[
https://issues.apache.org/jira/browse/FLINK-3254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15126345#comment-15126345
]
ASF GitHub Bot commented on FLINK-3254:
---------------------------------------
Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/1568#discussion_r51427575
--- Diff:
flink-java/src/main/java/org/apache/flink/api/java/operators/GroupReduceOperator.java
---
@@ -193,7 +199,11 @@ public SingleInputSemanticProperties
getSemanticProperties() {
protected GroupReduceOperatorBase<?, OUT, ?>
translateToDataFlow(Operator<IN> input) {
String name = getName() != null ? getName() : "GroupReduce at "
+ defaultName;
-
+
+ this.function = (function instanceof CombineFunction) ?
+ new CombineToGroupCombineWrapper((CombineFunction)
function) :
--- End diff --
This does also add a bit of overhead to the `reduce` method because it is
wrapped. However, this should be OK since it is called just once per group.
> CombineFunction interface not respected
> ---------------------------------------
>
> Key: FLINK-3254
> URL: https://issues.apache.org/jira/browse/FLINK-3254
> Project: Flink
> Issue Type: Bug
> Components: DataSet API
> Affects Versions: 1.0.0, 0.10.1
> Reporter: Fabian Hueske
> Assignee: Kostas
> Priority: Blocker
> Fix For: 1.0.0
>
>
> The DataSet API offers a {{CombineFunction}} interface, which differs from
> the {{GroupCombineFunction}} interface by being restricted to return a single
> value instead of returning arbitrary many values through a {{Collector}}.
> The JavaDocs of the {{GroupCombineFunction}} point to the {{CombineFunction}}
> interface, advertising it as more efficient.
> However, the {{CombineFunction}} interface is nor respected by Flink, i.e., a
> {{GroupReduceFunction}} that implements this interface is executed without
> leveraging the combine method.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
