Spark资源调度

一:任务调度和资源调度的区别:

任务调度是指通过DAGScheduler,TaskScheduler,SchedulerBackend完成的job的调度

资源调度是指应用程序获取资源的调度,他是通过schedule方法完成的

二:资源调度解密

因为master负责资源管理和调度,所以资源调度的方法schedule位于master.scala这个了类中,当注册程序或者资源发生改变的都会导致schedule的调用,例如注册程序的时候(包括worker,driver和application的注册等,注意executor是向SparkDeploySchedulerBackend注册的)

case RegisterApplication(description, driver) => {
  // TODO Prevent repeated registrations from some driver
  if (state == RecoveryState.STANDBY) {
    // ignore, don't send response
  } else {
    logInfo("Registering app " + description.name)
    val app = createApplication(description, driver)
    registerApplication(app)
    logInfo("Registered app " + description.name + " with ID " + app.id)
    persistenceEngine.addApplication(app)
    driver.send(RegisteredApplication(app.id, self))
    schedule()
  }
**
 * Schedule the currently available resources among waiting apps. This method will be called
 * every time a new app joins or resource availability changes.
 */

每当新的应用程序加入或者可用资源发生改变(比如exccutor或者worker增加或者减少的时候)的时候,该方法都会发生响应

private def schedule(): Unit = {
  if (state != RecoveryState.ALIVE) { return }//判断Master的状态是否为ALIVE,如果不是,则调度没有任何意义
  // Drivers take strict precedence over executors
  val shuffledWorkers = Random.shuffle(workers) // Randomization helps balance drivers
 
//将workers随机化,有利于负载均衡
  for (worker <- shuffledWorkers if worker.state == WorkerState.ALIVE) {//判断worker的状态,只有Alive级别的worker才能参与资源的分配工作
    for (driver <- waitingDrivers) {//循环遍历等待中的driver,当然这里指的是cluster模式,如果是client模式的话,driver就自动启动了。
      if (worker.memoryFree >= driver.desc.mem && worker.coresFree >= driver.desc.cores) {
 
//当worker的free内存和cpu比driver所需要的多的时候,将driver放到workers中随机的一个worker,启动driver
        launchDriver(worker, driver)
        waitingDrivers -= driver//将启动的driver在等待队列中移除。
      }
    }
  }
  startExecutorsOnWorkers()
}

schedule的代码解析(简单的就放在上面的代码注释里了)

Random.shuffle(workers)  将worker在master缓存数据结构中的顺序打乱
 
def shuffle[T, CC[X] <: TraversableOnce[X]](xs: CC[T])(implicit bf: CanBuildFrom[CC[T], T, CC[T]]): CC[T] = {
  val buf = new ArrayBuffer[T] ++= xs//构建一个临时的缓冲数组
 
  def swap(i1: Int, i2: Int) {//交换数组中指定下表的两个元素
    val tmp = buf(i1)
    buf(i1) = buf(i2)
    buf(i2) = tmp
  }
 
  for (n <- buf.length to 2 by -1) {//生成随机数,并不停交换,打乱了数组中元素的顺序
    val k = nextInt(n)
    swap(n - 1, k)
  }
 
  (bf(xs) ++= buf).result//返回随机化的新集合(这里就是workers的集合了)
}

2 waitingDrivers

private val waitingDrivers = new ArrayBuffer[DriverInfo]
 
可以看到这里waitingDrivers是一个数据元素为DriverInfo的数组,DriverInfo包含了driver的信息startTime(启动时间),id,desc(driver的描述信息),submitDate(提交日期)
 
private[deploy] class DriverInfo(
    val startTime: Long,
    val id: String,
    val desc: DriverDescription,
    val submitDate: Date)
  extends Serializable {
 
其中描述信息包含了一下内容
 
private[deploy] case class DriverDescription(
    jarUrl: String,//jar包地址
    mem: Int,//内存信息
    cores: Int,//CPU
    supervise: Boolean//当spark-submit指定driver在cluster模式下运行的话如果设定了supervise,driver挂掉的时候回自动重启,
    command: Command) {//一些环境信息
 
  override def toString: String = s"DriverDescription (${command.mainClass})"
}

3 launchDriver spark只有先启动driver才能进行后面具体的调度

private def launchDriver(worker: WorkerInfo, driver: DriverInfo) {
  logInfo("Launching driver " + driver.id + " on worker " + worker.id)
  worker.addDriver(driver)//表明driver运行的worker
  driver.worker = Some(worker)//driver和worker的相互引用
  worker.endpoint.send(LaunchDriver(driver.id, driver.desc))//master通过远程rpc发指令给worker,让worker启动driver。
  driver.state = DriverState.RUNNING//启动之后将driver的状态转为RUNNING
}

4 startExecutorsOnWorkers 先进先出的队列方式进行简单调度,spark默认启动Executor的方式是FIFO的方式,只有前一个app满足了资源分配的基础上,才会为下一个应用程序分配资源

/**
 * Schedule and launch executors on workers
 */
private def startExecutorsOnWorkers(): Unit = {
  // Right now this is a very simple FIFO scheduler. We keep trying to fit in the first app
  // in the queue, then the second app, etc.
  for (app <- waitingApps if app.coresLeft > 0) {//为应用程序具体分配Executor之前会判断当前应用程序是否还需要cores,
如果不需要则不会为应用程序分配Executor
val coresPerExecutor: Option[Int] = app.desc.coresPerExecutor//应用程序所需要的cores // Filter out workers that don't have enough resources to launch an executor //过滤掉不满足条件的worker,条件为:worker的状态必须是AlIVE的,worker的内存和cpu必须比每一个Executor所需要的大。 //过滤完之后,按照可用cores进行排序,并将大的放到前面,最优的最先使用。 val usableWorkers = workers.toArray.filter(_.state == WorkerState.ALIVE) .filter(worker => worker.memoryFree >= app.desc.memoryPerExecutorMB && worker.coresFree >= coresPerExecutor.getOrElse(1)) .sortBy(_.coresFree).reverse //这里采用spreadOutApps的方式来让应用程序尽可能分散的运行在每一个Node上,这种方式往往能顺便带来更好的数据本地性,通常数据是分散的分布在各台机器上,这种方式通常也是默认的。这方法返回的是每一个分配给每一个worker的cores的数组。具体的在分配cores的时候回尽可能的满足当前所需的 val assignedCores = scheduleExecutorsOnWorkers(app, usableWorkers, spreadOutApps) // Now that we've decided how many cores to allocate on each worker, let's allocate them //下面进行真正的分配Executors,Master通过远程通信发指令给Worker来启动ExecutorBackend进程,向driver发送ExecutorAdded通信。 for (pos <- 0 until usableWorkers.length if assignedCores(pos) > 0) { allocateWorkerResourceToExecutors( app, assignedCores(pos), coresPerExecutor, usableWorkers(pos)) } } }

 

posted @ 2018-11-15 23:26  大葱拌豆腐  阅读(398)  评论(0编辑  收藏  举报