Github user markhamstra commented on a diff in the pull request:
https://github.com/apache/spark/pull/1362#discussion_r15076386
--- Diff: core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala
---
@@ -1107,7 +1106,6 @@ class DAGScheduler(
case shufDep: ShuffleDependency[_, _, _] =>
val mapStage = getShuffleMapStage(shufDep, stage.jobId)
if (!mapStage.isAvailable) {
- visitedStages += mapStage
--- End diff --
@mateiz It looks to me that this is what was always intended, but has
actually been missing the `visitedStages` check for the past couple of years:
```scala
if (!mapStage.isAvailable && !visitedStages(mapStage)) {
visitedStages += mapStage
visit(mapStage.rdd)
} // Otherwise there's no need to follow the dependency back
```
Making that change works for me in the sense that all of the test suites
continue to pass, but I haven't yet got any relative performance numbers.
Proabably needs to go in a separate JIRA & PR.
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
