Github user tgravescs commented on the issue:
https://github.com/apache/spark/pull/22112
Thanks for the clarification, but I guess my point is with your last
statement:
> - with assumption that we will expand solution to cover all later.
If we document this and say we support unordered operations with the caveat
that failures could result in different results, my assumption is we don't
necessarily have to do anything else ever (this is what I am proposing). We
could decide to for instance add an option to sort, or if its not a result
stage fail more tasks to try handle the situation, but strictly speaking we
wouldn't have to.
If you think we have to fix those operations that can result in unordered
then I think it comes back to we just don't support unordered operations at all
and we should say that and probably force the sort on all these operations and
possibly on all operations where user could cause it to be different order on
rerun.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
