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
|
/*
* 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 spark.deploy.master
import akka.actor._
import akka.actor.Terminated
import akka.remote.{RemoteClientLifeCycleEvent, RemoteClientDisconnected, RemoteClientShutdown}
import akka.util.duration._
import java.text.SimpleDateFormat
import java.util.Date
import scala.collection.mutable.{ArrayBuffer, HashMap, HashSet}
import spark.deploy._
import spark.{Logging, SparkException, Utils}
import spark.metrics.MetricsSystem
import spark.util.AkkaUtils
import ui.MasterWebUI
private[spark] class Master(host: String, port: Int, webUiPort: Int) extends Actor with Logging {
val DATE_FORMAT = new SimpleDateFormat("yyyyMMddHHmmss") // For application IDs
val WORKER_TIMEOUT = System.getProperty("spark.worker.timeout", "60").toLong * 1000
var nextAppNumber = 0
val workers = new HashSet[WorkerInfo]
val idToWorker = new HashMap[String, WorkerInfo]
val actorToWorker = new HashMap[ActorRef, WorkerInfo]
val addressToWorker = new HashMap[Address, WorkerInfo]
val apps = new HashSet[ApplicationInfo]
val idToApp = new HashMap[String, ApplicationInfo]
val actorToApp = new HashMap[ActorRef, ApplicationInfo]
val addressToApp = new HashMap[Address, ApplicationInfo]
val waitingApps = new ArrayBuffer[ApplicationInfo]
val completedApps = new ArrayBuffer[ApplicationInfo]
var firstApp: Option[ApplicationInfo] = None
val webUi = new MasterWebUI(self, webUiPort)
Utils.checkHost(host, "Expected hostname")
val masterSource = new MasterSource(this)
val masterPublicAddress = {
val envVar = System.getenv("SPARK_PUBLIC_DNS")
if (envVar != null) envVar else host
}
// As a temporary workaround before better ways of configuring memory, we allow users to set
// a flag that will perform round-robin scheduling across the nodes (spreading out each app
// among all the nodes) instead of trying to consolidate each app onto a small # of nodes.
val spreadOutApps = System.getProperty("spark.deploy.spreadOut", "true").toBoolean
override def preStart() {
logInfo("Starting Spark master at spark://" + host + ":" + port)
// Listen for remote client disconnection events, since they don't go through Akka's watch()
context.system.eventStream.subscribe(self, classOf[RemoteClientLifeCycleEvent])
webUi.start()
context.system.scheduler.schedule(0 millis, WORKER_TIMEOUT millis)(timeOutDeadWorkers())
Master.metricsSystem.registerSource(masterSource)
Master.metricsSystem.start()
}
override def postStop() {
webUi.stop()
}
override def receive = {
case RegisterWorker(id, host, workerPort, cores, memory, worker_webUiPort, publicAddress) => {
logInfo("Registering worker %s:%d with %d cores, %s RAM".format(
host, workerPort, cores, Utils.memoryMegabytesToString(memory)))
if (idToWorker.contains(id)) {
sender ! RegisterWorkerFailed("Duplicate worker ID")
} else {
addWorker(id, host, workerPort, cores, memory, worker_webUiPort, publicAddress)
context.watch(sender) // This doesn't work with remote actors but helps for testing
sender ! RegisteredWorker("http://" + masterPublicAddress + ":" + webUi.boundPort.get)
schedule()
}
}
case RegisterApplication(description) => {
logInfo("Registering app " + description.name)
val app = addApplication(description, sender)
logInfo("Registered app " + description.name + " with ID " + app.id)
waitingApps += app
context.watch(sender) // This doesn't work with remote actors but helps for testing
sender ! RegisteredApplication(app.id)
schedule()
}
case ExecutorStateChanged(appId, execId, state, message, exitStatus) => {
val execOption = idToApp.get(appId).flatMap(app => app.executors.get(execId))
execOption match {
case Some(exec) => {
exec.state = state
exec.application.driver ! ExecutorUpdated(execId, state, message, exitStatus)
if (ExecutorState.isFinished(state)) {
val appInfo = idToApp(appId)
// Remove this executor from the worker and app
logInfo("Removing executor " + exec.fullId + " because it is " + state)
appInfo.removeExecutor(exec)
exec.worker.removeExecutor(exec)
// Only retry certain number of times so we don't go into an infinite loop.
if (appInfo.incrementRetryCount < ApplicationState.MAX_NUM_RETRY) {
schedule()
} else {
logError("Application %s with ID %s failed %d times, removing it".format(
appInfo.desc.name, appInfo.id, appInfo.retryCount))
removeApplication(appInfo, ApplicationState.FAILED)
}
}
}
case None =>
logWarning("Got status update for unknown executor " + appId + "/" + execId)
}
}
case Heartbeat(workerId) => {
idToWorker.get(workerId) match {
case Some(workerInfo) =>
workerInfo.lastHeartbeat = System.currentTimeMillis()
case None =>
logWarning("Got heartbeat from unregistered worker " + workerId)
}
}
case Terminated(actor) => {
// The disconnected actor could've been either a worker or an app; remove whichever of
// those we have an entry for in the corresponding actor hashmap
actorToWorker.get(actor).foreach(removeWorker)
actorToApp.get(actor).foreach(finishApplication)
}
case RemoteClientDisconnected(transport, address) => {
// The disconnected client could've been either a worker or an app; remove whichever it was
addressToWorker.get(address).foreach(removeWorker)
addressToApp.get(address).foreach(finishApplication)
}
case RemoteClientShutdown(transport, address) => {
// The disconnected client could've been either a worker or an app; remove whichever it was
addressToWorker.get(address).foreach(removeWorker)
addressToApp.get(address).foreach(finishApplication)
}
case RequestMasterState => {
sender ! MasterState(host, port, workers.toArray, apps.toArray, completedApps.toArray)
}
}
/**
* Can an app use the given worker? True if the worker has enough memory and we haven't already
* launched an executor for the app on it (right now the standalone backend doesn't like having
* two executors on the same worker).
*/
def canUse(app: ApplicationInfo, worker: WorkerInfo): Boolean = {
worker.memoryFree >= app.desc.memoryPerSlave && !worker.hasExecutor(app)
}
/**
* Schedule the currently available resources among waiting apps. This method will be called
* every time a new app joins or resource availability changes.
*/
def schedule() {
// Right now this is a very simple FIFO scheduler. We keep trying to fit in the first app
// in the queue, then the second app, etc.
if (spreadOutApps) {
// Try to spread out each app among all the nodes, until it has all its cores
for (app <- waitingApps if app.coresLeft > 0) {
val usableWorkers = workers.toArray.filter(_.state == WorkerState.ALIVE)
.filter(canUse(app, _)).sortBy(_.coresFree).reverse
val numUsable = usableWorkers.length
val assigned = new Array[Int](numUsable) // Number of cores to give on each node
var toAssign = math.min(app.coresLeft, usableWorkers.map(_.coresFree).sum)
var pos = 0
while (toAssign > 0) {
if (usableWorkers(pos).coresFree - assigned(pos) > 0) {
toAssign -= 1
assigned(pos) += 1
}
pos = (pos + 1) % numUsable
}
// Now that we've decided how many cores to give on each node, let's actually give them
for (pos <- 0 until numUsable) {
if (assigned(pos) > 0) {
val exec = app.addExecutor(usableWorkers(pos), assigned(pos))
launchExecutor(usableWorkers(pos), exec, app.desc.sparkHome)
app.state = ApplicationState.RUNNING
}
}
}
} else {
// Pack each app into as few nodes as possible until we've assigned all its cores
for (worker <- workers if worker.coresFree > 0 && worker.state == WorkerState.ALIVE) {
for (app <- waitingApps if app.coresLeft > 0) {
if (canUse(app, worker)) {
val coresToUse = math.min(worker.coresFree, app.coresLeft)
if (coresToUse > 0) {
val exec = app.addExecutor(worker, coresToUse)
launchExecutor(worker, exec, app.desc.sparkHome)
app.state = ApplicationState.RUNNING
}
}
}
}
}
}
def launchExecutor(worker: WorkerInfo, exec: ExecutorInfo, sparkHome: String) {
logInfo("Launching executor " + exec.fullId + " on worker " + worker.id)
worker.addExecutor(exec)
worker.actor ! LaunchExecutor(exec.application.id, exec.id, exec.application.desc, exec.cores, exec.memory, sparkHome)
exec.application.driver ! ExecutorAdded(exec.id, worker.id, worker.hostPort, exec.cores, exec.memory)
}
def addWorker(id: String, host: String, port: Int, cores: Int, memory: Int, webUiPort: Int,
publicAddress: String): WorkerInfo = {
// There may be one or more refs to dead workers on this same node (w/ different ID's), remove them.
workers.filter(w => (w.host == host && w.port == port) && (w.state == WorkerState.DEAD)).foreach(workers -= _)
val worker = new WorkerInfo(id, host, port, cores, memory, sender, webUiPort, publicAddress)
workers += worker
idToWorker(worker.id) = worker
actorToWorker(sender) = worker
addressToWorker(sender.path.address) = worker
return worker
}
def removeWorker(worker: WorkerInfo) {
logInfo("Removing worker " + worker.id + " on " + worker.host + ":" + worker.port)
worker.setState(WorkerState.DEAD)
idToWorker -= worker.id
actorToWorker -= worker.actor
addressToWorker -= worker.actor.path.address
for (exec <- worker.executors.values) {
logInfo("Telling app of lost executor: " + exec.id)
exec.application.driver ! ExecutorUpdated(exec.id, ExecutorState.LOST, Some("worker lost"), None)
exec.application.removeExecutor(exec)
}
}
def addApplication(desc: ApplicationDescription, driver: ActorRef): ApplicationInfo = {
val now = System.currentTimeMillis()
val date = new Date(now)
val app = new ApplicationInfo(now, newApplicationId(date), desc, date, driver, desc.appUiUrl)
apps += app
idToApp(app.id) = app
actorToApp(driver) = app
addressToApp(driver.path.address) = app
if (firstApp == None) {
firstApp = Some(app)
}
val workersAlive = workers.filter(_.state == WorkerState.ALIVE).toArray
if (workersAlive.size > 0 && !workersAlive.exists(_.memoryFree >= desc.memoryPerSlave)) {
logWarning("Could not find any workers with enough memory for " + firstApp.get.id)
}
return app
}
def finishApplication(app: ApplicationInfo) {
removeApplication(app, ApplicationState.FINISHED)
}
def removeApplication(app: ApplicationInfo, state: ApplicationState.Value) {
if (apps.contains(app)) {
logInfo("Removing app " + app.id)
apps -= app
idToApp -= app.id
actorToApp -= app.driver
addressToApp -= app.driver.path.address
completedApps += app // Remember it in our history
waitingApps -= app
for (exec <- app.executors.values) {
exec.worker.removeExecutor(exec)
exec.worker.actor ! KillExecutor(exec.application.id, exec.id)
exec.state = ExecutorState.KILLED
}
app.markFinished(state)
if (state != ApplicationState.FINISHED) {
app.driver ! ApplicationRemoved(state.toString)
}
schedule()
}
}
/** Generate a new app ID given a app's submission date */
def newApplicationId(submitDate: Date): String = {
val appId = "app-%s-%04d".format(DATE_FORMAT.format(submitDate), nextAppNumber)
nextAppNumber += 1
appId
}
/** Check for, and remove, any timed-out workers */
def timeOutDeadWorkers() {
// Copy the workers into an array so we don't modify the hashset while iterating through it
val expirationTime = System.currentTimeMillis() - WORKER_TIMEOUT
val toRemove = workers.filter(_.lastHeartbeat < expirationTime).toArray
for (worker <- toRemove) {
logWarning("Removing %s because we got no heartbeat in %d seconds".format(
worker.id, WORKER_TIMEOUT))
removeWorker(worker)
}
}
override def postStop() {
Master.metricsSystem.stop()
}
}
private[spark] object Master {
private val systemName = "sparkMaster"
private val actorName = "Master"
private val sparkUrlRegex = "spark://([^:]+):([0-9]+)".r
private val metricsSystem = MetricsSystem.createMetricsSystem("master")
def main(argStrings: Array[String]) {
val args = new MasterArguments(argStrings)
val (actorSystem, _) = startSystemAndActor(args.host, args.port, args.webUiPort)
actorSystem.awaitTermination()
}
/** Returns an `akka://...` URL for the Master actor given a sparkUrl `spark://host:ip`. */
def toAkkaUrl(sparkUrl: String): String = {
sparkUrl match {
case sparkUrlRegex(host, port) =>
"akka://%s@%s:%s/user/%s".format(systemName, host, port, actorName)
case _ =>
throw new SparkException("Invalid master URL: " + sparkUrl)
}
}
def startSystemAndActor(host: String, port: Int, webUiPort: Int): (ActorSystem, Int) = {
val (actorSystem, boundPort) = AkkaUtils.createActorSystem(systemName, host, port)
val actor = actorSystem.actorOf(Props(new Master(host, boundPort, webUiPort)), name = actorName)
(actorSystem, boundPort)
}
}
|