Github user mridulm commented on the issue:
https://github.com/apache/spark/pull/22112
@tgravescs I was specifically in agreement with
> Personally I don't want to talk about implementation until we decide what
we want our semantics to be around the unordered operations because that
affects any implementation.
and
> I would propose we fix the things that are using the round robin type
partitioning (repartition) but then unordered things like zip/MapPartitions
(via user code) we document or perhaps give the user the option to sort.
IMO a fix in spark core for repartition should work for most (if not all)
order dependent closures - we might choose not to implement for others due to
time constraints; but basic idea should be fairly similar.
Given this, I am fine with documenting the potential issue for others and
fix for a core subset - with assumption that we will expand solution to cover
all later.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
