Github user felixcheung commented on the issue: https://github.com/apache/spark/pull/19816 I think likely this is a general problem with default shuffle size being too big on limited resources when in tests (particularly on Windows/AppVeyor). The additional test failures are expected, I guess, since less partition can alter the result. Let's do a more isolated change for now? But I agree new tests can cause problems in the future. Perhaps we could identify a size that is smaller than the default that could get "close enough" to the result we have and set that for tests by default (16? 32? 64?).
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org