[
https://issues.apache.org/jira/browse/FLINK-1001?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14053832#comment-14053832
]
Stephan Ewen commented on FLINK-1001:
-------------------------------------
Actually, though I think the aggregators are susceptible to mutable object
bugs, the problem is in your code here:
The reduce function you write is a "minBy(0)", while the aggregation is
"min(0)". The first one selects the tuple where field 0 is smallest, the second
one picks for field 0 the minimum value (like in SQL). The same way as in SQL,
other non-grouped and non-aggregated fields are non deterministic.
> Aggregate Min/Max return unexpected values.
> -------------------------------------------
>
> Key: FLINK-1001
> URL: https://issues.apache.org/jira/browse/FLINK-1001
> Project: Flink
> Issue Type: Bug
> Reporter: Bastian Köcher
> Attachments: src.zip
>
>
> I wanted to replace my simple MinReducer
> public static class MinSimilarityReducer extends
> ReduceFunction<ClusterPair> {
> @Override
> public ClusterPair reduce(ClusterPair value1, ClusterPair
> value2) throws Exception {
> if (value1.getSimilarity().doubleValue() <
> value2.getSimilarity().doubleValue()) {
> return value1;
> }
> return value2;
> }
> }
> With Aggregate.Min, but Aggregate.Min delivers a value which should already
> be removed in my WorkSet.
> If I use my own Reducer it works as expected, but with an Aggregate this
> doesn't work anymore :/
--
This message was sent by Atlassian JIRA
(v6.2#6252)
