Github user squito commented on the issue: https://github.com/apache/spark/pull/21698 yeah, you'd have to sort the entire record. I think originally it didn't seem like that would work, because you don't know that `T` is sortable for `RDD[T]`. But after a sort, you've got bytes, so you can at least sort the serialized bytes of `T`. Note that you don't want to do the entire repartitioning based on that, since you won't get a good distribution for skewed data. But it can give you a deterministic order as an *additional* sort, just within one partition.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org