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
|
package spark
import org.scalatest.FunSuite
import akka.actor._
import spark.scheduler.MapStatus
import spark.storage.BlockManagerId
import spark.util.AkkaUtils
class MapOutputTrackerSuite extends FunSuite with LocalSparkContext {
test("compressSize") {
assert(MapOutputTracker.compressSize(0L) === 0)
assert(MapOutputTracker.compressSize(1L) === 1)
assert(MapOutputTracker.compressSize(2L) === 8)
assert(MapOutputTracker.compressSize(10L) === 25)
assert((MapOutputTracker.compressSize(1000000L) & 0xFF) === 145)
assert((MapOutputTracker.compressSize(1000000000L) & 0xFF) === 218)
// This last size is bigger than we can encode in a byte, so check that we just return 255
assert((MapOutputTracker.compressSize(1000000000000000000L) & 0xFF) === 255)
}
test("decompressSize") {
assert(MapOutputTracker.decompressSize(0) === 0)
for (size <- Seq(2L, 10L, 100L, 50000L, 1000000L, 1000000000L)) {
val size2 = MapOutputTracker.decompressSize(MapOutputTracker.compressSize(size))
assert(size2 >= 0.99 * size && size2 <= 1.11 * size,
"size " + size + " decompressed to " + size2 + ", which is out of range")
}
}
test("master start and stop") {
val actorSystem = ActorSystem("test")
val tracker = new MapOutputTracker(actorSystem, true)
tracker.stop()
}
test("master register and fetch") {
val actorSystem = ActorSystem("test")
val tracker = new MapOutputTracker(actorSystem, true)
tracker.registerShuffle(10, 2)
val compressedSize1000 = MapOutputTracker.compressSize(1000L)
val compressedSize10000 = MapOutputTracker.compressSize(10000L)
val size1000 = MapOutputTracker.decompressSize(compressedSize1000)
val size10000 = MapOutputTracker.decompressSize(compressedSize10000)
tracker.registerMapOutput(10, 0, new MapStatus(BlockManagerId("a", "hostA", 1000),
Array(compressedSize1000, compressedSize10000)))
tracker.registerMapOutput(10, 1, new MapStatus(BlockManagerId("b", "hostB", 1000),
Array(compressedSize10000, compressedSize1000)))
val statuses = tracker.getServerStatuses(10, 0)
assert(statuses.toSeq === Seq((BlockManagerId("a", "hostA", 1000), size1000),
(BlockManagerId("b", "hostB", 1000), size10000)))
tracker.stop()
}
test("master register and unregister and fetch") {
val actorSystem = ActorSystem("test")
val tracker = new MapOutputTracker(actorSystem, true)
tracker.registerShuffle(10, 2)
val compressedSize1000 = MapOutputTracker.compressSize(1000L)
val compressedSize10000 = MapOutputTracker.compressSize(10000L)
val size1000 = MapOutputTracker.decompressSize(compressedSize1000)
val size10000 = MapOutputTracker.decompressSize(compressedSize10000)
tracker.registerMapOutput(10, 0, new MapStatus(BlockManagerId("a", "hostA", 1000),
Array(compressedSize1000, compressedSize1000, compressedSize1000)))
tracker.registerMapOutput(10, 1, new MapStatus(BlockManagerId("b", "hostB", 1000),
Array(compressedSize10000, compressedSize1000, compressedSize1000)))
// As if we had two simulatenous fetch failures
tracker.unregisterMapOutput(10, 0, BlockManagerId("a", "hostA", 1000))
tracker.unregisterMapOutput(10, 0, BlockManagerId("a", "hostA", 1000))
// The remaining reduce task might try to grab the output despite the shuffle failure;
// this should cause it to fail, and the scheduler will ignore the failure due to the
// stage already being aborted.
intercept[FetchFailedException] { tracker.getServerStatuses(10, 1) }
}
test("remote fetch") {
try {
System.clearProperty("spark.driver.host") // In case some previous test had set it
val (actorSystem, boundPort) =
AkkaUtils.createActorSystem("test", "localhost", 0)
System.setProperty("spark.driver.port", boundPort.toString)
val masterTracker = new MapOutputTracker(actorSystem, true)
val slaveTracker = new MapOutputTracker(actorSystem, false)
masterTracker.registerShuffle(10, 1)
masterTracker.incrementGeneration()
slaveTracker.updateGeneration(masterTracker.getGeneration)
intercept[FetchFailedException] { slaveTracker.getServerStatuses(10, 0) }
val compressedSize1000 = MapOutputTracker.compressSize(1000L)
val size1000 = MapOutputTracker.decompressSize(compressedSize1000)
masterTracker.registerMapOutput(10, 0, new MapStatus(
BlockManagerId("a", "hostA", 1000), Array(compressedSize1000)))
masterTracker.incrementGeneration()
slaveTracker.updateGeneration(masterTracker.getGeneration)
assert(slaveTracker.getServerStatuses(10, 0).toSeq ===
Seq((BlockManagerId("a", "hostA", 1000), size1000)))
masterTracker.unregisterMapOutput(10, 0, BlockManagerId("a", "hostA", 1000))
masterTracker.incrementGeneration()
slaveTracker.updateGeneration(masterTracker.getGeneration)
intercept[FetchFailedException] { slaveTracker.getServerStatuses(10, 0) }
// failure should be cached
intercept[FetchFailedException] { slaveTracker.getServerStatuses(10, 0) }
} finally {
System.clearProperty("spark.driver.port")
}
}
}
|