cloud-fan edited a comment on pull request #32328: URL: https://github.com/apache/spark/pull/32328#issuecomment-827637229
The overhead of skew join handling is to replicate the non-skew side, and the overhead is larger in SHJ because we need to build the hash relation. But the overhead of skew should be more significant than this. I don't think this PR makes OOM much more likely to happen. It does add more tasks that need to build a hash relation, but the parallelism on the executor JVM is not increased. E.g. we may still run 8 tasks at the same time on one executor to build hash relations, though the total number of tasks is increased from 100 to 110. -- 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]
