zentol commented on PR #22778: URL: https://github.com/apache/flink/pull/22778#issuecomment-1592810226
> MinMaxSelector I assume you are referring to the partitioner. > I don't understand your confusion on why the test succeeded with the kryo serialization. My assumption is that, based on the javadocs, range partitioning takes the _value ranges_ into account (aka, isn't just hash-partitioning data), but Kryo records are opaque and there isn't a real value that Flink can inspect, in contrast to the POJO type where Flink can directly read the long values. If you don't have access to the values, then you either must use hashes or base it on the serialized data; using the serialized data of an opaque data type doesn't really make sense to me, and hashes shouldn't (?) be able to ensure that the partitioning works as nicely as the test requires. Maybe we just sort the set locally (since the POJO does implement Comparable) and slice that into partitions :shrug: -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
