[
https://issues.apache.org/jira/browse/FLINK-1176?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14176393#comment-14176393
]
Fabian Hueske commented on FLINK-1176:
--------------------------------------
Please correct me if I am wrong, but is a local reducer in this case not the
same as a combiner and the second reducer like a reducer? If that's the case,
we should be careful to use the same terminology for the stream and the batch
layers.
Does this issue only consider key-less reducers (All-Reduce in batch terms) or
are also reducers on key-groups addressed?
> Add operator for global reduce/aggregations
> -------------------------------------------
>
> Key: FLINK-1176
> URL: https://issues.apache.org/jira/browse/FLINK-1176
> Project: Flink
> Issue Type: Improvement
> Components: Streaming
> Reporter: Gyula Fora
> Priority: Minor
>
> Currently streaming reduce operators (like .reduce or .window(..).reduce,
> sum, etc.) only allow local reduces/aggregations. While one can achieve
> global aggregation by setting the parallelism of these operators to 1 it can
> cause a serious bottleneck in the program.
> For window/batch reduces and aggregations a global reducer can be added, with
> parallelism 1, to further reduce partial results. This global reduce option
> could be introduced as an optional boolean parameter for the intended
> operators.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
