swuferhong commented on PR #23423: URL: https://github.com/apache/flink/pull/23423#issuecomment-1754162270
> FLINK-27898 > @swuferhong Thanks for the pr. Overall, LGTM in code level. But I have one concern about it, in batch mode, it's total right to reduce the empty partition to a single Values node. But in stream mode, is it expected? I mean in optimize phase, no match partitions, but user expect to monitor the match partitions continuously. If reduce it to a single Values node, user will lose the ability. A similar issue it reported in [FLINK-27898](https://issues.apache.org/jira/browse/FLINK-27898). Hi, @luoyuxia . I think it's another issue. The problem is that the partition we push down is static and has already been determined during optimize phase, for streaming mode, it will not work for some cases. However, I think these two issues need consider separately. Currently, this PR is only an optimization under the existing `SupportsPartitionPushDown`. If we want to better support partition push down in streaming mode, I think we need new flip and pr to improve the existing interface. -- 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: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org