohui Mai
>
> This jira tracks the effort of implement the support `GROUP BY` clause in
> StormSQL.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
