Spark学习之3：SparkSubmit启动应用程序主类过程

本文主要讲述在standalone模式下，从bin/spark-submit脚本到SparkSubmit类启动应用程序主类的过程。

1 调用流程图

2 启动脚本

2.1 bin/spark-submit

# For client mode, the driver will be launched in the same JVM that launches
# SparkSubmit, so we may need to read the properties file for any extra class
# paths, library paths, java options and memory early on. Otherwise, it will
# be too late by the time the driver JVM has started.
if [[ "$SPARK_SUBMIT_DEPLOY_MODE" == "client" && -f "$SPARK_SUBMIT_PROPERTIES_FILE" ]]; then
  # Parse the properties file only if the special configs exist
  contains_special_configs=$(
    grep -e "spark.driver.extra*\|spark.driver.memory" "$SPARK_SUBMIT_PROPERTIES_FILE" | \
    grep -v "^[[:space:]]*#"
  )
  if [ -n "$contains_special_configs" ]; then
    export SPARK_SUBMIT_BOOTSTRAP_DRIVER=1
  fi
fi
exec "$SPARK_HOME"/bin/spark-class org.apache.spark.deploy.SparkSubmit "${ORIG_ARGS[@]}"

（1）在client模式，且spark属性配置文件中包含spark.driver*等配置信息，则export SPARK_SUBMIT_BOOTSTRAP_DRIVER=1；

（2）调用bin/spark-class；

2.2 bin/spark-class

# In Spark submit client mode, the driver is launched in the same JVM as Spark submit itself.
# Here we must parse the properties file for relevant "spark.driver.*" configs before launching
# the driver JVM itself. Instead of handling this complexity in Bash, we launch a separate JVM
# to prepare the launch environment of this driver JVM.
if [ -n "$SPARK_SUBMIT_BOOTSTRAP_DRIVER" ]; then
  # This is used only if the properties file actually contains these special configs
  # Export the environment variables needed by SparkSubmitDriverBootstrapper
  export RUNNER
  export CLASSPATH
  export JAVA_OPTS
  export OUR_JAVA_MEM
  export SPARK_CLASS=1
  shift # Ignore main class (org.apache.spark.deploy.SparkSubmit) and use our own
  exec "$RUNNER" org.apache.spark.deploy.SparkSubmitDriverBootstrapper "$@"
else
  # Note: The format of this command is closely echoed in SparkSubmitDriverBootstrapper.scala
  if [ -n "$SPARK_PRINT_LAUNCH_COMMAND" ]; then
    echo -n "Spark Command: " 1>&2
    echo "$RUNNER" -cp "$CLASSPATH" $JAVA_OPTS "$@" 1>&2
    echo -e "========================================\n" 1>&2
  fi
  exec "$RUNNER" -cp "$CLASSPATH" $JAVA_OPTS "$@"
fi

（1）如果$SPARK_SUBMIT_BOOTSTRAP_DRIVER不为空，则从参数列表中移除SparkSubmit，启动SparkSubmitDriverBootstrapper；SparkSubmitDriverBootstrapper最终会调用SparkSubmit；

（2）直接启动SparkSubmit；

3 应用程序主类的启动方式

应用程序主类有三种启动方式：

（1）有SparkSubmit直接启动；

（2）在Worker节点创建新进程DriverWrapper，由它来启动应用程序主类；

（3）通过StandaloneRestClient向StandaloneRestServer消息，后面的步骤同第2中方式（这种方式是spark1.3新增）。

4 SparkSubmit

4.1 main

  def main(args: Array[String]): Unit = {
    val appArgs = new SparkSubmitArguments(args)
    if (appArgs.verbose) {
      printStream.println(appArgs)
    }
    appArgs.action match {
      case SparkSubmitAction.SUBMIT => submit(appArgs)
      case SparkSubmitAction.KILL => kill(appArgs)
      case SparkSubmitAction.REQUEST_STATUS => requestStatus(appArgs)
    }
  }

（1）创建SparkSubmitArguments对象，并解析参数来初始化对象成员；

（2）这里只分析submit过程。

4.2 SparkSubmitArguments

此类封装了Spark参数。

  // Set parameters from command line arguments
  parseOpts(args.toList)
  // Populate `sparkProperties` map from properties file
  mergeDefaultSparkProperties()
  // Use `sparkProperties` map along with env vars to fill in any missing parameters
  loadEnvironmentArguments()
  validateArguments()

（1）解析命令行参数；

（2）合并属性文件（默认为spark-defaults.conf）中配置的参数；

（3）loadEnvironmentArguments方法会设置action参数，默认为SUBMIT；

    // Action should be SUBMIT unless otherwise specified
    action = Option(action).getOrElse(SUBMIT)

（4）验证参数，不同的action有不同的验证方法；

  /** Ensure that required fields exists. Call this only once all defaults are loaded. */
  private def validateArguments(): Unit = {
    action match {
      case SUBMIT => validateSubmitArguments()
      case KILL => validateKillArguments()
      case REQUEST_STATUS => validateStatusRequestArguments()
    }
  }

4.3 Spark属性参数优先级

spark提供了多种设置属性的方式（优先级从高到底）：

（1）在应用程序中设置属性参数；

（2）在命令行中设置参数；

（3）在配置文件中设置参数（默认为spark-defaults.conf）；

（4）Spark提供的默认属性值。

4.4 SparkSubmit.submit

    val (childArgs, childClasspath, sysProps, childMainClass) = prepareSubmitEnvironment(args)
    def doRunMain(): Unit = {
      if (args.proxyUser != null) {
       ......
      } else {
        runMain(childArgs, childClasspath, sysProps, childMainClass, args.verbose)
      }
    }
     // In standalone cluster mode, there are two submission gateways:
     //   (1) The traditional Akka gateway using o.a.s.deploy.Client as a wrapper
     //   (2) The new REST-based gateway introduced in Spark 1.3
     // The latter is the default behavior as of Spark 1.3, but Spark submit will fail over
     // to use the legacy gateway if the master endpoint turns out to be not a REST server.
    if (args.isStandaloneCluster && args.useRest) {
      try {
        printStream.println("Running Spark using the REST application submission protocol.")
        doRunMain()
      } catch {
        // Fail over to use the legacy submission gateway
        case e: SubmitRestConnectionException =>
          printWarning(s"Master endpoint ${args.master} was not a REST server. " +
            "Falling back to legacy submission gateway instead.")
          args.useRest = false
          submit(args)
      }
    // In all other modes, just run the main class as prepared
    } else {
      doRunMain()
    }

（1）prepareSubmitEnvironment其中一个职责就是设置childMainClass，它决定了应用程序主类的调用方式；

（2）调用doRunMain内部方法，它将调用runMain方法。

4.4.1 SparkSubmit.prepareSubmitEnvironment

（0）设置应用程序部署方式：

    // Set the deploy mode; default is client mode
    var deployMode: Int = args.deployMode match {
      case "client" | null => CLIENT
      case "cluster" => CLUSTER
      case _ => printErrorAndExit("Deploy mode must be either client or cluster"); -1
    }

（1）将childMainClass设置为应用程序主类名：

    // In client mode, launch the application main class directly
    // In addition, add the main application jar and any added jars (if any) to the classpath
    if (deployMode == CLIENT) {
      childMainClass = args.mainClass
      if (isUserJar(args.primaryResource)) {
        childClasspath += args.primaryResource
      }
      if (args.jars != null) { childClasspath ++= args.jars.split(",") }
      if (args.childArgs != null) { childArgs ++= args.childArgs }
    }

（2）将childMainClass设置为StandaloneRestClient或Client：

    // In standalone cluster mode, use the REST client to submit the application (Spark 1.3+).
    // All Spark parameters are expected to be passed to the client through system properties.
    if (args.isStandaloneCluster) {
      if (args.useRest) {
        childMainClass = "org.apache.spark.deploy.rest.StandaloneRestClient"
        childArgs += (args.primaryResource, args.mainClass)
      } else {
        // In legacy standalone cluster mode, use Client as a wrapper around the user class
        childMainClass = "org.apache.spark.deploy.Client"
        if (args.supervise) { childArgs += "--supervise" }
        Option(args.driverMemory).foreach { m => childArgs += ("--memory", m) }
        Option(args.driverCores).foreach { c => childArgs += ("--cores", c) }
        childArgs += "launch"
        childArgs += (args.master, args.primaryResource, args.mainClass)
      }
      if (args.childArgs != null) {
        childArgs ++= args.childArgs
      }
    }

（3）将 childMainClass设置为org.apache.spark.deploy.yarn.Client：

    // In yarn-cluster mode, use yarn.Client as a wrapper around the user class
    if (isYarnCluster) {
      childMainClass = "org.apache.spark.deploy.yarn.Client"
      ......

4.4.2 SparkSubmit.runMain

（0）方法签名：

  private def runMain(
      childArgs: Seq[String],
      childClasspath: Seq[String],
      sysProps: Map[String, String],
      childMainClass: String,
      verbose: Boolean): Unit

（1）可在命令行中设置verbose参数，将runMain的参数输出到控制台：

    if (verbose) {
      printStream.println(s"Main class:\n$childMainClass")
      printStream.println(s"Arguments:\n${childArgs.mkString("\n")}")
      printStream.println(s"System properties:\n${sysProps.mkString("\n")}")
      printStream.println(s"Classpath elements:\n${childClasspath.mkString("\n")}")
      printStream.println("\n")
    }

（2）加载jar

    for (jar <- childClasspath) {
      addJarToClasspath(jar, loader)
    }

（3）将Spark属性参数设置为系统属性（很多地方采用从System属性中获取参数）：

    for ((key, value) <- sysProps) {
      System.setProperty(key, value)
    }

（4）创建childMainClass的类对象：

    try {
      mainClass = Class.forName(childMainClass, true, loader)
    } catch {
      ...
    }

（5）获取main方法对象：

    val mainMethod = mainClass.getMethod("main", new Array[String](0).getClass)

（6）调用main方法：

    try {
      mainMethod.invoke(null, childArgs.toArray)
    } catch {
      case t: Throwable =>
        throw findCause(t)
    }

到此，就已经调用prepareSubmitEnvironment方法设置的childMainClass类了。

childMainClass值为：

• 应用程序主类名；
• org.apache.spark.deploy.rest.StandaloneRestClient；
  
• org.apache.spark.deploy.Client；
  
• org.apache.spark.deploy.yarn.Client。

接下来就是通过Master、Worker启动DriverWrapper进程，进而启动应用程序主类的过程。