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
|
/*
* 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 scala.collection.mutable
import org.apache.spark.annotation.DeveloperApi
import org.apache.spark.executor.TaskMetrics
import org.apache.spark.internal.Logging
import org.apache.spark.util.{Distribution, Utils}
/**
* :: DeveloperApi ::
* Simple SparkListener that logs a few summary statistics when each stage completes.
*/
@DeveloperApi
class StatsReportListener extends SparkListener with Logging {
import org.apache.spark.scheduler.StatsReportListener._
private val taskInfoMetrics = mutable.Buffer[(TaskInfo, TaskMetrics)]()
override def onTaskEnd(taskEnd: SparkListenerTaskEnd) {
val info = taskEnd.taskInfo
val metrics = taskEnd.taskMetrics
if (info != null && metrics != null) {
taskInfoMetrics += ((info, metrics))
}
}
override def onStageCompleted(stageCompleted: SparkListenerStageCompleted) {
implicit val sc = stageCompleted
this.logInfo(s"Finished stage: ${getStatusDetail(stageCompleted.stageInfo)}")
showMillisDistribution("task runtime:", (info, _) => Some(info.duration), taskInfoMetrics)
// Shuffle write
showBytesDistribution("shuffle bytes written:",
(_, metric) => metric.shuffleWriteMetrics.map(_.bytesWritten), taskInfoMetrics)
// Fetch & I/O
showMillisDistribution("fetch wait time:",
(_, metric) => metric.shuffleReadMetrics.map(_.fetchWaitTime), taskInfoMetrics)
showBytesDistribution("remote bytes read:",
(_, metric) => metric.shuffleReadMetrics.map(_.remoteBytesRead), taskInfoMetrics)
showBytesDistribution("task result size:",
(_, metric) => Some(metric.resultSize), taskInfoMetrics)
// Runtime breakdown
val runtimePcts = taskInfoMetrics.map { case (info, metrics) =>
RuntimePercentage(info.duration, metrics)
}
showDistribution("executor (non-fetch) time pct: ",
Distribution(runtimePcts.map(_.executorPct * 100)), "%2.0f %%")
showDistribution("fetch wait time pct: ",
Distribution(runtimePcts.flatMap(_.fetchPct.map(_ * 100))), "%2.0f %%")
showDistribution("other time pct: ", Distribution(runtimePcts.map(_.other * 100)), "%2.0f %%")
taskInfoMetrics.clear()
}
private def getStatusDetail(info: StageInfo): String = {
val failureReason = info.failureReason.map("(" + _ + ")").getOrElse("")
val timeTaken = info.submissionTime.map(
x => info.completionTime.getOrElse(System.currentTimeMillis()) - x
).getOrElse("-")
s"Stage(${info.stageId}, ${info.attemptId}); Name: '${info.name}'; " +
s"Status: ${info.getStatusString}$failureReason; numTasks: ${info.numTasks}; " +
s"Took: $timeTaken msec"
}
}
private[spark] object StatsReportListener extends Logging {
// For profiling, the extremes are more interesting
val percentiles = Array[Int](0, 5, 10, 25, 50, 75, 90, 95, 100)
val probabilities = percentiles.map(_ / 100.0)
val percentilesHeader = "\t" + percentiles.mkString("%\t") + "%"
def extractDoubleDistribution(
taskInfoMetrics: Seq[(TaskInfo, TaskMetrics)],
getMetric: (TaskInfo, TaskMetrics) => Option[Double]): Option[Distribution] = {
Distribution(taskInfoMetrics.flatMap { case (info, metric) => getMetric(info, metric) })
}
// Is there some way to setup the types that I can get rid of this completely?
def extractLongDistribution(
taskInfoMetrics: Seq[(TaskInfo, TaskMetrics)],
getMetric: (TaskInfo, TaskMetrics) => Option[Long]): Option[Distribution] = {
extractDoubleDistribution(
taskInfoMetrics,
(info, metric) => { getMetric(info, metric).map(_.toDouble) })
}
def showDistribution(heading: String, d: Distribution, formatNumber: Double => String) {
val stats = d.statCounter
val quantiles = d.getQuantiles(probabilities).map(formatNumber)
logInfo(heading + stats)
logInfo(percentilesHeader)
logInfo("\t" + quantiles.mkString("\t"))
}
def showDistribution(
heading: String,
dOpt: Option[Distribution],
formatNumber: Double => String) {
dOpt.foreach { d => showDistribution(heading, d, formatNumber)}
}
def showDistribution(heading: String, dOpt: Option[Distribution], format: String) {
def f(d: Double): String = format.format(d)
showDistribution(heading, dOpt, f _)
}
def showDistribution(
heading: String,
format: String,
getMetric: (TaskInfo, TaskMetrics) => Option[Double],
taskInfoMetrics: Seq[(TaskInfo, TaskMetrics)]) {
showDistribution(heading, extractDoubleDistribution(taskInfoMetrics, getMetric), format)
}
def showBytesDistribution(
heading: String,
getMetric: (TaskInfo, TaskMetrics) => Option[Long],
taskInfoMetrics: Seq[(TaskInfo, TaskMetrics)]) {
showBytesDistribution(heading, extractLongDistribution(taskInfoMetrics, getMetric))
}
def showBytesDistribution(heading: String, dOpt: Option[Distribution]) {
dOpt.foreach { dist => showBytesDistribution(heading, dist) }
}
def showBytesDistribution(heading: String, dist: Distribution) {
showDistribution(heading, dist, (d => Utils.bytesToString(d.toLong)): Double => String)
}
def showMillisDistribution(heading: String, dOpt: Option[Distribution]) {
showDistribution(heading, dOpt,
(d => StatsReportListener.millisToString(d.toLong)): Double => String)
}
def showMillisDistribution(
heading: String,
getMetric: (TaskInfo, TaskMetrics) => Option[Long],
taskInfoMetrics: Seq[(TaskInfo, TaskMetrics)]) {
showMillisDistribution(heading, extractLongDistribution(taskInfoMetrics, getMetric))
}
val seconds = 1000L
val minutes = seconds * 60
val hours = minutes * 60
/**
* Reformat a time interval in milliseconds to a prettier format for output
*/
def millisToString(ms: Long): String = {
val (size, units) =
if (ms > hours) {
(ms.toDouble / hours, "hours")
} else if (ms > minutes) {
(ms.toDouble / minutes, "min")
} else if (ms > seconds) {
(ms.toDouble / seconds, "s")
} else {
(ms.toDouble, "ms")
}
"%.1f %s".format(size, units)
}
}
private case class RuntimePercentage(executorPct: Double, fetchPct: Option[Double], other: Double)
private object RuntimePercentage {
def apply(totalTime: Long, metrics: TaskMetrics): RuntimePercentage = {
val denom = totalTime.toDouble
val fetchTime = metrics.shuffleReadMetrics.map(_.fetchWaitTime)
val fetch = fetchTime.map(_ / denom)
val exec = (metrics.executorRunTime - fetchTime.getOrElse(0L)) / denom
val other = 1.0 - (exec + fetch.getOrElse(0d))
RuntimePercentage(exec, fetch, other)
}
}
|
