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
|
/*
* 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.Properties
import org.mockito.Matchers.any
import org.mockito.Mockito._
import org.scalatest.BeforeAndAfter
import org.apache.spark._
import org.apache.spark.executor.{Executor, TaskMetrics, TaskMetricsSuite}
import org.apache.spark.memory.TaskMemoryManager
import org.apache.spark.metrics.source.JvmSource
import org.apache.spark.network.util.JavaUtils
import org.apache.spark.rdd.RDD
import org.apache.spark.util._
class TaskContextSuite extends SparkFunSuite with BeforeAndAfter with LocalSparkContext {
test("provide metrics sources") {
val filePath = getClass.getClassLoader.getResource("test_metrics_config.properties").getFile
val conf = new SparkConf(loadDefaults = false)
.set("spark.metrics.conf", filePath)
sc = new SparkContext("local", "test", conf)
val rdd = sc.makeRDD(1 to 1)
val result = sc.runJob(rdd, (tc: TaskContext, it: Iterator[Int]) => {
tc.getMetricsSources("jvm").count {
case source: JvmSource => true
case _ => false
}
}).sum
assert(result > 0)
}
test("calls TaskCompletionListener after failure") {
TaskContextSuite.completed = false
sc = new SparkContext("local", "test")
val rdd = new RDD[String](sc, List()) {
override def getPartitions = Array[Partition](StubPartition(0))
override def compute(split: Partition, context: TaskContext) = {
context.addTaskCompletionListener(context => TaskContextSuite.completed = true)
sys.error("failed")
}
}
val closureSerializer = SparkEnv.get.closureSerializer.newInstance()
val func = (c: TaskContext, i: Iterator[String]) => i.next()
val taskBinary = sc.broadcast(JavaUtils.bufferToArray(closureSerializer.serialize((rdd, func))))
val task = new ResultTask[String, String](
0, 0, taskBinary, rdd.partitions(0), Seq.empty, 0, new Properties, new TaskMetrics)
intercept[RuntimeException] {
task.run(0, 0, null)
}
assert(TaskContextSuite.completed === true)
}
test("calls TaskFailureListeners after failure") {
TaskContextSuite.lastError = null
sc = new SparkContext("local", "test")
val rdd = new RDD[String](sc, List()) {
override def getPartitions = Array[Partition](StubPartition(0))
override def compute(split: Partition, context: TaskContext) = {
context.addTaskFailureListener((context, error) => TaskContextSuite.lastError = error)
sys.error("damn error")
}
}
val closureSerializer = SparkEnv.get.closureSerializer.newInstance()
val func = (c: TaskContext, i: Iterator[String]) => i.next()
val taskBinary = sc.broadcast(JavaUtils.bufferToArray(closureSerializer.serialize((rdd, func))))
val task = new ResultTask[String, String](
0, 0, taskBinary, rdd.partitions(0), Seq.empty, 0, new Properties, new TaskMetrics)
intercept[RuntimeException] {
task.run(0, 0, null)
}
assert(TaskContextSuite.lastError.getMessage == "damn error")
}
test("all TaskCompletionListeners should be called even if some fail") {
val context = TaskContext.empty()
val listener = mock(classOf[TaskCompletionListener])
context.addTaskCompletionListener(_ => throw new Exception("blah"))
context.addTaskCompletionListener(listener)
context.addTaskCompletionListener(_ => throw new Exception("blah"))
intercept[TaskCompletionListenerException] {
context.markTaskCompleted()
}
verify(listener, times(1)).onTaskCompletion(any())
}
test("all TaskFailureListeners should be called even if some fail") {
val context = TaskContext.empty()
val listener = mock(classOf[TaskFailureListener])
context.addTaskFailureListener((_, _) => throw new Exception("exception in listener1"))
context.addTaskFailureListener(listener)
context.addTaskFailureListener((_, _) => throw new Exception("exception in listener3"))
val e = intercept[TaskCompletionListenerException] {
context.markTaskFailed(new Exception("exception in task"))
}
// Make sure listener 2 was called.
verify(listener, times(1)).onTaskFailure(any(), any())
// also need to check failure in TaskFailureListener does not mask earlier exception
assert(e.getMessage.contains("exception in listener1"))
assert(e.getMessage.contains("exception in listener3"))
assert(e.getMessage.contains("exception in task"))
}
test("TaskContext.attemptNumber should return attempt number, not task id (SPARK-4014)") {
sc = new SparkContext("local[1,2]", "test") // use maxRetries = 2 because we test failed tasks
// Check that attemptIds are 0 for all tasks' initial attempts
val attemptIds = sc.parallelize(Seq(1, 2), 2).mapPartitions { iter =>
Seq(TaskContext.get().attemptNumber).iterator
}.collect()
assert(attemptIds.toSet === Set(0))
// Test a job with failed tasks
val attemptIdsWithFailedTask = sc.parallelize(Seq(1, 2), 2).mapPartitions { iter =>
val attemptId = TaskContext.get().attemptNumber
if (iter.next() == 1 && attemptId == 0) {
throw new Exception("First execution of task failed")
}
Seq(attemptId).iterator
}.collect()
assert(attemptIdsWithFailedTask.toSet === Set(0, 1))
}
test("accumulators are updated on exception failures") {
// This means use 1 core and 4 max task failures
sc = new SparkContext("local[1,4]", "test")
val param = AccumulatorParam.LongAccumulatorParam
// Create 2 accumulators, one that counts failed values and another that doesn't
val acc1 = new Accumulator(0L, param, Some("x"), countFailedValues = true)
val acc2 = new Accumulator(0L, param, Some("y"), countFailedValues = false)
// Fail first 3 attempts of every task. This means each task should be run 4 times.
sc.parallelize(1 to 10, 10).map { i =>
acc1 += 1
acc2 += 1
if (TaskContext.get.attemptNumber() <= 2) {
throw new Exception("you did something wrong")
} else {
0
}
}.count()
// The one that counts failed values should be 4x the one that didn't,
// since we ran each task 4 times
assert(AccumulatorContext.get(acc1.id).get.value === 40L)
assert(AccumulatorContext.get(acc2.id).get.value === 10L)
}
test("failed tasks collect only accumulators whose values count during failures") {
sc = new SparkContext("local", "test")
val acc1 = AccumulatorSuite.createLongAccum("x", false)
val acc2 = AccumulatorSuite.createLongAccum("y", true)
acc1.add(1)
acc2.add(1)
// Create a dummy task. We won't end up running this; we just want to collect
// accumulator updates from it.
val taskMetrics = TaskMetrics.empty
val task = new Task[Int](0, 0, 0) {
context = new TaskContextImpl(0, 0, 0L, 0,
new TaskMemoryManager(SparkEnv.get.memoryManager, 0L),
new Properties,
SparkEnv.get.metricsSystem,
taskMetrics)
taskMetrics.registerAccumulator(acc1)
taskMetrics.registerAccumulator(acc2)
override def runTask(tc: TaskContext): Int = 0
}
// First, simulate task success. This should give us all the accumulators.
val accumUpdates1 = task.collectAccumulatorUpdates(taskFailed = false)
TaskMetricsSuite.assertUpdatesEquals(accumUpdates1.takeRight(2), Seq(acc1, acc2))
// Now, simulate task failures. This should give us only the accums that count failed values.
val accumUpdates2 = task.collectAccumulatorUpdates(taskFailed = true)
TaskMetricsSuite.assertUpdatesEquals(accumUpdates2.takeRight(1), Seq(acc2))
}
test("only updated internal accumulators will be sent back to driver") {
sc = new SparkContext("local", "test")
// Create a dummy task. We won't end up running this; we just want to collect
// accumulator updates from it.
val taskMetrics = TaskMetrics.empty
val task = new Task[Int](0, 0, 0) {
context = new TaskContextImpl(0, 0, 0L, 0,
new TaskMemoryManager(SparkEnv.get.memoryManager, 0L),
new Properties,
SparkEnv.get.metricsSystem,
taskMetrics)
taskMetrics.incMemoryBytesSpilled(10)
override def runTask(tc: TaskContext): Int = 0
}
val updatedAccums = task.collectAccumulatorUpdates()
assert(updatedAccums.length == 2)
// the RESULT_SIZE accumulator will be sent back anyway.
assert(updatedAccums(0).name == Some(InternalAccumulator.RESULT_SIZE))
assert(updatedAccums(0).value == 0)
assert(updatedAccums(1).name == Some(InternalAccumulator.MEMORY_BYTES_SPILLED))
assert(updatedAccums(1).value == 10)
}
test("localProperties are propagated to executors correctly") {
sc = new SparkContext("local", "test")
sc.setLocalProperty("testPropKey", "testPropValue")
val res = sc.parallelize(Array(1), 1).map(i => i).map(i => {
val inTask = TaskContext.get().getLocalProperty("testPropKey")
val inDeser = Executor.taskDeserializationProps.get().getProperty("testPropKey")
s"$inTask,$inDeser"
}).collect()
assert(res === Array("testPropValue,testPropValue"))
}
}
private object TaskContextSuite {
@volatile var completed = false
@volatile var lastError: Throwable = _
}
private case class StubPartition(index: Int) extends Partition
|