1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
|
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.spark.scheduler
import java.util.Random
import scala.collection.Map
import scala.collection.mutable
import scala.collection.mutable.ArrayBuffer
import org.apache.spark._
import org.apache.spark.internal.Logging
import org.apache.spark.util.ManualClock
class FakeDAGScheduler(sc: SparkContext, taskScheduler: FakeTaskScheduler)
extends DAGScheduler(sc) {
override def taskStarted(task: Task[_], taskInfo: TaskInfo) {
taskScheduler.startedTasks += taskInfo.index
}
override def taskEnded(
task: Task[_],
reason: TaskEndReason,
result: Any,
accumUpdates: Seq[AccumulableInfo],
taskInfo: TaskInfo) {
taskScheduler.endedTasks(taskInfo.index) = reason
}
override def executorAdded(execId: String, host: String) {}
override def executorLost(execId: String) {}
override def taskSetFailed(
taskSet: TaskSet,
reason: String,
exception: Option[Throwable]): Unit = {
taskScheduler.taskSetsFailed += taskSet.id
}
}
// Get the rack for a given host
object FakeRackUtil {
private val hostToRack = new mutable.HashMap[String, String]()
def cleanUp() {
hostToRack.clear()
}
def assignHostToRack(host: String, rack: String) {
hostToRack(host) = rack
}
def getRackForHost(host: String): Option[String] = {
hostToRack.get(host)
}
}
/**
* A mock TaskSchedulerImpl implementation that just remembers information about tasks started and
* feedback received from the TaskSetManagers. Note that it's important to initialize this with
* a list of "live" executors and their hostnames for isExecutorAlive and hasExecutorsAliveOnHost
* to work, and these are required for locality in TaskSetManager.
*/
class FakeTaskScheduler(sc: SparkContext, liveExecutors: (String, String)* /* execId, host */)
extends TaskSchedulerImpl(sc)
{
val startedTasks = new ArrayBuffer[Long]
val endedTasks = new mutable.HashMap[Long, TaskEndReason]
val finishedManagers = new ArrayBuffer[TaskSetManager]
val taskSetsFailed = new ArrayBuffer[String]
val executors = new mutable.HashMap[String, String]
for ((execId, host) <- liveExecutors) {
addExecutor(execId, host)
}
for ((execId, host) <- liveExecutors; rack <- getRackForHost(host)) {
hostsByRack.getOrElseUpdate(rack, new mutable.HashSet[String]()) += host
}
dagScheduler = new FakeDAGScheduler(sc, this)
def removeExecutor(execId: String) {
executors -= execId
val host = executorIdToHost.get(execId)
assert(host != None)
val hostId = host.get
val executorsOnHost = executorsByHost(hostId)
executorsOnHost -= execId
for (rack <- getRackForHost(hostId); hosts <- hostsByRack.get(rack)) {
hosts -= hostId
if (hosts.isEmpty) {
hostsByRack -= rack
}
}
}
override def taskSetFinished(manager: TaskSetManager): Unit = finishedManagers += manager
override def isExecutorAlive(execId: String): Boolean = executors.contains(execId)
override def hasExecutorsAliveOnHost(host: String): Boolean = executors.values.exists(_ == host)
override def hasHostAliveOnRack(rack: String): Boolean = {
hostsByRack.get(rack) != None
}
def addExecutor(execId: String, host: String) {
executors.put(execId, host)
val executorsOnHost = executorsByHost.getOrElseUpdate(host, new mutable.HashSet[String])
executorsOnHost += execId
executorIdToHost += execId -> host
for (rack <- getRackForHost(host)) {
hostsByRack.getOrElseUpdate(rack, new mutable.HashSet[String]()) += host
}
}
override def getRackForHost(value: String): Option[String] = FakeRackUtil.getRackForHost(value)
}
/**
* A Task implementation that results in a large serialized task.
*/
class LargeTask(stageId: Int) extends Task[Array[Byte]](stageId, 0, 0, Seq.empty) {
val randomBuffer = new Array[Byte](TaskSetManager.TASK_SIZE_TO_WARN_KB * 1024)
val random = new Random(0)
random.nextBytes(randomBuffer)
override def runTask(context: TaskContext): Array[Byte] = randomBuffer
override def preferredLocations: Seq[TaskLocation] = Seq[TaskLocation]()
}
class TaskSetManagerSuite extends SparkFunSuite with LocalSparkContext with Logging {
import TaskLocality.{ANY, PROCESS_LOCAL, NO_PREF, NODE_LOCAL, RACK_LOCAL}
private val conf = new SparkConf
val LOCALITY_WAIT_MS = conf.getTimeAsMs("spark.locality.wait", "3s")
val MAX_TASK_FAILURES = 4
override def beforeEach() {
super.beforeEach()
FakeRackUtil.cleanUp()
}
test("TaskSet with no preferences") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"))
val taskSet = FakeTask.createTaskSet(1)
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
val accumUpdates = taskSet.tasks.head.initialAccumulators.map { a => a.toInfo(Some(0L), None) }
// Offer a host with NO_PREF as the constraint,
// we should get a nopref task immediately since that's what we only have
val taskOption = manager.resourceOffer("exec1", "host1", NO_PREF)
assert(taskOption.isDefined)
// Tell it the task has finished
manager.handleSuccessfulTask(0, createTaskResult(0, accumUpdates))
assert(sched.endedTasks(0) === Success)
assert(sched.finishedManagers.contains(manager))
}
test("multiple offers with no preferences") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"))
val taskSet = FakeTask.createTaskSet(3)
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES)
val accumUpdatesByTask: Array[Seq[AccumulableInfo]] = taskSet.tasks.map { task =>
task.initialAccumulators.map { a => a.toInfo(Some(0L), None) }
}
// First three offers should all find tasks
for (i <- 0 until 3) {
val taskOption = manager.resourceOffer("exec1", "host1", NO_PREF)
assert(taskOption.isDefined)
val task = taskOption.get
assert(task.executorId === "exec1")
}
assert(sched.startedTasks.toSet === Set(0, 1, 2))
// Re-offer the host -- now we should get no more tasks
assert(manager.resourceOffer("exec1", "host1", NO_PREF) === None)
// Finish the first two tasks
manager.handleSuccessfulTask(0, createTaskResult(0, accumUpdatesByTask(0)))
manager.handleSuccessfulTask(1, createTaskResult(1, accumUpdatesByTask(1)))
assert(sched.endedTasks(0) === Success)
assert(sched.endedTasks(1) === Success)
assert(!sched.finishedManagers.contains(manager))
// Finish the last task
manager.handleSuccessfulTask(2, createTaskResult(2, accumUpdatesByTask(2)))
assert(sched.endedTasks(2) === Success)
assert(sched.finishedManagers.contains(manager))
}
test("skip unsatisfiable locality levels") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("execA", "host1"), ("execC", "host2"))
val taskSet = FakeTask.createTaskSet(1, Seq(TaskLocation("host1", "execB")))
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// An executor that is not NODE_LOCAL should be rejected.
assert(manager.resourceOffer("execC", "host2", ANY) === None)
// Because there are no alive PROCESS_LOCAL executors, the base locality level should be
// NODE_LOCAL. So, we should schedule the task on this offered NODE_LOCAL executor before
// any of the locality wait timers expire.
assert(manager.resourceOffer("execA", "host1", ANY).get.index === 0)
}
test("basic delay scheduling") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"), ("exec2", "host2"))
val taskSet = FakeTask.createTaskSet(4,
Seq(TaskLocation("host1", "exec1")),
Seq(TaskLocation("host2", "exec2")),
Seq(TaskLocation("host1"), TaskLocation("host2", "exec2")),
Seq() // Last task has no locality prefs
)
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// First offer host1, exec1: first task should be chosen
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 0)
assert(manager.resourceOffer("exec1", "host1", PROCESS_LOCAL) == None)
clock.advance(LOCALITY_WAIT_MS)
// Offer host1, exec1 again, at NODE_LOCAL level: the node local (task 2) should
// get chosen before the noPref task
assert(manager.resourceOffer("exec1", "host1", NODE_LOCAL).get.index == 2)
// Offer host2, exec3 again, at NODE_LOCAL level: we should choose task 2
assert(manager.resourceOffer("exec2", "host2", NODE_LOCAL).get.index == 1)
// Offer host2, exec3 again, at NODE_LOCAL level: we should get noPref task
// after failing to find a node_Local task
assert(manager.resourceOffer("exec2", "host2", NODE_LOCAL) == None)
clock.advance(LOCALITY_WAIT_MS)
assert(manager.resourceOffer("exec2", "host2", NO_PREF).get.index == 3)
}
test("we do not need to delay scheduling when we only have noPref tasks in the queue") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"), ("exec3", "host2"))
val taskSet = FakeTask.createTaskSet(3,
Seq(TaskLocation("host1", "exec1")),
Seq(TaskLocation("host2", "exec3")),
Seq() // Last task has no locality prefs
)
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// First offer host1, exec1: first task should be chosen
assert(manager.resourceOffer("exec1", "host1", PROCESS_LOCAL).get.index === 0)
assert(manager.resourceOffer("exec3", "host2", PROCESS_LOCAL).get.index === 1)
assert(manager.resourceOffer("exec3", "host2", NODE_LOCAL) == None)
assert(manager.resourceOffer("exec3", "host2", NO_PREF).get.index === 2)
}
test("delay scheduling with fallback") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc,
("exec1", "host1"), ("exec2", "host2"), ("exec3", "host3"))
val taskSet = FakeTask.createTaskSet(5,
Seq(TaskLocation("host1")),
Seq(TaskLocation("host2")),
Seq(TaskLocation("host2")),
Seq(TaskLocation("host3")),
Seq(TaskLocation("host2"))
)
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// First offer host1: first task should be chosen
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 0)
// Offer host1 again: nothing should get chosen
assert(manager.resourceOffer("exec1", "host1", ANY) === None)
clock.advance(LOCALITY_WAIT_MS)
// Offer host1 again: second task (on host2) should get chosen
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 1)
// Offer host1 again: third task (on host2) should get chosen
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 2)
// Offer host2: fifth task (also on host2) should get chosen
assert(manager.resourceOffer("exec2", "host2", ANY).get.index === 4)
// Now that we've launched a local task, we should no longer launch the task for host3
assert(manager.resourceOffer("exec2", "host2", ANY) === None)
clock.advance(LOCALITY_WAIT_MS)
// After another delay, we can go ahead and launch that task non-locally
assert(manager.resourceOffer("exec2", "host2", ANY).get.index === 3)
}
test("delay scheduling with failed hosts") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"), ("exec2", "host2"),
("exec3", "host3"))
val taskSet = FakeTask.createTaskSet(3,
Seq(TaskLocation("host1")),
Seq(TaskLocation("host2")),
Seq(TaskLocation("host3"))
)
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// First offer host1: first task should be chosen
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 0)
// After this, nothing should get chosen, because we have separated tasks with unavailable
// preference from the noPrefPendingTasks
assert(manager.resourceOffer("exec1", "host1", ANY) === None)
// Now mark host2 as dead
sched.removeExecutor("exec2")
manager.executorLost("exec2", "host2", SlaveLost())
// nothing should be chosen
assert(manager.resourceOffer("exec1", "host1", ANY) === None)
clock.advance(LOCALITY_WAIT_MS * 2)
// task 1 and 2 would be scheduled as nonLocal task
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 1)
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 2)
// all finished
assert(manager.resourceOffer("exec1", "host1", ANY) === None)
assert(manager.resourceOffer("exec2", "host2", ANY) === None)
}
test("task result lost") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"))
val taskSet = FakeTask.createTaskSet(1)
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 0)
// Tell it the task has finished but the result was lost.
manager.handleFailedTask(0, TaskState.FINISHED, TaskResultLost)
assert(sched.endedTasks(0) === TaskResultLost)
// Re-offer the host -- now we should get task 0 again.
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 0)
}
test("repeated failures lead to task set abortion") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"))
val taskSet = FakeTask.createTaskSet(1)
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// Fail the task MAX_TASK_FAILURES times, and check that the task set is aborted
// after the last failure.
(1 to manager.maxTaskFailures).foreach { index =>
val offerResult = manager.resourceOffer("exec1", "host1", ANY)
assert(offerResult.isDefined,
"Expect resource offer on iteration %s to return a task".format(index))
assert(offerResult.get.index === 0)
manager.handleFailedTask(offerResult.get.taskId, TaskState.FINISHED, TaskResultLost)
if (index < MAX_TASK_FAILURES) {
assert(!sched.taskSetsFailed.contains(taskSet.id))
} else {
assert(sched.taskSetsFailed.contains(taskSet.id))
}
}
}
test("executors should be blacklisted after task failure, in spite of locality preferences") {
val rescheduleDelay = 300L
val conf = new SparkConf().
set("spark.scheduler.executorTaskBlacklistTime", rescheduleDelay.toString).
// don't wait to jump locality levels in this test
set("spark.locality.wait", "0")
sc = new SparkContext("local", "test", conf)
// two executors on same host, one on different.
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"),
("exec1.1", "host1"), ("exec2", "host2"))
// affinity to exec1 on host1 - which we will fail.
val taskSet = FakeTask.createTaskSet(1, Seq(TaskLocation("host1", "exec1")))
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, 4, clock)
{
val offerResult = manager.resourceOffer("exec1", "host1", PROCESS_LOCAL)
assert(offerResult.isDefined, "Expect resource offer to return a task")
assert(offerResult.get.index === 0)
assert(offerResult.get.executorId === "exec1")
// Cause exec1 to fail : failure 1
manager.handleFailedTask(offerResult.get.taskId, TaskState.FINISHED, TaskResultLost)
assert(!sched.taskSetsFailed.contains(taskSet.id))
// Ensure scheduling on exec1 fails after failure 1 due to blacklist
assert(manager.resourceOffer("exec1", "host1", PROCESS_LOCAL).isEmpty)
assert(manager.resourceOffer("exec1", "host1", NODE_LOCAL).isEmpty)
assert(manager.resourceOffer("exec1", "host1", RACK_LOCAL).isEmpty)
assert(manager.resourceOffer("exec1", "host1", ANY).isEmpty)
}
// Run the task on exec1.1 - should work, and then fail it on exec1.1
{
val offerResult = manager.resourceOffer("exec1.1", "host1", NODE_LOCAL)
assert(offerResult.isDefined,
"Expect resource offer to return a task for exec1.1, offerResult = " + offerResult)
assert(offerResult.get.index === 0)
assert(offerResult.get.executorId === "exec1.1")
// Cause exec1.1 to fail : failure 2
manager.handleFailedTask(offerResult.get.taskId, TaskState.FINISHED, TaskResultLost)
assert(!sched.taskSetsFailed.contains(taskSet.id))
// Ensure scheduling on exec1.1 fails after failure 2 due to blacklist
assert(manager.resourceOffer("exec1.1", "host1", NODE_LOCAL).isEmpty)
}
// Run the task on exec2 - should work, and then fail it on exec2
{
val offerResult = manager.resourceOffer("exec2", "host2", ANY)
assert(offerResult.isDefined, "Expect resource offer to return a task")
assert(offerResult.get.index === 0)
assert(offerResult.get.executorId === "exec2")
// Cause exec2 to fail : failure 3
manager.handleFailedTask(offerResult.get.taskId, TaskState.FINISHED, TaskResultLost)
assert(!sched.taskSetsFailed.contains(taskSet.id))
// Ensure scheduling on exec2 fails after failure 3 due to blacklist
assert(manager.resourceOffer("exec2", "host2", ANY).isEmpty)
}
// After reschedule delay, scheduling on exec1 should be possible.
clock.advance(rescheduleDelay)
{
val offerResult = manager.resourceOffer("exec1", "host1", PROCESS_LOCAL)
assert(offerResult.isDefined, "Expect resource offer to return a task")
assert(offerResult.get.index === 0)
assert(offerResult.get.executorId === "exec1")
assert(manager.resourceOffer("exec1", "host1", PROCESS_LOCAL).isEmpty)
// Cause exec1 to fail : failure 4
manager.handleFailedTask(offerResult.get.taskId, TaskState.FINISHED, TaskResultLost)
}
// we have failed the same task 4 times now : task id should now be in taskSetsFailed
assert(sched.taskSetsFailed.contains(taskSet.id))
}
test("new executors get added and lost") {
// Assign host2 to rack2
FakeRackUtil.assignHostToRack("host2", "rack2")
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc)
val taskSet = FakeTask.createTaskSet(4,
Seq(TaskLocation("host1", "execA")),
Seq(TaskLocation("host1", "execB")),
Seq(TaskLocation("host2", "execC")),
Seq())
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// Only ANY is valid
assert(manager.myLocalityLevels.sameElements(Array(NO_PREF, ANY)))
// Add a new executor
sched.addExecutor("execD", "host1")
manager.executorAdded()
// Valid locality should contain NODE_LOCAL and ANY
assert(manager.myLocalityLevels.sameElements(Array(NODE_LOCAL, NO_PREF, ANY)))
// Add another executor
sched.addExecutor("execC", "host2")
manager.executorAdded()
// Valid locality should contain PROCESS_LOCAL, NODE_LOCAL, RACK_LOCAL and ANY
assert(manager.myLocalityLevels.sameElements(
Array(PROCESS_LOCAL, NODE_LOCAL, NO_PREF, RACK_LOCAL, ANY)))
// test if the valid locality is recomputed when the executor is lost
sched.removeExecutor("execC")
manager.executorLost("execC", "host2", SlaveLost())
assert(manager.myLocalityLevels.sameElements(Array(NODE_LOCAL, NO_PREF, ANY)))
sched.removeExecutor("execD")
manager.executorLost("execD", "host1", SlaveLost())
assert(manager.myLocalityLevels.sameElements(Array(NO_PREF, ANY)))
}
test("Executors exit for reason unrelated to currently running tasks") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc)
val taskSet = FakeTask.createTaskSet(4,
Seq(TaskLocation("host1", "execA")),
Seq(TaskLocation("host1", "execB")),
Seq(TaskLocation("host2", "execC")),
Seq())
val manager = new TaskSetManager(sched, taskSet, 1, new ManualClock)
sched.addExecutor("execA", "host1")
manager.executorAdded()
sched.addExecutor("execC", "host2")
manager.executorAdded()
assert(manager.resourceOffer("exec1", "host1", ANY).isDefined)
sched.removeExecutor("execA")
manager.executorLost(
"execA",
"host1",
ExecutorExited(143, false, "Terminated for reason unrelated to running tasks"))
assert(!sched.taskSetsFailed.contains(taskSet.id))
assert(manager.resourceOffer("execC", "host2", ANY).isDefined)
sched.removeExecutor("execC")
manager.executorLost(
"execC", "host2", ExecutorExited(1, true, "Terminated due to issue with running tasks"))
assert(sched.taskSetsFailed.contains(taskSet.id))
}
test("test RACK_LOCAL tasks") {
// Assign host1 to rack1
FakeRackUtil.assignHostToRack("host1", "rack1")
// Assign host2 to rack1
FakeRackUtil.assignHostToRack("host2", "rack1")
// Assign host3 to rack2
FakeRackUtil.assignHostToRack("host3", "rack2")
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc,
("execA", "host1"), ("execB", "host2"), ("execC", "host3"))
val taskSet = FakeTask.createTaskSet(2,
Seq(TaskLocation("host1", "execA")),
Seq(TaskLocation("host1", "execA")))
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
assert(manager.myLocalityLevels.sameElements(Array(PROCESS_LOCAL, NODE_LOCAL, RACK_LOCAL, ANY)))
// Set allowed locality to ANY
clock.advance(LOCALITY_WAIT_MS * 3)
// Offer host3
// No task is scheduled if we restrict locality to RACK_LOCAL
assert(manager.resourceOffer("execC", "host3", RACK_LOCAL) === None)
// Task 0 can be scheduled with ANY
assert(manager.resourceOffer("execC", "host3", ANY).get.index === 0)
// Offer host2
// Task 1 can be scheduled with RACK_LOCAL
assert(manager.resourceOffer("execB", "host2", RACK_LOCAL).get.index === 1)
}
test("do not emit warning when serialized task is small") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"))
val taskSet = FakeTask.createTaskSet(1)
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES)
assert(!manager.emittedTaskSizeWarning)
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 0)
assert(!manager.emittedTaskSizeWarning)
}
test("emit warning when serialized task is large") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"))
val taskSet = new TaskSet(Array(new LargeTask(0)), 0, 0, 0, null)
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES)
assert(!manager.emittedTaskSizeWarning)
assert(manager.resourceOffer("exec1", "host1", ANY).get.index === 0)
assert(manager.emittedTaskSizeWarning)
}
test("Not serializable exception thrown if the task cannot be serialized") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("exec1", "host1"))
val taskSet = new TaskSet(
Array(new NotSerializableFakeTask(1, 0), new NotSerializableFakeTask(0, 1)), 0, 0, 0, null)
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES)
intercept[TaskNotSerializableException] {
manager.resourceOffer("exec1", "host1", ANY)
}
assert(manager.isZombie)
}
test("abort the job if total size of results is too large") {
val conf = new SparkConf().set("spark.driver.maxResultSize", "2m")
sc = new SparkContext("local", "test", conf)
def genBytes(size: Int): (Int) => Array[Byte] = { (x: Int) =>
val bytes = Array.ofDim[Byte](size)
scala.util.Random.nextBytes(bytes)
bytes
}
// multiple 1k result
val r = sc.makeRDD(0 until 10, 10).map(genBytes(1024)).collect()
assert(10 === r.size)
// single 10M result
val thrown = intercept[SparkException] {sc.makeRDD(genBytes(10 << 20)(0), 1).collect()}
assert(thrown.getMessage().contains("bigger than spark.driver.maxResultSize"))
// multiple 1M results
val thrown2 = intercept[SparkException] {
sc.makeRDD(0 until 10, 10).map(genBytes(1 << 20)).collect()
}
assert(thrown2.getMessage().contains("bigger than spark.driver.maxResultSize"))
}
test("speculative and noPref task should be scheduled after node-local") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(
sc, ("execA", "host1"), ("execB", "host2"), ("execC", "host3"))
val taskSet = FakeTask.createTaskSet(4,
Seq(TaskLocation("host1", "execA")),
Seq(TaskLocation("host2"), TaskLocation("host1")),
Seq(),
Seq(TaskLocation("host3", "execC")))
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
assert(manager.resourceOffer("execA", "host1", PROCESS_LOCAL).get.index === 0)
assert(manager.resourceOffer("execA", "host1", NODE_LOCAL) == None)
assert(manager.resourceOffer("execA", "host1", NO_PREF).get.index == 1)
manager.speculatableTasks += 1
clock.advance(LOCALITY_WAIT_MS)
// schedule the nonPref task
assert(manager.resourceOffer("execA", "host1", NO_PREF).get.index === 2)
// schedule the speculative task
assert(manager.resourceOffer("execB", "host2", NO_PREF).get.index === 1)
clock.advance(LOCALITY_WAIT_MS * 3)
// schedule non-local tasks
assert(manager.resourceOffer("execB", "host2", ANY).get.index === 3)
}
test("node-local tasks should be scheduled right away " +
"when there are only node-local and no-preference tasks") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(
sc, ("execA", "host1"), ("execB", "host2"), ("execC", "host3"))
val taskSet = FakeTask.createTaskSet(4,
Seq(TaskLocation("host1")),
Seq(TaskLocation("host2")),
Seq(),
Seq(TaskLocation("host3")))
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// node-local tasks are scheduled without delay
assert(manager.resourceOffer("execA", "host1", NODE_LOCAL).get.index === 0)
assert(manager.resourceOffer("execA", "host2", NODE_LOCAL).get.index === 1)
assert(manager.resourceOffer("execA", "host3", NODE_LOCAL).get.index === 3)
assert(manager.resourceOffer("execA", "host3", NODE_LOCAL) === None)
// schedule no-preference after node local ones
assert(manager.resourceOffer("execA", "host3", NO_PREF).get.index === 2)
}
test("SPARK-4939: node-local tasks should be scheduled right after process-local tasks finished")
{
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("execA", "host1"), ("execB", "host2"))
val taskSet = FakeTask.createTaskSet(4,
Seq(TaskLocation("host1")),
Seq(TaskLocation("host2")),
Seq(ExecutorCacheTaskLocation("host1", "execA")),
Seq(ExecutorCacheTaskLocation("host2", "execB")))
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// process-local tasks are scheduled first
assert(manager.resourceOffer("execA", "host1", NODE_LOCAL).get.index === 2)
assert(manager.resourceOffer("execB", "host2", NODE_LOCAL).get.index === 3)
// node-local tasks are scheduled without delay
assert(manager.resourceOffer("execA", "host1", NODE_LOCAL).get.index === 0)
assert(manager.resourceOffer("execB", "host2", NODE_LOCAL).get.index === 1)
assert(manager.resourceOffer("execA", "host1", NODE_LOCAL) == None)
assert(manager.resourceOffer("execB", "host2", NODE_LOCAL) == None)
}
test("SPARK-4939: no-pref tasks should be scheduled after process-local tasks finished") {
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc, ("execA", "host1"), ("execB", "host2"))
val taskSet = FakeTask.createTaskSet(3,
Seq(),
Seq(ExecutorCacheTaskLocation("host1", "execA")),
Seq(ExecutorCacheTaskLocation("host2", "execB")))
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// process-local tasks are scheduled first
assert(manager.resourceOffer("execA", "host1", PROCESS_LOCAL).get.index === 1)
assert(manager.resourceOffer("execB", "host2", PROCESS_LOCAL).get.index === 2)
// no-pref tasks are scheduled without delay
assert(manager.resourceOffer("execA", "host1", PROCESS_LOCAL) == None)
assert(manager.resourceOffer("execA", "host1", NODE_LOCAL) == None)
assert(manager.resourceOffer("execA", "host1", NO_PREF).get.index === 0)
assert(manager.resourceOffer("execA", "host1", ANY) == None)
}
test("Ensure TaskSetManager is usable after addition of levels") {
// Regression test for SPARK-2931
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc)
val taskSet = FakeTask.createTaskSet(2,
Seq(TaskLocation("host1", "execA")),
Seq(TaskLocation("host2", "execB.1")))
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
// Only ANY is valid
assert(manager.myLocalityLevels.sameElements(Array(ANY)))
// Add a new executor
sched.addExecutor("execA", "host1")
sched.addExecutor("execB.2", "host2")
manager.executorAdded()
assert(manager.pendingTasksWithNoPrefs.size === 0)
// Valid locality should contain PROCESS_LOCAL, NODE_LOCAL and ANY
assert(manager.myLocalityLevels.sameElements(Array(PROCESS_LOCAL, NODE_LOCAL, ANY)))
assert(manager.resourceOffer("execA", "host1", ANY) !== None)
clock.advance(LOCALITY_WAIT_MS * 4)
assert(manager.resourceOffer("execB.2", "host2", ANY) !== None)
sched.removeExecutor("execA")
sched.removeExecutor("execB.2")
manager.executorLost("execA", "host1", SlaveLost())
manager.executorLost("execB.2", "host2", SlaveLost())
clock.advance(LOCALITY_WAIT_MS * 4)
sched.addExecutor("execC", "host3")
manager.executorAdded()
// Prior to the fix, this line resulted in an ArrayIndexOutOfBoundsException:
assert(manager.resourceOffer("execC", "host3", ANY) !== None)
}
test("Test that locations with HDFSCacheTaskLocation are treated as PROCESS_LOCAL.") {
// Regression test for SPARK-2931
sc = new SparkContext("local", "test")
val sched = new FakeTaskScheduler(sc,
("execA", "host1"), ("execB", "host2"), ("execC", "host3"))
val taskSet = FakeTask.createTaskSet(3,
Seq(TaskLocation("host1")),
Seq(TaskLocation("host2")),
Seq(TaskLocation("hdfs_cache_host3")))
val clock = new ManualClock
val manager = new TaskSetManager(sched, taskSet, MAX_TASK_FAILURES, clock)
assert(manager.myLocalityLevels.sameElements(Array(PROCESS_LOCAL, NODE_LOCAL, ANY)))
sched.removeExecutor("execA")
manager.executorAdded()
assert(manager.myLocalityLevels.sameElements(Array(PROCESS_LOCAL, NODE_LOCAL, ANY)))
sched.removeExecutor("execB")
manager.executorAdded()
assert(manager.myLocalityLevels.sameElements(Array(PROCESS_LOCAL, NODE_LOCAL, ANY)))
sched.removeExecutor("execC")
manager.executorAdded()
assert(manager.myLocalityLevels.sameElements(Array(ANY)))
}
test("Test TaskLocation for different host type.") {
assert(TaskLocation("host1") === HostTaskLocation("host1"))
assert(TaskLocation("hdfs_cache_host1") === HDFSCacheTaskLocation("host1"))
assert(TaskLocation("executor_host1_3") === ExecutorCacheTaskLocation("host1", "3"))
}
private def createTaskResult(
id: Int,
accumUpdates: Seq[AccumulableInfo] = Seq.empty[AccumulableInfo]): DirectTaskResult[Int] = {
val valueSer = SparkEnv.get.serializer.newInstance()
new DirectTaskResult[Int](valueSer.serialize(id), accumUpdates)
}
}
|