Github user squito commented on a diff in the pull request:
https://github.com/apache/spark/pull/5636#discussion_r29691601
--- Diff:
core/src/test/scala/org/apache/spark/scheduler/DAGSchedulerSuite.scala ---
@@ -475,7 +475,148 @@ class DAGSchedulerSuite
assert(results === Map(0 -> 42, 1 -> 43))
assertDataStructuresEmpty()
}
+
+ test("Multiple consecutive stage failures should lead to stage being
aborted.") {
+ // Create a new Listener to confirm that the listenerBus sees the
JobEnd message
+ // when we abort the stage. This message will also be consumed by the
EventLoggingListener
+ // so this will propagate up to the user.
+ var ended = false
+ var jobResult : JobResult = null
+ class EndListener extends SparkListener {
+ override def onJobEnd(jobEnd: SparkListenerJobEnd): Unit = {
+ jobResult = jobEnd.jobResult
+ ended = true
+ }
+ }
+
+ sc.listenerBus.addListener(new EndListener())
+
+ val shuffleMapRdd = new MyRDD(sc, 2, Nil)
+ val shuffleDep = new ShuffleDependency(shuffleMapRdd, null)
+ val shuffleId = shuffleDep.shuffleId
+ val reduceRdd = new MyRDD(sc, 2, List(shuffleDep))
+ submit(reduceRdd, Array(0, 1))
+
+ complete(taskSets(0), Seq(
+ (Success, makeMapStatus("hostA", 1)),
+ (Success, makeMapStatus("hostB", 1))))
+
+ for (x <- 1 to Stage.MAX_STAGE_FAILURES) {
+ // the 2nd ResultTask failed
+ complete(taskSets(1), Seq(
+ (Success, 42),
+ (FetchFailed(makeBlockManagerId("hostA"), shuffleId, 0, 0,
"ignored"), null)))
+
+ scheduler.resubmitFailedStages()
+ if (x < Stage.MAX_STAGE_FAILURES) {
+ assert(scheduler.runningStages.nonEmpty)
+ assert(!ended)
+ } else {
+ // Stage has been aborted and removed from running stages
+ assertDataStructuresEmpty()
+ sc.listenerBus.waitUntilEmpty(1000)
+ assert(ended)
+ assert(jobResult.isInstanceOf[JobFailed])
+ }
+ }
+ }
+
+
+ test("Multiple consecutive Fetch failures in a stage triggers an
abort.") {
+ // Create a new Listener to confirm that the listenerBus sees the
JobEnd message
+ // when we abort the stage. This message will also be consumed by the
EventLoggingListener
+ // so this will propagate up to the user.
+ var ended = false
+ var jobResult : JobResult = null
+ class EndListener extends SparkListener {
+ override def onJobEnd(jobEnd: SparkListenerJobEnd): Unit = {
+ jobResult = jobEnd.jobResult
+ ended = true
+ }
+ }
+
+ sc.listenerBus.addListener(new EndListener())
+
+ val shuffleMapRdd = new MyRDD(sc, 8, Nil)
+ val shuffleDep = new ShuffleDependency(shuffleMapRdd, null)
+ val shuffleId = shuffleDep.shuffleId
+ val reduceRdd = new MyRDD(sc, 8, List(shuffleDep))
+ submit(reduceRdd, Array(0, 1, 2, 3, 4, 5, 6, 7))
+
+ complete(taskSets(0), Seq(
+ (Success, makeMapStatus("hostA", 1)),
+ (Success, makeMapStatus("hostA", 1)),
+ (Success, makeMapStatus("hostA", 1)),
+ (Success, makeMapStatus("hostA", 1)),
+ (Success, makeMapStatus("hostA", 1)),
+ (Success, makeMapStatus("hostA", 1)),
+ (Success, makeMapStatus("hostA", 1)),
+ (Success, makeMapStatus("hostB", 1))))
+
+ complete(taskSets(1), Seq(
+ (Success, 42),
+ (FetchFailed(makeBlockManagerId("hostA"), shuffleId, 0, 0,
"ignored"), null),
+ (FetchFailed(makeBlockManagerId("hostA"), shuffleId, 0, 0,
"ignored1"), null),
+ (FetchFailed(makeBlockManagerId("hostA"), shuffleId, 0, 0,
"ignored2"), null),
+ (FetchFailed(makeBlockManagerId("hostA"), shuffleId, 0, 0,
"ignored3"), null),
+ (FetchFailed(makeBlockManagerId("hostA"), shuffleId, 0, 0,
"ignored4"), null),
+ (FetchFailed(makeBlockManagerId("hostA"), shuffleId, 0, 0,
"ignored5"), null),
+ (FetchFailed(makeBlockManagerId("hostA"), shuffleId, 0, 0,
"ignored6"), null)))
+
+ scheduler.resubmitFailedStages()
+ assertDataStructuresEmpty()
+ sc.listenerBus.waitUntilEmpty(1000)
+ assert(ended)
+ assert(jobResult.isInstanceOf[JobFailed])
--- End diff --
I think in this case, we actually want to make sure the job *succeeds*.
The idea is that after a bunch of fetch failures on attempt 0, we shouldn't
give up. Attempt 1 should still get tried. And if our tests has all tasks in
attempt 1 complete successfully, the job should pass.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
