cloud-fan commented on code in PR #39170:
URL: https://github.com/apache/spark/pull/39170#discussion_r1163913638
##########
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/InjectRuntimeFilter.scala:
##########
@@ -114,51 +115,89 @@ object InjectRuntimeFilter extends Rule[LogicalPlan] with
PredicateHelper with J
}
/**
- * Returns whether the plan is a simple filter over scan and the filter is
likely selective
+ * Extracts the plan exists a simple filter over scan and the filter is
likely selective
* Also check if the plan only has simple expressions (attribute reference,
literals) so that we
* do not add a subquery that might have an expensive computation
*/
- private def isSelectiveFilterOverScan(plan: LogicalPlan): Boolean = {
+ private def extractSelectiveFilterOverScan(
+ plan: LogicalPlan,
+ filterCreationSideExp: Expression): Option[LogicalPlan] = {
@tailrec
def isSelective(
p: LogicalPlan,
predicateReference: AttributeSet,
hasHitFilter: Boolean,
- hasHitSelectiveFilter: Boolean): Boolean = p match {
- case Project(projectList, child) =>
- if (hasHitFilter) {
- // We need to make sure all expressions referenced by filter
predicates are simple
- // expressions.
- val referencedExprs = projectList.filter(predicateReference.contains)
- referencedExprs.forall(isSimpleExpression) &&
- isSelective(
- child,
- referencedExprs.map(_.references).foldLeft(AttributeSet.empty)(_
++ _),
- hasHitFilter,
- hasHitSelectiveFilter)
+ hasHitSelectiveFilter: Boolean): Option[LogicalPlan] = p match {
+ case Project(projectList, child) if hasHitFilter =>
+ // We need to make sure all expressions referenced by filter
predicates are simple
+ // expressions.
+ val referencedExprs = projectList.filter(predicateReference.contains)
+ if (referencedExprs.forall(isSimpleExpression)) {
+ isSelective(
+ child,
+ referencedExprs.map(_.references).foldLeft(AttributeSet.empty)(_
++ _),
+ hasHitFilter,
+ hasHitSelectiveFilter)
} else {
- assert(predicateReference.isEmpty && !hasHitSelectiveFilter)
- isSelective(child, predicateReference, hasHitFilter,
hasHitSelectiveFilter)
+ None
}
- case Filter(condition, child) =>
- isSimpleExpression(condition) && isSelective(
+ case Project(_, child) =>
+ assert(predicateReference.isEmpty && !hasHitSelectiveFilter)
+ isSelective(child, predicateReference, hasHitFilter,
hasHitSelectiveFilter)
+ case Filter(condition, child) if isSimpleExpression(condition) =>
+ isSelective(
child,
predicateReference ++ condition.references,
hasHitFilter = true,
hasHitSelectiveFilter = hasHitSelectiveFilter ||
isLikelySelective(condition))
- case _: LeafNode => hasHitSelectiveFilter
- case _ => false
+ case ExtractEquiJoinKeys(joinType, _, _, _, _, left, right, hint) =>
+ // Runtime filters use one side of the [[Join]] to build a set of join
key values and prune
+ // the other side of the [[Join]]. It's also OK to use a superset of
the join key values to
+ // do the pruning.
+ if (isLeftSideSuperset(joinType, left, filterCreationSideExp) &&
Review Comment:
we can simplify the condition here. Join children always output a superset
of join output (if null is ignored), except for the right child of left joins.
However, for left joins, the right side plan output is not included in the join
output, so `left/right.output.exists(_.semanticEquals(filterCreationSideExp))`
should just work.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
