Github user marmbrus commented on a diff in the pull request:
https://github.com/apache/spark/pull/837#discussion_r13520365
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/planning/patterns.scala
---
@@ -119,6 +119,11 @@ object HashFilteredJoin extends Logging with
PredicateHelper {
case FilteredOperation(predicates, join @ Join(left, right, Inner,
condition)) =>
logger.debug(s"Considering hash inner join on: ${predicates ++
condition}")
splitPredicates(predicates ++ condition, join)
+ // All predicates can be evaluated for left semi join (those that are
in the WHERE
+ // clause can only from left table, so they can all be pushed down.)
--- End diff --
I think in general we should avoid making too many assumptions in the
planner about what optimizations have occurred. For example, in the future we
might avoid pushing down predicates that are very expensive to evaluate as it
might be cheaper to run them on an already filtered set of data. However, in
the case of LEFT SEMI JOIN, I think it is actually okay to push all evaluation
into the join condition, even if they only refer to the left table. Is that
true?
---
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.
---
