Github user marmbrus commented on a diff in the pull request:
https://github.com/apache/spark/pull/11235#discussion_r55403683
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -146,6 +147,56 @@ object EliminateSerialization extends
Rule[LogicalPlan] {
}
/**
+ * Add Filter to left and right of a EquiJoin to filter out rows with null
keys.
+ * So we may not need to check nullability of keys while joining. Besides,
by filtering
+ * out keys with null, we can also reduce data size in Join.
+ */
+object AddNullFilterForEquiJoin extends Rule[LogicalPlan] with
PredicateHelper {
+ def apply(plan: LogicalPlan): LogicalPlan = plan transform {
+ case ExtractEquiJoinKeys(Inner, leftKeys, rightKeys, condition, left,
right) =>
+ val leftConditions = leftKeys.distinct.map { l =>
+ IsNotNull(l)
+ }.reduceLeft(And)
+
+ val rightConditions = rightKeys.distinct.map { r =>
+ IsNotNull(r)
+ }.reduceLeft(And)
+
+ val keysConditions = ExpressionSet(leftKeys.zip(rightKeys).map { lr
=>
+ EqualTo(lr._2, lr._1)
+ }).reduceLeft(And)
+
+ val finalConditions = if (condition.isDefined) {
+ And(keysConditions, condition.get)
+ } else {
+ keysConditions
+ }
+
+ val leftHasAllNotNullPredicate = left.constraints.nonEmpty &&
+ left.constraints.filter(_.isInstanceOf[IsNotNull])
+ .forall(expr =>
leftConditions.references.intersect(expr.references).nonEmpty)
+
+ val rightHasAllNotNullPredicate = right.constraints.nonEmpty &&
+ right.constraints.filter(_.isInstanceOf[IsNotNull])
+ .forall(expr =>
rightConditions.references.intersect(expr.references).nonEmpty)
--- End diff --
This seems complicated and definitely needs more comments. Why are we not
just adding IsNotNull to the left/right when it doesn't already exist in the
conditions?
---
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]
