上篇文章我们讲解了master调度driver和executor(application)的过程,但是对于executor在worker上的创建过程没有讲,这里我们接着上篇文章继续讲。
首先我们从worker接收到master发送过来的LaunchExecutor事件开始,源码如下:
case LaunchExecutor(masterUrl, appId, execId, appDesc, cores_, memory_, resources_) =>
//验证当前masterurl是否有效
if (masterUrl != activeMasterUrl) {
logWarning("Invalid Master (" + masterUrl + ") attempted to launch executor.")
} else {
try {
logInfo("Asked to launch executor %s/%d for %s".format(appId, execId, appDesc.name))
// 创建executor工作目录
val executorDir = new File(workDir, appId + "/" + execId)
if (!executorDir.mkdirs()) {
throw new IOException("Failed to create directory " + executorDir)
}
// 创建application目录,在application执行结束后由worker删除
val appLocalDirs = appDirectories.getOrElse(appId, {
val localRootDirs = Utils.getOrCreateLocalRootDirs(conf)
val dirs = localRootDirs.flatMap { dir =>
try {
val appDir = Utils.createDirectory(dir, namePrefix = "executor")
Utils.chmod700(appDir)
Some(appDir.getAbsolutePath())
} catch {
case e: IOException =>
logWarning(s"${e.getMessage}. Ignoring this directory.")
None
}
}.toSeq
if (dirs.isEmpty) {
throw new IOException("No subfolder can be created in " +
s"${localRootDirs.mkString(",")}.")
}
dirs
})
//存储application和其目录的对应关系
appDirectories(appId) = appLocalDirs
//重点:创建ExecutorRunner对象并启动
val manager = new ExecutorRunner(
appId,
execId,
appDesc.copy(command = Worker.maybeUpdateSSLSettings(appDesc.command, conf)),
cores_,
memory_,
self,
workerId,
webUi.scheme,
host,
webUi.boundPort,
publicAddress,
sparkHome,
executorDir,
workerUri,
conf,
appLocalDirs,
ExecutorState.LAUNCHING,
resources_)
executors(appId + "/" + execId) = manager
manager.start()
//executor启动后,当前worker被使用的资源都加上executor所占用的
coresUsed += cores_
memoryUsed += memory_
addResourcesUsed(resources_)
} catch {
case e: Exception =>
logError(s"Failed to launch executor $appId/$execId for ${appDesc.name}.", e)
if (executors.contains(appId + "/" + execId)) {
executors(appId + "/" + execId).kill()
executors -= appId + "/" + execId
}
sendToMaster(ExecutorStateChanged(appId, execId, ExecutorState.FAILED,
Some(e.toString), None))
}
}
可以看到,executor的创建主要是创建一个ExecutorRunner对象并调用start启动,创建对象没有什么可说的,我们直接追踪其start方法,源码如下:
private[worker] def start(): Unit = {
workerThread = new Thread("ExecutorRunner for " + fullId) {
//重点:通过新建一个线程专门用于创建executor
override def run(): Unit = { fetchAndRunExecutor() }
}
workerThread.start()
// 钩子函数处理一些收尾操作
shutdownHook = ShutdownHookManager.addShutdownHook { () =>
//到这一步后,Executor应该早就结束了构建发起状态,如果还是发起态,说明Executor出现了异常,所以设置其状态为失败态
if (state == ExecutorState.LAUNCHING) {
state = ExecutorState.FAILED
}
killProcess(Some("Worker shutting down")) }
}
在ExecutorRunner的启动方法中,会创建一个java线程,通过该线程创建executor。接下来我们再来看看这个线程里面的操作内容,如下:
private def fetchAndRunExecutor(): Unit = {
try {
//准备环境
val resourceFileOpt = prepareResourcesFile(SPARK_EXECUTOR_PREFIX, resources, executorDir)
// 拼接命令参数
val arguments = appDesc.command.arguments ++ resourceFileOpt.map(f =>
Seq("--resourcesFile", f.getAbsolutePath)).getOrElse(Seq.empty)
val subsOpts = appDesc.command.javaOpts.map {
Utils.substituteAppNExecIds(_, appId, execId.toString)
}
val subsCommand = appDesc.command.copy(arguments = arguments, javaOpts = subsOpts)
val builder = CommandUtils.buildProcessBuilder(subsCommand, new SecurityManager(conf),
memory, sparkHome.getAbsolutePath, substituteVariables)
//构建命令
val command = builder.command()
val redactedCommand = Utils.redactCommandLineArgs(conf, command.asScala)
.mkString("\"", "\" \"", "\"")
logInfo(s"Launch command: $redactedCommand")
//环境准备
builder.directory(executorDir)
builder.environment.put("SPARK_EXECUTOR_DIRS", appLocalDirs.mkString(File.pathSeparator))
builder.environment.put("SPARK_LAUNCH_WITH_SCALA", "0")
// Add webUI log urls
val baseUrl =
if (conf.get(UI_REVERSE_PROXY)) {
s"/proxy/$workerId/logPage/?appId=$appId&executorId=$execId&logType="
} else {
s"$webUiScheme$publicAddress:$webUiPort/logPage/?appId=$appId&executorId=$execId&logType="
}
builder.environment.put("SPARK_LOG_URL_STDERR", s"${baseUrl}stderr")
builder.environment.put("SPARK_LOG_URL_STDOUT", s"${baseUrl}stdout")
//发起命令,此时会有一个新的运行态进程,这个进程就是spark具体执行任务的executor
process = builder.start()
//重点一:将启动命令输出到stderr文件流中(这里之所以标记重点,是因为这块代码不好debug查看,但是我们可以去日志中查看其具体执行的命令是什么,这样我们也可以推测出后续的处理流程)
val header = "Spark Executor Command: %s\n%s\n\n".format(
redactedCommand, "=" * 40)
// 重定向输出流
val stdout = new File(executorDir, "stdout")
stdoutAppender = FileAppender(process.getInputStream, stdout, conf)
val stderr = new File(executorDir, "stderr")
Files.write(header, stderr, StandardCharsets.UTF_8)
stderrAppender = FileAppender(process.getErrorStream, stderr, conf)
//修改Executor状态为运行态
state = ExecutorState.RUNNING
//通知当前worker此Executor的状态变化
worker.send(ExecutorStateChanged(appId, execId, state, None, None))
// 线程阻塞,等待executor的退出
val exitCode = process.waitFor()
// 阻塞结束,说明executor退出了
state = ExecutorState.EXITED
val message = "Command exited with code " + exitCode
//再次向worker通知其状态变化
worker.send(ExecutorStateChanged(appId, execId, state, Some(message), Some(exitCode)))
} catch {
case interrupted: InterruptedException =>
logInfo("Runner thread for executor " + fullId + " interrupted")
state = ExecutorState.KILLED
killProcess(None)
case e: Exception =>
logError("Error running executor", e)
state = ExecutorState.FAILED
killProcess(Some(e.toString))
}
}
可以看到executor是通过命令专门发起的一个进程,因为debug比较困难,所以我们没法实时看到发起的细节,但是好在其会打印出执行命令,我们通过命令也可以继续追踪后续的操作。所以在日志打印那行源码我也加了个重点标识。
这一步其实还有一个比较重要的操作,就是通知当前worker此Executor的状态变化,worker接收该事件后会先通知Master,随后自己有一系列的记录和相关的处理操作,不过由于这块代码比较简单易懂,且和我们的源码阅读目的关联不大,所以这就不专门介绍。
下面我们详细看下executor的命令构建及其后续操作:
重点:打印executor构建命令
我们到stderr中查看下启动命令:

这个命令很简单,主要是调用CoarseGrainedExecutorBackend这个类,我们知道jvm调用一个类时,程序的入口就是main函数,所以我们去看下该类的main函数源码:
def main(args: Array[String]): Unit = {
//定义一个变量函数,其会创建CoarseGrainedExecutorBackend 对象
val createFn: (RpcEnv, Arguments, SparkEnv, ResourceProfile) =>
CoarseGrainedExecutorBackend = { case (rpcEnv, arguments, env, resourceProfile) =>
new CoarseGrainedExecutorBackend(rpcEnv, arguments.driverUrl, arguments.executorId,
arguments.bindAddress, arguments.hostname, arguments.cores, arguments.userClassPath, env,
arguments.resourcesFileOpt, resourceProfile)
}
//核心逻辑
run(parseArguments(args, this.getClass.getCanonicalName.stripSuffix("$")), createFn)
System.exit(0)
}
main方法中需要留意的是定义一个变量函数,这个函数在调用时会创建CoarseGrainedExecutorBackend 对象,随后是调用该对象的run方法,我们继续深入看下:
def run(
arguments: Arguments,
backendCreateFn: (RpcEnv, Arguments, SparkEnv, ResourceProfile) =>
CoarseGrainedExecutorBackend): Unit = {
Utils.initDaemon(log)
//获取driver的rpc客户端
SparkHadoopUtil.get.runAsSparkUser { () =>
Utils.checkHost(arguments.hostname)
val executorConf = new SparkConf
val fetcher = RpcEnv.create(
"driverPropsFetcher",
arguments.bindAddress,
arguments.hostname,
-1,
executorConf,
new SecurityManager(executorConf),
numUsableCores = 0,
clientMode = true)
var driver: RpcEndpointRef = null
val nTries = 3
for (i <- 0 until nTries if driver == null) {
try {
driver = fetcher.setupEndpointRefByURI(arguments.driverUrl)
} catch {
case e: Throwable => if (i == nTries - 1) {
throw e
}
}
}
//获取driver的配置和属性
val cfg = driver.askSync[SparkAppConfig](RetrieveSparkAppConfig(arguments.resourceProfileId))
val props = cfg.sparkProperties ++ Seq[(String, String)](("spark.app.id", arguments.appId))
fetcher.shutdown()
// 根据我们从driver获取的信息构建SparkEnv
val driverConf = new SparkConf()
for ((key, value) <- props) {
if (SparkConf.isExecutorStartupConf(key)) {
driverConf.setIfMissing(key, value)
} else {
driverConf.set(key, value)
}
}
cfg.hadoopDelegationCreds.foreach { tokens =>
SparkHadoopUtil.get.addDelegationTokens(tokens, driverConf)
}
driverConf.set(EXECUTOR_ID, arguments.executorId)
val env = SparkEnv.createExecutorEnv(driverConf, arguments.executorId, arguments.bindAddress,
arguments.hostname, arguments.cores, cfg.ioEncryptionKey, isLocal = false)
//重点:创建Executor节点,而这个节点就是CoarseGrainedExecutorBackend对象
env.rpcEnv.setupEndpoint("Executor",
backendCreateFn(env.rpcEnv, arguments, env, cfg.resourceProfile))
//创建worker监控节点,用于worker和executor间的信息交互(具体不是很明白)
arguments.workerUrl.foreach { url =>
env.rpcEnv.setupEndpoint("WorkerWatcher", new WorkerWatcher(env.rpcEnv, url))
}
env.rpcEnv.awaitTermination()
}
}
run方法中主要是根据driver的配置创建SparkEnv对象,随后创建了Executor节点,这是一个RPC节点,可以接受TaskScheduler发布过来的任务。如下:

至此,关于Executor的启动过程我们已经知道了,但是我们还不知道Executor的执行本质是什么,所以我们接着看下Executor RPC节点启动的时候会有哪些操作。
override def onStart(): Unit = {
logInfo("Connecting to driver: " + driverUrl)
try {
//资源准备
_resources = parseOrFindResources(resourcesFileOpt)
} catch {
case NonFatal(e) =>
exitExecutor(1, "Unable to create executor due to " + e.getMessage, e)
}
rpcEnv.asyncSetupEndpointRefByURI(driverUrl).flatMap { ref =>
driver = Some(ref)
//重点一:向Driver发送RegisterExecutor事件
ref.ask[Boolean](RegisterExecutor(executorId, self, hostname, cores, extractLogUrls,
extractAttributes, _resources, resourceProfile.id))
}(ThreadUtils.sameThread).onComplete {
case Success(_) =>
//重点二:向driver注册成功Executor后,向自己发送Executor节点
self.send(RegisteredExecutor)
case Failure(e) =>
exitExecutor(1, s"Cannot register with driver: $driverUrl", e, notifyDriver = false)
}(ThreadUtils.sameThread)
}
这块的代码比较简单,首先是准备资源,其次是分别向Driver和自己本身发送Executor的注册事件。下面我们来分别看下:
重点一:向Driver发送RegisterExecutor事件
如果看过我上篇文章的小伙伴看到这肯定会对上篇文章遗留的问题有所触动,那我们就来看下吧:

注意看,这次driver是CoarseGrainedSchedulerBackend对象,并不是我们再driver期间创建的三个节点之一,但是查看下继承关系可以知道,该对象被StandaloneSchedulerBackend对象继承。到这基本可以和之前的内容连上了。我们接着看下事件的大致处理流程:
case RegisterExecutor(executorId, executorRef, hostname, cores, logUrls,
attributes, resources, resourceProfileId) =>
//如果当前executor以及被注册过,则返回异常
if (executorDataMap.contains(executorId)) {
context.sendFailure(new IllegalStateException(s"Duplicate executor ID: $executorId"))
//如果当前executor被设置为黑名单,则同样返回异常
} else if (scheduler.nodeBlacklist.contains(hostname) ||
isBlacklisted(executorId, hostname)) {
logInfo(s"Rejecting $executorId as it has been blacklisted.")
context.sendFailure(new IllegalStateException(s"Executor is blacklisted: $executorId"))
} else {
//获取executor的地址信息
val executorAddress = if (executorRef.address != null) {
executorRef.address
} else {
context.senderAddress
}
logInfo(s"Registered executor $executorRef ($executorAddress) with ID $executorId")
//记录executor地址和executorId的映射关系
addressToExecutorId(executorAddress) = executorId
totalCoreCount.addAndGet(cores)
totalRegisteredExecutors.addAndGet(1)
val resourcesInfo = resources.map{ case (k, v) =>
(v.name,
new ExecutorResourceInfo(v.name, v.addresses,
taskResourceNumParts.getOrElse(v.name, 1)))
}
//封装executor的信息为一个对象
val data = new ExecutorData(executorRef, executorAddress, hostname,
0, cores, logUrlHandler.applyPattern(logUrls, attributes), attributes,
resourcesInfo, resourceProfileId)
//记录当前executor的信息
CoarseGrainedSchedulerBackend.this.synchronized {
executorDataMap.put(executorId, data)
if (currentExecutorIdCounter < executorId.toInt) {
currentExecutorIdCounter = executorId.toInt
}
if (numPendingExecutors > 0) {
numPendingExecutors -= 1
logDebug(s"Decremented number of pending executors ($numPendingExecutors left)")
}
}
listenerBus.post(
SparkListenerExecutorAdded(System.currentTimeMillis(), executorId, data))
context.reply(true)
}
可以看到,StandaloneSchedulerBackend接收到Executor的注册信息后,将记录Executor的信息,不过这还不能够完全解答我们上一章的遗留问题,要想解决需要阅读task的任务提交源码才行,这个我们放到后面章节讲解,但是这里可能也是其铺垫之一,所以这块也要留心下。
重点二:向driver注册成功Executor后,向自己发送Executor节点
到目前为止Executor的启动基本已经结束,但是Executor的执行本质我们还没看到,而这是整体Executor构建流程的最后一步了,所以我们的答案就在这块,首先还是先看下源码:
case RegisteredExecutor =>
logInfo("Successfully registered with driver")
try {
//重点:创建一个Executor对象,其内部封装一个线程池,Task任务也是提交到它的线程池运行
executor = new Executor(executorId, hostname, env, userClassPath, isLocal = false,
resources = _resources)
//想driver发送LaunchedExecutor时间
driver.get.send(LaunchedExecutor(executorId))
} catch {
case NonFatal(e) =>
exitExecutor(1, "Unable to create executor due to " + e.getMessage, e)
}
这块的代码只有两步,一个是创建Executor对象,一个是通知driver Executor已经启动好了,随时可以接受task任务。通知driver没什么好讲的,几乎是老套路,这里不再过多介绍,我们来看下Executor的线程池信息:

Executor有一个私有的成员属性,他就是我们常用的缓冲线程池,所以到此为止,我们本章的源码阅读目的已经达成,我们了解了executor创建的完整流程,并且知道了executor执行的本质其实是用缓冲线程池执行任务。
总结:
1、SparkContext初始化时创建了三个核心RPC节点,分别是SchedulerBackend(实现类是StandaloneSchedulerBackend)、TaskScheduler(实现类是TaskSchedulerImpl)、StandaloneAppClient,原则上这三个RPC节点都处于driver运行期间,泛称上都可以称为driver节点。
2、executor内部封装有缓冲线程池,所以task任务提交到executor执行的本质是最终使用线程池进行处理。
本文深入解析Spark Executor的启动过程,从worker接收到LaunchExecutor事件开始,详细阐述Executor创建ExecutorRunner,启动java线程,构建并执行executor进程的步骤。Executor本质是通过线程池执行Task任务,注册到Driver后等待任务分配。
executor在worker上的创建过程,executor本质是什么,是线程池吗?&spm=1001.2101.3001.5002&articleId=125355262&d=1&t=3&u=fa97d808889b407ea04032d23cfb8e19)
2528

被折叠的 条评论
为什么被折叠?



