Github user mgaido91 commented on a diff in the pull request:
https://github.com/apache/spark/pull/21109#discussion_r188242730
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/planning/patterns.scala
---
@@ -131,13 +134,101 @@ object ExtractEquiJoinKeys extends Logging with
PredicateHelper {
if (joinKeys.nonEmpty) {
val (leftKeys, rightKeys) = joinKeys.unzip
- logDebug(s"leftKeys:$leftKeys | rightKeys:$rightKeys")
- Some((joinType, leftKeys, rightKeys,
otherPredicates.reduceOption(And), left, right))
+
+ // Find any simple range expressions between two columns
+ // (and involving only those two columns)
+ // of the two tables being joined,
+ // which are not used in the equijoin expressions,
+ // and which can be used for secondary sort optimizations.
+ val rangePreds: mutable.Set[Expression] = mutable.Set.empty
+ var rangeConditions: Seq[BinaryComparison] =
+ if (SQLConf.get.useSmjInnerRangeOptimization) { // &&
SQLConf.get.wholeStageEnabled) {
+ otherPredicates.flatMap {
+ case p@LessThan(l, r) => isValidRangeCondition(l, r, left,
right, joinKeys) match {
+ case "asis" => rangePreds.add(p); Some(LessThan(l, r))
+ case "vs" => rangePreds.add(p); Some(GreaterThan(r, l))
+ case _ => None
+ }
+ case p@LessThanOrEqual(l, r) =>
+ isValidRangeCondition(l, r, left, right, joinKeys) match {
+ case "asis" => rangePreds.add(p);
Some(LessThanOrEqual(l, r))
+ case "vs" => rangePreds.add(p);
Some(GreaterThanOrEqual(r, l))
+ case _ => None
+ }
+ case p@GreaterThan(l, r) => isValidRangeCondition(l, r,
left, right, joinKeys) match {
+ case "asis" => rangePreds.add(p); Some(GreaterThan(l, r))
+ case "vs" => rangePreds.add(p); Some(LessThan(r, l))
+ case _ => None
+ }
+ case p@GreaterThanOrEqual(l, r) =>
+ isValidRangeCondition(l, r, left, right, joinKeys) match {
+ case "asis" => rangePreds.add(p);
Some(GreaterThanOrEqual(l, r))
+ case "vs" => rangePreds.add(p); Some(LessThanOrEqual(r,
l))
+ case _ => None
+ }
+ case _ => None
+ }
+ }
+ else {
+ Nil
+ }
+
+ // Only using secondary join optimization when both lower and
upper conditions
+ // are specified (e.g. t1.a < t2.b + x and t1.a > t2.b - x)
+ if(rangeConditions.size != 2 ||
+ // Looking for one < and one > comparison:
+ rangeConditions.filter(x => x.isInstanceOf[LessThan] ||
+ x.isInstanceOf[LessThanOrEqual]).size == 0 ||
+ rangeConditions.filter(x => x.isInstanceOf[GreaterThan] ||
+ x.isInstanceOf[GreaterThanOrEqual]).size == 0 ||
+ // Check if both comparisons reference the same columns:
+ rangeConditions.flatMap(c =>
c.left.references.toSeq.distinct).distinct.size != 1 ||
+ rangeConditions.flatMap(c =>
c.right.references.toSeq.distinct).distinct.size != 1) {
+ logDebug("Inner range optimization conditions not met. Clearing
range conditions")
+ rangeConditions = Nil
+ rangePreds.clear()
+ }
+
+ Some((joinType, leftKeys, rightKeys, rangeConditions,
+
otherPredicates.filterNot(rangePreds.contains(_)).reduceOption(And), left,
right))
} else {
None
}
case _ => None
}
+
+ private def isValidRangeCondition(l : Expression, r : Expression,
+ left : LogicalPlan, right :
LogicalPlan,
+ joinKeys : Seq[(Expression,
Expression)]) = {
+ val (lattrs, rattrs) = (l.references.toSeq, r.references.toSeq)
+ if(lattrs.size != 1 || rattrs.size != 1) {
+ "none"
+ }
+ else if (canEvaluate(l, left) && canEvaluate(r, right)) {
+ val equiset = joinKeys.filter{ case (ljk : Expression, rjk :
Expression) =>
+ ljk.references.toSeq.contains(lattrs(0)) &&
rjk.references.toSeq.contains(rattrs(0)) }
+ if (equiset.isEmpty) {
+ "asis"
--- End diff --
can we return something more meaningful than this string? Maybe a
`Option(Bool)` in enough which tells whether to reverse or not.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
