Github user squito commented on a diff in the pull request:
https://github.com/apache/spark/pull/4055#discussion_r35220904
--- Diff:
core/src/test/scala/org/apache/spark/scheduler/DAGSchedulerSuite.scala ---
@@ -739,6 +742,88 @@ class DAGSchedulerSuite
assertDataStructuresEmpty()
}
+ test("verify not submit next stage while not have registered mapStatus")
{
+ val firstRDD = new MyRDD(sc, 3, Nil)
+ val firstShuffleDep = new ShuffleDependency(firstRDD, null)
+ val firstShuffleId = firstShuffleDep.shuffleId
+ val shuffleMapRdd = new MyRDD(sc, 3, List(firstShuffleDep))
+ val shuffleDep = new ShuffleDependency(shuffleMapRdd, null)
+ val reduceRdd = new MyRDD(sc, 1, List(shuffleDep))
+ submit(reduceRdd, Array(0))
+
+ // things start out smoothly, stage 0 completes with no issues
+ complete(taskSets(0), Seq(
+ (Success, makeMapStatus("hostB", shuffleMapRdd.partitions.size)),
+ (Success, makeMapStatus("hostB", shuffleMapRdd.partitions.size)),
+ (Success, makeMapStatus("hostA", shuffleMapRdd.partitions.size))
+ ))
+
+ // then one executor dies, and a task fails in stage 1
+ runEvent(ExecutorLost("exec-hostA"))
+ runEvent(CompletionEvent(taskSets(1).tasks(0),
+ FetchFailed(null, firstShuffleId, 2, 0, "Fetch failed"),
+ null, null, createFakeTaskInfo(), null))
+
+ // so we resubmit stage 0, which completes happily
+ Thread.sleep(1000)
+ val stage0Resubmit = taskSets(2)
+ assert(stage0Resubmit.stageId == 0)
+ assert(stage0Resubmit.stageAttemptId === 1)
+ val task = stage0Resubmit.tasks(0)
+ assert(task.partitionId === 2)
+ runEvent(CompletionEvent(task, Success,
+ makeMapStatus("hostC", shuffleMapRdd.partitions.size), null,
createFakeTaskInfo(), null))
+
+ // now here is where things get tricky : we will now have a task set
representing
+ // the second attempt for stage 1, but we *also* have some tasks for
the first attempt for
+ // stage 1 still going
+ val stage1Resubmit = taskSets(3)
+ assert(stage1Resubmit.stageId == 1)
+ assert(stage1Resubmit.stageAttemptId === 1)
+ assert(stage1Resubmit.tasks.length === 3)
+
+ // we'll have some tasks finish from the first attempt, and some
finish from the second attempt,
+ // so that we actually have all stage outputs, though no attempt has
completed all its
+ // tasks
+ runEvent(CompletionEvent(taskSets(3).tasks(0), Success,
+ makeMapStatus("hostC", reduceRdd.partitions.size), null,
createFakeTaskInfo(), null))
+ runEvent(CompletionEvent(taskSets(3).tasks(1), Success,
+ makeMapStatus("hostC", reduceRdd.partitions.size), null,
createFakeTaskInfo(), null))
+ // late task finish from the first attempt
+ runEvent(CompletionEvent(taskSets(1).tasks(2), Success,
+ makeMapStatus("hostB", reduceRdd.partitions.size), null,
createFakeTaskInfo(), null))
+
+ // What should happen now is that we submit stage 2. However, we
might not see an error
+ // b/c of DAGScheduler's error handling (it tends to swallow errors
and just log them). But
+ // we can check some conditions.
+ // Note that the really important thing here is not so much that we
submit stage 2 *immediately*
+ // but that we don't end up with some error from these interleaved
completions. It would also
+ // be OK (though sub-optimal) if stage 2 simply waited until the
resubmission of stage 1 had
+ // all its tasks complete
+
+ // check that we have all the map output for stage 0 (it should have
been there even before
+ // the last round of completions from stage 1, but just to double
check it hasn't been messed
+ // up)
+ (0 until 3).foreach { reduceIdx =>
+ val arr = mapOutputTracker.getServerStatuses(0, reduceIdx)
+ assert(arr != null)
+ assert(arr.nonEmpty)
--- End diff --
we don't use `try / case e: Exception => fail("")` to fail tests when there
is an exception -- we just let the exception fail the test directly. You get
more info in the stack trace that way. So I think its better to just leave it
bare.
You could just put in a comment explaining what the point is:
```scala
// this would throw an exception if the map status hadn't been registered
mapOutputTracker.getMapSizesByExecutorId(0, reduceIdx)
```
I still slightly prefer leaving the asserts in there. Yes, they are kinda
pointless with the current behavior of `getMapSizesByExecutorId` -- but I'd
just like to be a bit more defensive, in case that behavior changes in the
future. (eg., maybe some future refactoring makes them stop throwing
exceptions for some reason).
Maybe to be very clear, you could include the asserts and more comments:
```scala
// this would throw an exception if the map status hadn't been registered
val statuses = mapOutputTracker.getMapSizesByExecutorId(0, reduceIdx)
// really we should have already thrown an exception rather than fail
either of these asserts, but just
// to be extra defensive let's double check the statuses are OK
assert(statuses != null)
assert(statuses.nonEmpty)
```
This is pretty minor, though, I don't feel strongly about it.
---
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]
