[
https://issues.apache.org/jira/browse/BEAM-2620?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16809184#comment-16809184
]
Luke Cwik commented on BEAM-2620:
---------------------------------
This seems very doable by a runner looking at execution time metrics and
swapping to perform combiner lifting if the number of keys is low.
> Revisit GroupByKey.createWithFewKeys
> ------------------------------------
>
> Key: BEAM-2620
> URL: https://issues.apache.org/jira/browse/BEAM-2620
> Project: Beam
> Issue Type: Bug
> Components: beam-model
> Reporter: Thomas Groh
> Priority: Major
> Labels: portability
>
> This doesn't have a parallel within the GroupByKeyPayload, so there's
> currently no way to send it through the Runner API.
> The place it will almost always be created is in a {{Combine.globally()}}.
> It's potentially useful as an optimizer hint. The Dataflow Runner in
> streaming mode disables combiner lifting unless the GroupByKey has the
> fewKeys property set to true.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
