ITPub博客

首页 > 大数据 > 数据分析 > 好程序员大数据分享Spark任务和集群启动流程

好程序员大数据分享Spark任务和集群启动流程

原创 数据分析 作者:好程序员IT 时间:2019-08-19 15:47:02 0 删除 编辑

  好程序员大数据分享Spark任务和集群启动流程,Spark集群启动流程

  1.调用start-all.sh脚本,开始启动Master

  2.Master启动以后,preStart方法调用了一个定时器,定时检查超时的Worker后删除

  3.启动脚本会解析slaves配置文件,找到启动Worker的相应节点.开始启动Worker

  4.Worker服务启动后开始调用preStart方法开始向所有的Master进行注册

  5.Master接收到Worker发送过来的注册信息,Master开始保存注册信息并把自己的URL响应给Worker

  6.Worker接收到Master的URL后并更新,开始调用一个定时器,定时的向Master发送心跳信息

 

任务提交流程

1.Driver端会通过spark-submit脚本启动SaparkSubmit进程,此时创建了一个非常重要的对象(SparkContext),开始向Master发送消息

2.Master接收到发送过来的信息后开始生成任务信息,并把任务信息放到一个对列里

3.Master把所有有效的Worker过滤出来,按照空闲的资源进行排序

4.Master开始向有效的Worker通知拿取任务信息并启动相应的Executor

5.Worker启动Executor并向Driver反向注册

6.Driver开始把生成的task发送给相应的Executor,Executor开始执行任务

 

集群启动流程

1.首先创建Master类

import akka.actor.{Actor, ActorSystem, Props}

import com.typesafe.config.{Config, ConfigFactory}

 

import scala.collection.mutable

import scala.concurrent.duration._

 

class Master(val masterHost: String, val masterPort: Int) extends Actor{

 

  // 用来存储Worker的注册信息

  val idToWorker = new mutable.HashMap[String, WorkerInfo]()

 

  // 用来存储Worker的信息

  val workers = new mutable.HashSet[WorkerInfo]()

 

  // Worker的超时时间间隔

  val checkInterval: Long = 15000

 

 

  // 生命周期方法,在构造器之后,receive方法之前只调用一次

  override def preStart(): Unit = {

    // 启动一个定时器,用来定时检查超时的Worker

    import context.dispatcher

    context.system.scheduler.schedule(0 millis, checkInterval millis, self, CheckTimeOutWorker)

  }

 

  // 在preStart方法之后,不断的重复调用

  override def receive: Receive = {

    // Worker -> Master

    case RegisterWorker(id, host, port, memory, cores) => {

      if (!idToWorker.contains(id)){

        val workerInfo = new WorkerInfo(id, host, port, memory, cores)

        idToWorker += (id -> workerInfo)

        workers += workerInfo

 

        println("a worker registered")

 

        sender ! RegisteredWorker(s"akka.tcp://${Master.MASTER_SYSTEM}" +

          s"@${masterHost}:${masterPort}/user/${Master.MASTER_ACTOR}")

      }

    }

    case HeartBeat(workerId) => {

      // 通过传过来的workerId获取对应的WorkerInfo

      val workerInfo: WorkerInfo = idToWorker(workerId)

      // 获取当前时间

      val currentTime = System.currentTimeMillis()

      // 更新最后一次心跳时间

      workerInfo.lastHeartbeatTime = currentTime

    }

    case CheckTimeOutWorker => {

      val currentTime = System.currentTimeMillis()

      val toRemove: mutable.HashSet[WorkerInfo] =

        workers.filter(w => currentTime - w.lastHeartbeatTime > checkInterval)

 

      // 将超时的Worker从idToWorker和workers中移除

      toRemove.foreach(deadWorker => {

        idToWorker -= deadWorker.id

        workers -= deadWorker

      })

 

      println(s"num of workers: ${workers.size}")

    }

  }

}

object Master{

  val MASTER_SYSTEM = "MasterSystem"

  val MASTER_ACTOR = "Master"

 

  def main(args: Array[String]): Unit = {

    val host = args(0)

    val port = args(1).toInt

 

    val configStr =

      s"""

         |akka.actor.provider = "akka.remote.RemoteActorRefProvider"

         |akka.remote.netty.tcp.hostname = "$host"

         |akka.remote.netty.tcp.port = "$port"

      """.stripMargin

 

    // 配置创建Actor需要的配置信息

    val config: Config = ConfigFactory.parseString(configStr)

 

    // 创建ActorSystem

    val actorSystem: ActorSystem = ActorSystem(MASTER_SYSTEM, config)

 

    // 用actorSystem实例创建Actor

    actorSystem.actorOf(Props(new Master(host, port)), MASTER_ACTOR)

 

    actorSystem.awaitTermination()

 

  }

}

2.创建RemoteMsg特质

trait RemoteMsg extends Serializable{

 

}

 

// Master -> self(Master)

case object CheckTimeOutWorker

 

// Worker -> Master

case class RegisterWorker(id: String, host: String,

                          port: Int, memory: Int, cores: Int) extends RemoteMsg

 

// Master -> Worker

case class RegisteredWorker(masterUrl: String) extends RemoteMsg

 

// Worker -> self

case object SendHeartBeat

 

// Worker -> Master(HeartBeat)

case class HeartBeat(workerId: String) extends RemoteMsg

3.创建Worker类

import java.util.UUID

 

import akka.actor.{Actor, ActorRef, ActorSelection, ActorSystem, Props}

import com.typesafe.config.{Config, ConfigFactory}

 

import scala.concurrent.duration._

 

class Worker(val host: String, val port: Int, val masterHost: String,

             val masterPort: Int, val memory: Int, val cores: Int) extends Actor{

 

  // 生成一个Worker ID

  val workerId = UUID.randomUUID().toString

 

  // 用来存储MasterURL

  var masterUrl: String = _

 

  // 心跳时间间隔

  val heartBeat_interval: Long = 10000

 

  // master的Actor

  var master: ActorSelection = _

 

  override def preStart(){

    // 获取Master的Actor

    master = context.actorSelection(s"akka.tcp://${Master.MASTER_SYSTEM}" +

      s"@${masterHost}:${masterPort}/user/${Master.MASTER_ACTOR}")

 

    master ! RegisterWorker(workerId, host, port, memory, cores)

  }

 

  override def receive: Receive = {

    // Worker接收到Master发送过来的注册成功的信息(masterUrl)

    case RegisteredWorker(masterUrl) => {

      this.masterUrl = masterUrl

      // 启动一个定时器,定时给Master发送心跳

      import context.dispatcher

      context.system.scheduler.schedule(0 millis, heartBeat_interval millis, self, SendHeartBeat)

    }

    case SendHeartBeat => {

      // 向Master发送心跳

      master ! HeartBeat(workerId)

    }

 

  }

 

}

object Worker{

  val WORKER_SYSTEM = "WorkerSystem"

  val WORKER_ACTOR = "Worker"

 

  def main(args: Array[String]): Unit = {

    val host = args(0)

    val port = args(1).toInt

    val masterHost = args(2)

    val masterPort = args(3).toInt

    val memory = args(4).toInt

    val cores = args(5).toInt

 

    val configStr =

      s"""

         |akka.actor.provider = "akka.remote.RemoteActorRefProvider"

         |akka.remote.netty.tcp.hostname = "$host"

         |akka.remote.netty.tcp.port = "$port"

      """.stripMargin

 

    // 配置创建Actor需要的配置信息

    val config: Config = ConfigFactory.parseString(configStr)

 

    // 创建ActorSystem

    val actorSystem: ActorSystem = ActorSystem(WORKER_SYSTEM, config)

 

    // 用actorSystem实例创建Actor

    val worker: ActorRef = actorSystem.actorOf(

      Props(new Worker(host, port, masterHost, masterPort, memory, cores)), WORKER_ACTOR)

 

    actorSystem.awaitTermination()

 

  }

}

4.创建初始化类

class WorkerInfo(val id: String, val host: String, val port: Int,

                 val memory: Int, val cores: Int) {

 

  // 初始化最后一次心跳的时间

  var lastHeartbeatTime: Long = _

 

}

5.本地测试需要传入参数:


来自 “ ITPUB博客 ” ,链接:http://blog.itpub.net/69913892/viewspace-2654145/,如需转载,请注明出处,否则将追究法律责任。

请登录后发表评论 登录
全部评论
欢迎关注公众号:好程序员特训营 web前端教程分享 723729549 大数据教程分享 703503210

注册时间:2019-03-20

  • 博文量
    372
  • 访问量
    240953