Github user rxin commented on a diff in the pull request:
https://github.com/apache/spark/pull/11130#discussion_r52838460
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoin.scala
---
@@ -105,75 +107,129 @@ case class BroadcastHashJoin(
val broadcastRelation = Await.result(broadcastFuture, timeout)
streamedPlan.execute().mapPartitions { streamedIter =>
- val hashedRelation = broadcastRelation.value
-
TaskContext.get().taskMetrics().incPeakExecutionMemory(hashedRelation.getMemorySize)
- hashJoin(streamedIter, hashedRelation, numOutputRows)
+ val joinedRow = new JoinedRow()
+ val hashTable = broadcastRelation.value
+
TaskContext.get().taskMetrics().incPeakExecutionMemory(hashTable.getMemorySize)
+ val keyGenerator = streamSideKeyGenerator
+ val resultProj = createResultProjection
+
+ joinType match {
+ case Inner =>
+ hashJoin(streamedIter, hashTable, numOutputRows)
+
+ case LeftOuter =>
+ streamedIter.flatMap(currentRow => {
+ val rowKey = keyGenerator(currentRow)
+ joinedRow.withLeft(currentRow)
+ leftOuterIterator(rowKey, joinedRow, hashTable.get(rowKey),
resultProj, numOutputRows)
+ })
+
+ case RightOuter =>
+ streamedIter.flatMap(currentRow => {
+ val rowKey = keyGenerator(currentRow)
+ joinedRow.withRight(currentRow)
+ rightOuterIterator(rowKey, hashTable.get(rowKey), joinedRow,
resultProj, numOutputRows)
+ })
+
+ case x =>
+ throw new IllegalArgumentException(
+ s"BroadcastHashJoin should not take $x as the JoinType")
+ }
}
}
- private var broadcastRelation: Broadcast[HashedRelation] = _
- // the term for hash relation
- private var relationTerm: String = _
-
override def upstream(): RDD[InternalRow] = {
streamedPlan.asInstanceOf[CodegenSupport].upstream()
}
override def doProduce(ctx: CodegenContext): String = {
+ streamedPlan.asInstanceOf[CodegenSupport].produce(ctx, this)
+ }
+
+ override def doConsume(ctx: CodegenContext, input: Seq[ExprCode]):
String = {
+ if (joinType == Inner) {
+ codegenInner(ctx, input)
+ } else {
+ // LeftOuter and RightOuter
+ codegenOuter(ctx, input)
+ }
+ }
+
+ private def prepareBroadcast(ctx: CodegenContext):
(Broadcast[HashedRelation], String) = {
// create a name for HashedRelation
- broadcastRelation = Await.result(broadcastFuture, timeout)
+ val broadcastRelation = Await.result(broadcastFuture, timeout)
val broadcast = ctx.addReferenceObj("broadcast", broadcastRelation)
- relationTerm = ctx.freshName("relation")
+ val relationTerm = ctx.freshName("relation")
val clsName = broadcastRelation.value.getClass.getName
ctx.addMutableState(clsName, relationTerm,
s"""
| $relationTerm = ($clsName) $broadcast.value();
| incPeakExecutionMemory($relationTerm.getMemorySize());
""".stripMargin)
-
- s"""
- | ${streamedPlan.asInstanceOf[CodegenSupport].produce(ctx, this)}
- """.stripMargin
+ (broadcastRelation, relationTerm)
}
- override def doConsume(ctx: CodegenContext, input: Seq[ExprCode]):
String = {
- // generate the key as UnsafeRow or Long
+ private def genJoinKey(ctx: CodegenContext, input: Seq[ExprCode]):
(ExprCode, String) = {
ctx.currentVars = input
- val (keyVal, anyNull) = if (canJoinKeyFitWithinLong) {
+ if (canJoinKeyFitWithinLong) {
+ // generate the join key as Long
val expr = rewriteKeyExpr(streamedKeys).head
val ev = BindReferences.bindReference(expr,
streamedPlan.output).gen(ctx)
(ev, ev.isNull)
} else {
+ // generate the join key as UnsafeRow
val keyExpr = streamedKeys.map(BindReferences.bindReference(_,
streamedPlan.output))
val ev = GenerateUnsafeProjection.createCode(ctx, keyExpr)
(ev, s"${ev.value}.anyNull()")
}
+ }
- // find the matches from HashedRelation
- val matched = ctx.freshName("matched")
-
- // create variables for output
+ private def genBuildSideVars(ctx: CodegenContext, matched: String):
Seq[ExprCode] = {
ctx.currentVars = null
ctx.INPUT_ROW = matched
- val buildColumns = buildPlan.output.zipWithIndex.map { case (a, i) =>
- BoundReference(i, a.dataType, a.nullable).gen(ctx)
+ buildPlan.output.zipWithIndex.map { case (a, i) =>
+ val ev = BoundReference(i, a.dataType, a.nullable).gen(ctx)
+ if (joinType == Inner) {
+ ev
+ } else {
+ // the variables are needed even there is no matched rows
--- End diff --
I'm really confused by this. What is this doing?
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
