Github user rxin commented on the issue:
https://github.com/apache/spark/pull/15575
@yhuai let me be more clear. It is actually fairly confusing to say "this
only changes if the operator doesn't shuffle data".
The thing is that we are relying on each operator's output partitioning to
determine injecting exchanges. So is it safe to depend on "child", which is
ambiguous in this context? Child could mean the current child before injecting
exchange, or it could mean the injected exchange operator. Can the query
planner incorrectly not injecting exchange because a plan node is deemed
already having the correct output partitioning? Or is this always guaranteed to
not happen due to certain ordering we enforce in planning.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
