sunchao commented on PR #7258: URL: https://github.com/apache/datafusion-comet/pull/7258#issuecomment-6503695287
@comphead, the current head (`48e1024bfc`) now has green CI, including native tests, the Spark 4.1 shuffle tests and Spark SQL suites. I updated the description with that result. The local correctness tests and complete local-shuffle query benchmark are also documented there. For the waiting use case, enable `spark.comet.exec.broadcastExchange.producerCoalesce.enabled=true` on this build and compare with the flag off. This path applies when the broadcast build reads local Comet native shuffle with the supported types. For performance attribution, apply #7310 in both runs, as in the current measurements, so the comparison measures the additional benefit from batching. The flag remains opt-in because the wide-string control regresses and whole-query timings vary with run order. Please share the query/plan and data shape for that use case so we can check whether it exercises this path and assess the workload result. -- 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]
