Github user cloud-fan commented on a diff in the pull request: https://github.com/apache/spark/pull/19781#discussion_r151967403 --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/joins/BroadcastHashJoinExec.scala --- @@ -83,11 +83,12 @@ case class BroadcastHashJoinExec( // them in a buffer, because these result rows share one UnsafeRow instance. Note that here // we wait for the broadcast to be finished, which is a no-op because it's already finished // when we wait it in `doProduce`. - !buildPlan.executeBroadcast[HashedRelation]().value.keyIsUnique + streamedPlan.asInstanceOf[CodegenSupport].needCopyResult || + !buildPlan.executeBroadcast[HashedRelation]().value.keyIsUnique // Other joins types(semi, anti, existence) can at most produce one result row for one input // row from the streamed side, so no need to copy the result rows. --- End diff -- and the 2 comments becomes: ``` case _: InnerLike | LeftOuter | RightOuter => // For inner and outer joins, one row from the streamed side may produce multiple result rows, // if the build side has duplicated keys. Note that here ... case _ => Other joins types(semi, anti, existence) can at most produce one result row for one input row from the streamed side. ```
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org