[GitHub] [spark] cloud-fan commented on a diff in pull request #38799: [SPARK-37099][SQL] Introduce the group limit of Window for rank-based filter to optimize top-k computation

[GitBox <gi...@apache.org>]

cloud-fan commented on code in PR #38799:
URL: https://github.com/apache/spark/pull/38799#discussion_r1036828144


##########
sql/core/src/main/scala/org/apache/spark/sql/execution/SparkStrategies.scala:
##########
@@ -627,6 +627,87 @@ abstract class SparkStrategies extends QueryPlanner[SparkPlan] {
     }
   }
 
+  /**
+   * Optimize the filter based on rank-like window function by reduce not required rows.
+   * This rule optimizes the following cases:
+   * {{{
+   *   SELECT *, ROW_NUMBER() OVER(PARTITION BY k ORDER BY a) AS rn FROM Tab1 WHERE rn = 5
+   *   SELECT *, ROW_NUMBER() OVER(PARTITION BY k ORDER BY a) AS rn FROM Tab1 WHERE 5 = rn
+   *   SELECT *, ROW_NUMBER() OVER(PARTITION BY k ORDER BY a) AS rn FROM Tab1 WHERE rn < 5
+   *   SELECT *, ROW_NUMBER() OVER(PARTITION BY k ORDER BY a) AS rn FROM Tab1 WHERE 5 > rn
+   *   SELECT *, ROW_NUMBER() OVER(PARTITION BY k ORDER BY a) AS rn FROM Tab1 WHERE rn <= 5
+   *   SELECT *, ROW_NUMBER() OVER(PARTITION BY k ORDER BY a) AS rn FROM Tab1 WHERE 5 >= rn
+   * }}}
+   */
+  object WindowGroupLimit extends Strategy with PredicateHelper {
+
+    /**
+     * Extract all the limit values from predicates.
+     */
+    def extractLimits(condition: Expression, attr: Attribute): Option[Int] = {
+      val limits = splitConjunctivePredicates(condition).collect {
+        case EqualTo(IntegerLiteral(limit), e) if e.semanticEquals(attr) => limit
+        case EqualTo(e, IntegerLiteral(limit)) if e.semanticEquals(attr) => limit
+        case LessThan(e, IntegerLiteral(limit)) if e.semanticEquals(attr) => limit - 1
+        case GreaterThan(IntegerLiteral(limit), e) if e.semanticEquals(attr) => limit - 1
+        case LessThanOrEqual(e, IntegerLiteral(limit)) if e.semanticEquals(attr) => limit
+        case GreaterThanOrEqual(IntegerLiteral(limit), e) if e.semanticEquals(attr) => limit
+      }
+
+      if (limits.nonEmpty) Some(limits.min) else None
+    }
+
+    private def supports(
+        windowExpressions: Seq[NamedExpression]): Boolean = windowExpressions.exists {
+      case Alias(WindowExpression(_: Rank | _: DenseRank | _: RowNumber, WindowSpecDefinition(_, _,
+      SpecifiedWindowFrame(RowFrame, UnboundedPreceding, CurrentRow))), _) => true
+      case _ => false
+    }
+
+    def apply(plan: LogicalPlan): Seq[SparkPlan] = {
+      if (conf.windowGroupLimitThreshold == -1) return Nil
+
+      plan match {
+        case filter @ Filter(condition,
+          window @ logical.Window(windowExpressions, partitionSpec, orderSpec, child))
+          if !child.isInstanceOf[logical.Window] &&
+            supports(windowExpressions) && orderSpec.nonEmpty =>

Review Comment:
   do we really require the window to only contain rank like functions?



-- 
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: reviews-unsubscribe@spark.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org