Spark资源调度机制源码分析--基于spreadOutApps及非spreadOutApps两种资源调度算法
1、spreadOutApp尽量平均分配到每个executor上;
2、非spreadOutApp尽量在使用单个executor的资源。
源码分析
org.apache.spark.deploy.master.Master
1、首先判断,master状态不是ALIVE的话,直接返回
2、调度driver
3、 Application的调度机制(核心之核心,重中之重)
源码如下:
/*
*schedule()解决了spark资源调度的问题
*/
private def schedule() {
//首先判断,master状态不是ALIVE的话,直接返回
//也就是说,stanby master是不会进行application等资源调度的
if (state != RecoveryState.ALIVE) { return }
// First schedule drivers, they take strict precedence over applications
// Randomization helps balance drivers
//Random.shuffle的原理,大家要清楚,就是对传入的集合的元素进行随机的打乱
//取出了workers中的所有之前注册上来的worker,进行过滤,必须是状态为ALIVE的worker
//对状态为ALIVE的worker,调用Random的shuffle方法进行随机的打乱
val shuffledAliveWorkers = Random.shuffle(workers.toSeq.filter(_.state == WorkerState.ALIVE))
val numWorkersAlive = shuffledAliveWorkers.size
var curPos = 0
//首先,调度driver
//为什么要调度driver,大家想一下,什么情况下,会注册driver,并且会导致driver被调度
//其实 ,只有用yarn-cluster模式提交的时候,才会注册driver;因为standalone和yarn-client模式,都会在本地直接
//启动driver,而不会来注册driver,就更不可能让master调度driver了
//driver调度机制
//遍历waittingDrivers ArrayBuffer
for (driver <- waitingDrivers.toList) { // iterate over a copy of waitingDrivers
// We assign workers to each waiting driver in a round-robin fashion. For each driver, we
// start from the last worker that was assigned a driver, and continue onwards until we have
// explored all alive workers.
var launched = false
var numWorkersVisited = 0
//while的条件,numWorkersVisited小于numWorkersAlive
//什么意思?就是说,只要还有活着的worker没有遍历到,那么就继续进行遍历
//而且,当前这个driver还没有被启动,也就是launched为false
while (numWorkersVisited < numWorkersAlive && !launched) {
val worker = shuffledAliveWorkers(curPos)
numWorkersVisited += 1
//如果当前这个worker的空闲内存量大于等于,driver需要的内存
//并且worker的空闲cpu数量,大于等于driver需要的cpu数量
if (worker.memoryFree >= driver.desc.mem && worker.coresFree >= driver.desc.cores) {
//启动driver
launchDriver(worker, driver)
//并且将driver从waitingDrivers队列中移除
waitingDrivers -= driver
launched = true
}
//将指针指向下一个worker
curPos = (curPos + 1) % numWorkersAlive
}
}
// 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.
// Application的调度机制(核心之核心,重中之重)
// 首先, application的调度算法有两种,一种是spreadOutApps,另一种是非spreadOutApps
if (spreadOutApps) {
// Try to spread out each app among all the nodes, until it has all its cores
//首先,遍历waitingApps中的ApplicationInfo,并且过滤出application还需要高度的cores的application
for (app <- waitingApps if app.coresLeft > 0) {
//从workers中,过滤状态为ALIVE的,再次过滤可以被Application使用的Worker,然后按照剩余cpu数量倒序排序
val usableWorkers = workers.toArray.filter(_.state == WorkerState.ALIVE)
.filter(canUse(app, _)).sortBy(_.coresFree).reverse
val numUsable = usableWorkers.length
//创建一个空数组,存储了要分配给每个worker的cpu数量
val assigned = new Array[Int](numUsable) // Number of cores to give on each node
//获取到底要分配多少cpu,取app剩余要分配的cpu的数量和worker总共可用cpu数量的最小值
var toAssign = math.min(app.coresLeft, usableWorkers.map(_.coresFree).sum)
//通过这种算法,其实会将每个application,要启动的executor,都平均分布到各个worker上去
//比如有20个cpu core要分配,那么实际会循环两遍worker,每次循环,给每个worker分配1个core
//最后每个worker分配了2个core
//while条件,只要要分配的cpu,还没有分配完,就继续循环
var pos = 0
while (toAssign > 0) {
//每一个worker,如果空闲的cpu数量大于,已经分配出去的cpu数量
//也就是说,worker还有可分配的cpu
if (usableWorkers(pos).coresFree - assigned(pos) > 0) {
//将总共要分配的cpu数量-1,因为这里已经决定在这个worker上分配一个cpu了
toAssign -= 1
//给这个worker分配的cpu数量,加1
assigned(pos) += 1
}
//指针移动到下一下worker
pos = (pos + 1) % numUsable
}
// Now that we've decided how many cores to give on each node, let's actually give them
// 给每个worker分配完application要求的cpu core之后
// 遍历worker
for (pos <- 0 until numUsable) {
//只要判断之前给这个worker分配到了core
if (assigned(pos) > 0) {
//首先,在application内部缓存结构中,添加executor
//并且创建ExecutorDesc对象,其中封装了,给这个executor分配多少个cpu core
//在spark-submit脚本中,可以指定要多少executor,每个execuor多少个cpu,多少内存
//那么基于源码机制,实际上,executor的实际数量,以及每个executor的cpu,可能与配置是不一样的
//因为,我人帝里基于总的cpu来分配的,就是比如,要求3个executor,每个要3个cpu,那么比如,有9个workers,每个有1个cpu
//那么其实总其知道,要分配9个core,其实根据这种算法,会给每个worker分配一个core,然后给每个worker启动一个executor
//最后会启动,9个executor,每个executor有1个cpu core
val exec = app.addExecutor(usableWorkers(pos), assigned(pos))
//那么就在worker上启动executor
launchExecutor(usableWorkers(pos), exec)
//将application状态设置为running
app.state = ApplicationState.RUNNING
}
}
}
} else {
// Pack each app into as few nodes as possible until we've assigned all its cores
//非spreadOutApps调度算法
//这种算法与spreadOutApps算法正好相反,1、spreadOutApp尽量平均分配到每个executor上;2、非spreadOutApp尽量在使用单个executor的资源。
//每个application,都尽可能分配到尽量少的worker上去,比如总其有10个worker,每个有10个core
//app总共要分配 20个core,那么其实,只会分配到两个worker上,每个worker都占满10个core
//那么,其余的app,就只能 分配到下一个worker了
//比如,spark-submit里,配置的是要10个executor,每个要2个core,那么总共是20个croe
//只会启动2个executor,每个有10个cores
//将每个Application,尽可能少的分配到worker上去
//首先,遍历worker,并且是状态为ALIVE,还有空闲cpu的worker
for (worker <- workers if worker.coresFree > 0 && worker.state == WorkerState.ALIVE) {
//遍历application,并且是还有城朵分配的core的application
for (app <- waitingApps if app.coresLeft > 0) {
//判断,如果当前这个worker可以被 application使用
if (canUse(app, worker)) {
//取worker剩余cpu数量,与app要分配的cpu数量的最小值
val coresToUse = math.min(worker.coresFree, app.coresLeft)
//如果Worker剩余cpu为0了,就不分配了
if (coresToUse > 0) {
// 给app添加一个executor
val exec = app.addExecutor(worker, coresToUse)
//在worker上启动executor
launchExecutor(worker, exec)
//将application状态设置为running
app.state = ApplicationState.RUNNING
}
}
}
}
}
}