c21 commented on a change in pull request #32476:
URL: https://github.com/apache/spark/pull/32476#discussion_r629027318
##########
File path:
sql/core/src/main/scala/org/apache/spark/sql/execution/joins/SortMergeJoinExec.scala
##########
@@ -418,115 +443,140 @@ case class SortMergeJoinExec(
// Inline mutable state since not many join operations in a task
val matches = ctx.addMutableState(clsName, "matches",
v => s"$v = new $clsName($inMemoryThreshold, $spillThreshold);",
forceInline = true)
- // Copy the left keys as class members so they could be used in next
function call.
- val matchedKeyVars = copyKeys(ctx, leftKeyVars)
+ // Copy the streamed keys as class members so they could be used in next
function call.
+ val matchedKeyVars = copyKeys(ctx, streamedKeyVars)
+
+ // Handle the case when streamed rows has any NULL keys.
+ val handleStreamedAnyNull = joinType match {
+ case _: InnerLike =>
+ // Skip streamed row.
+ s"""
+ |$streamedRow = null;
+ |continue;
+ """.stripMargin
+ case LeftOuter | RightOuter =>
+ // Eagerly return streamed row.
+ s"""
+ |if (!$matches.isEmpty()) {
+ | $matches.clear();
+ |}
+ |return false;
+ """.stripMargin
+ case x =>
+ throw new IllegalArgumentException(
+ s"SortMergeJoin.genScanner should not take $x as the JoinType")
+ }
- ctx.addNewFunction("findNextInnerJoinRows",
+ // Handle the case when streamed keys less than buffered keys.
+ val handleStreamedLessThanBuffered = joinType match {
+ case _: InnerLike =>
+ // Skip streamed row.
+ s"$streamedRow = null;"
+ case LeftOuter | RightOuter =>
+ // Eagerly return with streamed row.
+ "return false;"
+ case x =>
+ throw new IllegalArgumentException(
+ s"SortMergeJoin.genScanner should not take $x as the JoinType")
+ }
+
+ ctx.addNewFunction("findNextJoinRows",
s"""
- |private boolean findNextInnerJoinRows(
- | scala.collection.Iterator leftIter,
- | scala.collection.Iterator rightIter) {
- | $leftRow = null;
+ |private boolean findNextJoinRows(
Review comment:
@maropu - No I think we need buffer anyway. The buffered rows has same
join keys with current streamed row. But there can be multiple followed
streamed rows having same join keys, as the buffered rows. Even though buffered
rows cannot match condition with current streamed row, they may match condition
with followed streamed rows. I think this is how current sort merge join
(code-gen & iterator) is designed.
--
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
