squito commented on a change in pull request #23677: [SPARK-26755][SCHEDULER] :
Optimize Spark Scheduler to dequeue speculative tasks…
URL: https://github.com/apache/spark/pull/23677#discussion_r303604083
##########
File path:
core/src/test/scala/org/apache/spark/scheduler/TaskSetManagerSuite.scala
##########
@@ -1655,4 +1657,70 @@ class TaskSetManagerSuite extends SparkFunSuite with
LocalSparkContext with Logg
// get removed inside TaskSchedulerImpl later.
assert(availableResources(GPU) sameElements Array("0", "1", "2", "3"))
}
+
+ test("SPARK-26755 Ensure that a speculative task is submitted only once for
execution") {
+ sc = new SparkContext("local", "test")
+ sched = new FakeTaskScheduler(sc, ("exec1", "host1"), ("exec2", "host2"))
+ val taskSet = FakeTask.createTaskSet(4)
+ // Set the speculation multiplier to be 0 so speculative tasks are
launched immediately
+ sc.conf.set(config.SPECULATION_MULTIPLIER, 0.0)
+ sc.conf.set(config.SPECULATION_ENABLED, true)
+ sc.conf.set(config.SPECULATION_QUANTILE, 0.5)
+ val clock = new ManualClock()
+ val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock
= clock)
+ val accumUpdatesByTask: Array[Seq[AccumulatorV2[_, _]]] =
taskSet.tasks.map { task =>
+ task.metrics.internalAccums
+ }
+ // Offer resources for 4 tasks to start
+ for ((k, v) <- List(
+ "exec1" -> "host1",
+ "exec1" -> "host1",
+ "exec2" -> "host2",
+ "exec2" -> "host2")) {
Review comment:
nit: would you mind switching this to use `foreach` instead? I think some
of the early scheduler code had more for loops, but the convention has since
changed -- (scala's for loops actually have some weird wrinkles). Also prefer
`Seq` over `List` in general.
```scala
// 4 offers -- 2 on each exec
Seq("exec1" -> "host1", "exec2" -> "host2").foreach { case (exec, host) =>
(0 until 2).foreach { _ =>
val taskOption = manager.resourceOffer(k, v, NO_PREF)
assert(taskOption.isDefined)
assert(taskOption.get.executorId === k)
}
}
```
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
