Github user squito commented on a diff in the pull request:
https://github.com/apache/spark/pull/15249#discussion_r81893799
--- Diff:
core/src/test/scala/org/apache/spark/scheduler/TaskSetManagerSuite.scala ---
@@ -809,32 +816,65 @@ class TaskSetManagerSuite extends SparkFunSuite with
LocalSparkContext with Logg
test("Kill other task attempts when one attempt belonging to the same
task succeeds") {
sc = new SparkContext("local", "test")
sched = new FakeTaskScheduler(sc, ("exec1", "host1"), ("exec2",
"host2"))
- val taskSet = FakeTask.createTaskSet(4)
+ val taskSet = FakeTask.createTaskSet(5)
// Set the speculation multiplier to be 0 so speculative tasks are
launched immediately
sc.conf.set("spark.speculation.multiplier", "0.0")
+ sc.conf.set("spark.speculation.quantile", "0.6")
val clock = new ManualClock()
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES,
clock)
val accumUpdatesByTask: Array[Seq[AccumulatorV2[_, _]]] =
taskSet.tasks.map { task =>
task.metrics.internalAccums
}
// Offer resources for 4 tasks to start
+ val tasks = new ArrayBuffer[TaskDescription]()
for ((k, v) <- List(
"exec1" -> "host1",
"exec1" -> "host1",
+ "exec1" -> "host1",
"exec2" -> "host2",
"exec2" -> "host2")) {
val taskOption = manager.resourceOffer(k, v, NO_PREF)
assert(taskOption.isDefined)
val task = taskOption.get
assert(task.executorId === k)
+ tasks += task
}
- assert(sched.startedTasks.toSet === Set(0, 1, 2, 3))
- // Complete the 3 tasks and leave 1 task in running
+ assert(sched.startedTasks.toSet === (0 until 5).toSet)
+ // Complete 3 tasks and leave 2 task in running
for (id <- Set(0, 1, 2)) {
manager.handleSuccessfulTask(id, createTaskResult(id,
accumUpdatesByTask(id)))
assert(sched.endedTasks(id) === Success)
}
+ def runningTaskForIndex(index: Int): TaskDescription = {
+ val t = tasks.find { task => task.index == index &&
!sched.endedTasks.contains(task.taskId) }
+ t match {
+ case Some(x) => x
+ case None =>
+ throw new RuntimeException(s"couldn't find index $index in " +
+ s"tasks: ${tasks.map{t => t.index -> t.taskId}} with
endedTasks:" +
+ s" ${sched.endedTasks.keys}")
+ }
+ }
+
+ // have each of the running tasks fail 3 times (not enough to abort
the stage)
+ (3 until 6).foreach { attempt =>
--- End diff --
it didn't *need* to, in the sense that the test would pass without these
changes. I was concerned about my concern to `TaskKilled.countTowardsFailure`
-- in particular, I wanted to make sure that when you have speculative tasks,
cancelling speculative tasks doesn't lead to aborting the taskset. This wasn't
tested at all before, so I expanded this test to cover that. So the additions
here are to make sure there are 4 task "failures", but one of them is just a
speculative task being canceled. The other changes to create 2 speculative
tasks are also just for better coverage -- I wanted the first successful
speculative task to *not* complete the taskset. Unfortunately that made some
of the bookkeeping more complicated.
In hindsight, its probably pointless to throw it all into the same test.
I'll break it out into another one. It'll need to do virtually everything of
the current test, so it will be redundant from a coverage perspective. But
hopefully the intent of the tests will be more clear.
---
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]
