Github user marmbrus commented on a diff in the pull request: https://github.com/apache/spark/pull/1147#discussion_r15725487 --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/joins.scala --- @@ -137,6 +137,185 @@ trait HashJoin { } /** + * Constant Value for Binary Join Node + */ +object HashOuterJoin { + val DUMMY_LIST = Seq[Row](null) + val EMPTY_LIST = Seq[Row]() +} + +/** + * :: DeveloperApi :: + * Performs a hash based outer join for two child relations by shuffling the data using + * the join keys. This operator requires loading the associated partition in both side into memory. + */ +@DeveloperApi +case class HashOuterJoin( + leftKeys: Seq[Expression], + rightKeys: Seq[Expression], + joinType: JoinType, + condition: Option[Expression], + left: SparkPlan, + right: SparkPlan) extends BinaryNode { + + override def outputPartitioning: Partitioning = left.outputPartitioning + + override def requiredChildDistribution = + ClusteredDistribution(leftKeys) :: ClusteredDistribution(rightKeys) :: Nil + + def output = left.output ++ right.output + + // TODO we need to rewrite all of the iterators with our own implementation instead of the Scala + // iterator for performance purpose. + + private[this] def leftOuterIterator( + key: Row, leftIter: Iterable[Row], rightIter: Iterable[Row]): Iterator[Row] = { + val joinedRow = new JoinedRow() + val rightNullRow = new GenericRow(right.output.length) + val boundCondition = + condition.map(newPredicate(_, left.output ++ right.output)).getOrElse((row: Row) => true) + + leftIter.iterator.flatMap { l => + joinedRow.withLeft(l) + var matched = false + (if (!key.anyNull) rightIter.collect { case r if (boundCondition(joinedRow.withRight(r))) => + matched = true + joinedRow.copy + } else { + Nil + }) ++ HashOuterJoin.DUMMY_LIST.filter(_ => !matched).map( _ => { + // HashOuterJoin.DUMMY_LIST.filter(_ => !matched) is a tricky way to add additional row, --- End diff -- Oh wow... thanks for the explanation. That is even more subtle than I thought.
--- 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. ---