Github user marmbrus commented on the pull request:
https://github.com/apache/spark/pull/3888#issuecomment-69110525
The problem with a hybrid approach is: how do you choose between them? If
we had good statistics I would agree with you that we should use those to
decide. However, we don't at this point. Given that, I think a safe option
that may be slower but never OOMs the driver is better than aiming for optimal
performance in the small data cases.
Also, were you testing with broadcast left semi join enabled and statistics
for your tables? Avoiding a shuffle there might speed things up.
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
