dwsmith1983 commented on issue #6530: URL: https://github.com/apache/datafusion-comet/issues/6530#issuecomment-5954764903
@comphead a separate cause in the same code is #6539: under AQE the final aggregate keeps a native plan serialized before its shuffle stage existed, so it reads its shuffle through the JVM decoder. While measuring that, a local prototype also had `convertNode` turn an `AQEShuffleReadExec` over a Comet shuffle stage into a `CometExchangeSink`, with `shouldUseShuffleScan` accepting it. An AQE skew join with `PartialReducerPartitionSpec` splits then ran as `CometSortMergeJoin` over native shuffle reads and matched Spark. I'm opening a PR for #6539 only, without the `AQEShuffleReadExec` conversion, so that part stays with this issue. Both changes sit in `CometExecRule` and `CometExchangeSink.shouldUseShuffleScan`. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
