作者：搖擺少年夢
微訊號： zhouzhihubeyond

spark-submit 指令碼應用程式提交流程

在執行Spar應用程式時，會將spark應用程式打包後使用spark-submit指令碼提交到Spark中執行，執行提交命令如下：

root@sparkmaster:/hadoopLearning/spark-1.5.0-bin-hadoop2.4/bin# 
./spark-submit --master spark://sparkmaster:7077 
--class SparkWordCount --executor-memory 1g
 /root/IdeaProjects/SparkWordCount/out
 
/artifacts/SparkWord
 Count_jar/SparkWordCount.jar  hdfs://ns1/README.md
 hdfs://ns1/SparkWordCountResult

為弄清楚整個流程，我們先來分析一下spark-submit指令碼，spark-submit指令碼內容如下：

#!/usr/bin/env bash

SPARK_HOME="$(cd "`dirname "$0"`"/..; pwd)"

# disable randomized hash for string in Python 3.3+
export PYTHONHASHSEED=0

#spark-submit最終呼叫的是spark-class指令碼
 

#傳入的類是org.apache.spark.deploy.SparkSubmit
#及其它傳入的引數，如deploy mode、executor-memory等
exec "$SPARK_HOME"/bin/spark-class org.apache.spark.deploy.SparkSubmit "[email protected]"

spark-class指令碼會載入spark配置的環境變數資訊、定位依賴包spark-assembly-1.5.0-hadoop2.4.0.jar檔案（以spark1.5.0為例）等，然後再呼叫org.apache.spark.launcher.Main正式啟動Spark應用程式的執行，具體如下：

# Figure out where Spark is installed
#定位SAPRK_HOME目錄
export SPARK_HOME="$(cd "`dirname "$0"`"/..; pwd)"

#載入load-spark-env.sh，執行環境相關資訊
#例如配置檔案conf下的spark-env.sh等
. "$SPARK_HOME"/bin/load-spark-env.sh

# Find the java binary
# 定位JAVA_HOME目錄
if [ -n "${JAVA_HOME}" ]; then
  RUNNER="${JAVA_HOME}/bin/java"
else
  if [ `command -v java` ]; then
    RUNNER="java"
  else
    echo "JAVA_HOME is not set" >&2
    exit 1
  fi
fi

# Find assembly jar
#定位spark-assembly-1.5.0-hadoop2.4.0.jar檔案（以spark1.5.0為例）
#這意味著任務提交時無需將該JAR檔案打包
SPARK_ASSEMBLY_JAR=
if [ -f "$SPARK_HOME/RELEASE" ]; then
  ASSEMBLY_DIR="$SPARK_HOME/lib"
else
  ASSEMBLY_DIR="$SPARK_HOME/assembly/target/scala-$SPARK_SCALA_VERSION"
fi

num_jars="$(ls -1 "$ASSEMBLY_DIR" | grep "^spark-assembly.*hadoop.*\.jar$" | wc -l)"
if [ "$num_jars" -eq "0" -a -z "$SPARK_ASSEMBLY_JAR" ]; then
  echo "Failed to find Spark assembly in $ASSEMBLY_DIR." 1>&2
  echo "You need to build Spark before running this program." 1>&2
  exit 1
fi
ASSEMBLY_JARS="$(ls -1 "$ASSEMBLY_DIR" | grep "^spark-assembly.*hadoop.*\.jar$" || true)"
if [ "$num_jars" -gt "1" ]; then
  echo "Found multiple Spark assembly jars in $ASSEMBLY_DIR:" 1>&2
  echo "$ASSEMBLY_JARS" 1>&2
  echo "Please remove all but one jar." 1>&2
  exit 1
fi

SPARK_ASSEMBLY_JAR="${ASSEMBLY_DIR}/${ASSEMBLY_JARS}"

LAUNCH_CLASSPATH="$SPARK_ASSEMBLY_JAR"

# Add the launcher build dir to the classpath if requested.
if [ -n "$SPARK_PREPEND_CLASSES" ]; then
  LAUNCH_CLASSPATH="$SPARK_HOME/launcher/target/scala-$SPARK_SCALA_VERSION/classes:$LAUNCH_CLASSPATH"
fi

export _SPARK_ASSEMBLY="$SPARK_ASSEMBLY_JAR"

# The launcher library will print arguments separated by a NULL character, to allow arguments with
# characters that would be otherwise interpreted by the shell. Read that in a while loop, populating
# an array that will be used to exec the final command.
#執行org.apache.spark.launcher.Main作為Spark應用程式的主入口
CMD=()
while IFS= read -d '' -r ARG; do
  CMD+=("$ARG")
done < <("$RUNNER" -cp "$LAUNCH_CLASSPATH" org.apache.spark.launcher.Main "[email protected]")
exec "${CMD[@]}"

從上述程式碼中，可以看到，通過org.apache.spark.launcher.Main類啟動org.apache.spark.deploy.SparkSubmit的執行，SparkSubmit部分原始碼如下：

//SparkSubmit Main方法
def main(args: Array[String]): Unit = {
    //任務提交時設定的引數，見圖2
    val appArgs = new SparkSubmitAarguments(args)
    if (appArgs.verbose) {
      // scalastyle:off println
      printStream.println(appArgs)
      // scalastyle:on println
    }
    appArgs.action match {
      //任務提交時，執行submit(appArgs)
      case SparkSubmitAction.SUBMIT => submit(appArgs)
      case SparkSubmitAction.KILL => kill(appArgs)
      case SparkSubmitAction.REQUEST_STATUS => requestStatus(appArgs)
    }
  }

圖1 appArgs = new SparkSubmitAarguments(args)引數

進入submit方法：

  /**
   * Submit the application using the provided parameters.
   *
   * This runs in two steps. First, we prepare the launch environment by setting up
   * the appropriate classpath, system properties, and application arguments for
   * running the child main class based on the cluster manager and the deploy mode.
   * Second, we use this launch environment to invoke the main method of the child
   * main class.
   */
  private def submit(args: SparkSubmitArguments): Unit = {
   //執行引數等資訊，見圖2
    val (childArgs, childClasspath, sysProps, childMainClass) = prepareSubmitEnvironment(args)
    //定義在submit方法中的方法doRunMain()
    def doRunMain(): Unit = {
      if (args.proxyUser != null) {
        val proxyUser = UserGroupInformation.createProxyUser(args.proxyUser,
          UserGroupInformation.getCurrentUser())
        try {
          proxyUser.doAs(new PrivilegedExceptionAction[Unit]() {
            override def run(): Unit = {
              //執行runMain方法
              runMain(childArgs, childClasspath, sysProps, childMainClass, args.verbose)
            }
          })
        } catch {
          case e: Exception =>
            // Hadoop's AuthorizationException suppresses the exception's stack trace, which
            // makes the message printed to the output by the JVM not very helpful. Instead,
            // detect exceptions with empty stack traces here, and treat them differently.
            if (e.getStackTrace().length == 0) {
              // scalastyle:off println
              printStream.println(s"ERROR: ${e.getClass().getName()}: ${e.getMessage()}")
              // scalastyle:on println
              exitFn(1)
            } else {
              throw e
            }
        }
      } else {
        //執行runMain方法
        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 {
        // scalastyle:off println
        printStream.println("Running Spark using the REST application submission protocol.")
        // scalastyle:on println
        //呼叫submit方法中的doRunMain方法
        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 {
       //呼叫submit方法中的doRunMain方法
      doRunMain()
    }
  }

圖2 任務提交時設定的引數，

從上面的程式碼可以看到，最終呼叫的是runMain方法，其原始碼如下：

/**
   * Run the main method of the child class using the provided launch environment.
   *
   * Note that this main class will not be the one provided by the user if we're
   * running cluster deploy mode or python applications.
   */
  private def runMain(
      childArgs: Seq[String],
      childClasspath: Seq[String],
      sysProps: Map[String, String],
      childMainClass: String,
      verbose: Boolean): Unit = {
    // scalastyle:off println
    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")
    }
    // scalastyle:on println

    val loader =
      if (sysProps.getOrElse("spark.driver.userClassPathFirst", "false").toBoolean) {
        new ChildFirstURLClassLoader(new Array[URL](0),
          Thread.currentThread.getContextClassLoader)
      } else {
        new MutableURLClassLoader(new Array[URL](0),
          Thread.currentThread.getContextClassLoader)
      }
    Thread.currentThread.setContextClassLoader(loader)

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

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

    var mainClass: Class[_] = null

    try {
      //複用反射載入childMainClass，這裡為SparkWordCount
      mainClass = Utils.classForName(childMainClass)
    } catch {
      case e: ClassNotFoundException =>
        e.printStackTrace(printStream)
        if (childMainClass.contains("thriftserver")) {
          // scalastyle:off println
          printStream.println(s"Failed to load main class $childMainClass.")
          printStream.println("You need to build Spark with -Phive and -Phive-thriftserver.")
          // scalastyle:on println
        }
        System.exit(CLASS_NOT_FOUND_EXIT_STATUS)
    }

    // SPARK-4170
    if (classOf[scala.App].isAssignableFrom(mainClass)) {
      printWarning("Subclasses of scala.App may not work correctly. Use a main() method instead.")
    }

    //呼叫反射機制載入main方法,即SparkWordCount中的main方法
    val mainMethod = mainClass.getMethod("main", new Array[String](0).getClass)
    if (!Modifier.isStatic(mainMethod.getModifiers)) {
      throw new IllegalStateException("The main method in the given main class must be static")
    }

    def findCause(t: Throwable): Throwable = t match {
      case e: UndeclaredThrowableException =>
        if (e.getCause() != null) findCause(e.getCause()) else e
      case e: InvocationTargetException =>
        if (e.getCause() != null) findCause(e.getCause()) else e
      case e: Throwable =>
        e
    }

    try {
      //執行main方法，即執行SparkWordCount
      mainMethod.invoke(null, childArgs.toArray)
    } catch {
      case t: Throwable =>
        throw findCause(t)
    }
  }

mainMethod.invoke(null, childArgs.toArray)方法執行完畢後，進入SparkWordCount的main方法，執行Spark應用程式，如下圖

至此，正式完成Spark應用程式執行的提交。