manuzhang commented on pull request #29797: URL: https://github.com/apache/spark/pull/29797#issuecomment-727132781
@maryannxue thanks for pointing out the fundamental problems while I'm not familiar with the design behind. When starting out, I was trying to solve the most urgent issue in our use cases, too many output files caused by LSR, which broke the downstream jobs on the pipeline. Meanwhile, disabling LSR entirely downgrades the performance of BHJ switched from SMJ. Hence, this PR with help of @cloud-fan. > But again, this is a "repartition" issue rather than the write command issue. Introducing this kind of hack in AQE plan does not make sense. 1. Is that ok or even good if repartition is optimized away in cases other than on write command ? 2. Moving forward, what's your suggestion ? Do you want me revert this PR ? ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
