Github user jiangxb1987 commented on a diff in the pull request: https://github.com/apache/spark/pull/21758#discussion_r207087848 --- Diff: core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala --- @@ -1349,6 +1339,29 @@ class DAGScheduler( s"longer running") } + if (mapStage.rdd.isBarrier()) { --- End diff -- Previously I was thinking that by unregistering shuffle output, we may avoid modify the submit missing tasks logic. Now I realized you have to launch all the tasks for taskSet of a barrier stage anyway, so maybe the approach you mentioned is cleaner, I'll try to submit a follow up PR on that. Thanks!
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org