Spark-Yarn任务提交流程解析

YARN架构解析：SparkSubmit、ApplicationMaster与ExecutorBackend详解

最新推荐文章于 2025-10-21 09:42:52 发布

原创最新推荐文章于 2025-10-21 09:42:52 发布 · 2.5k 阅读

4 ·

CC 4.0 BY-SA版权

文章标签：

#spark

spark 专栏收录该内容

4 篇文章

订阅专栏

本文详细解读了SparkSubmit的运行流程，从提交脚本到YarnClusterApplication的实例化，再到ApplicationMaster的启动、资源分配和ExecutorBackend的执行，展示了分布式计算框架中的关键组件交互。

一、SparkSubmit

bin/spark-submit \
--class org.apache.spark.examples.SparkPi \
--master yarn \
--deploy-mode cluster \
./examples/jars/spark-examples_2.12-3.0.0.jar \
10

bin/spark-submit \
--class org.apache.spark.examples.SparkPi \
--master yarn \
--deploy-mode client \
./examples/jars/spark-examples_2.12-3.0.0.jar \
10

打开spark-submit 脚本

进入 org.apache.spark.deploy.yarn.YarnClusterApplication

查看 submitApplication

查看 createContainerLaunchContext

小结：

SparkSubmit

   -- main

       -- doSubmit

           // 解析参数
           -- parseArguments

               // master => --master => yarn
               // mainClass => --class => SparkPi(WordCount)
               -- parse

           -- submit

               -- doRunMain

               -- runMain

                   // (childArgs, childClasspath, sparkConf, childMainClass)
                   // 【Cluster】childMainClass => org.apache.spark.deploy.yarn.YarnClusterApplication
                   // 【Client 】childMainClass => SparkPi(WordCount)
                   -- prepareSubmitEnvironment

                   // Class.forName("xxxxxxx")
                   -- mainClass : Class = Utils.classForName(childMainClass)

                   // classOf[SparkApplication].isAssignableFrom(mainClass)
                   -- 【Cluster】a). mainClass.getConstructor().newInstance().asInstanceOf[SparkApplication]
                   -- 【Client 】b). new JavaMainApplication(mainClass)

                   -- app.start

YarnClusterApplication

   -- start

       // new ClientArguments
       // --class => userClass => SparkPI(WordCount)
       -- new Client

       -- client.run

           -- submitApplication

               // 【Cluster】 org.apache.spark.deploy.yarn.ApplicationMaster
               // 【Client 】 org.apache.spark.deploy.yarn.ExecutorLauncher
               --createContainerLaunchContext
               --createApplicationSubmissionContext

二、ApplicationMaster

这里不得不提一下，scala的模式匹配真是优雅

启动Driver

注册AM

资源分配

启动 NodeManager

小结

ApplicationMaster

   -- main

       // --class => userClass => SparkPi(WordCount)
       -- new ApplicationMasterArguments

       -- master = new ApplicationMaster

       -- master.run()

       -- 【Client 】runExecutorLauncher
       -- 【Cluster】runDriver

           -- userClassThread = startUserApplication()

               -- ClassLoader.loadClass(args.userClass)
                   getMethod("main")

               -- new Thread().start()

                   -- run
                       -- mainMethod.invoke(null, userArgs.toArray)

                           -- WordCount

                               -- new SparkContext() // 创建Spark的环境
                               -- 【blocking....................】

           // Blocking -----------------------------------------
           -- val sc = ThreadUtils.awaitResult

           // 注册
           -- registerAM

           // 资源分配
           -- createAllocator

               -- allocator.allocateResources

                   -- handleAllocatedContainers

                       -- runAllocatedContainers

                           // bin/java org.apache.spark.executor.YarnCoarseGrainedExecutorBackend
                           -- prepareCommand

           -- resumeDriver // 执行Spark作业
           -- userClassThread.join()

三、ExecutorBackend

小结

YarnCoarseGrainedExecutorBackend

   -- main

       -- CoarseGrainedExecutorBackend.run

           -- val env = SparkEnv.createExecutorEnv

           -- env.rpcEnv.setupEndpoint("Executor")