参考《Spark内核设计的艺术:架构设计与实现——耿嘉安》
NettyRpcEnv概述
Spark的NettyRpc环境的一些重要组件:
private[netty] val transportConf = SparkTransportConf.fromSparkConf(...)
private val dispatcher: Dispatcher = new Dispatcher(this, numUsableCores)
private val streamManager = new NettyStreamManager(this)
private val transportContext = new TransportContext(transportConf,
new NettyRpcHandler(dispatcher, this, streamManager))
//用于创建TransportClient的工厂类
private val clientFactory = transportContext.createClientFactory(createClientBootstraps())
//volatile 关键字保证变量在多线程之间的可见性
@volatile private var server: TransportServer = _
绪:RpcEndpoint&RpcEndpointRef
RpcEndpoint
RpcEndpoint是对能够处理RPC请求,给某一特定服务提供本地及跨节点调用的RPC组件的抽象,所有运行于RPC框架上的实体都应该继承trait RPCEndpoint。
package org.apache.spark.rpc
import org.apache.spark.SparkException
//创建RpcEnv的工厂类,必须有一个空构造函数才能通过反射创建
private[spark] trait RpcEnvFactory {
def create(config: RpcEnvConfig): RpcEnv
}
private[spark] trait RpcEndpoint {
//当前RpcEndpoint所属的RpcEnv
val rpcEnv: RpcEnv
//获取RpcEndpoint相关联的RpcEndpointRef
final def self: RpcEndpointRef = {
require(rpcEnv != null, "rpcEnv has not been initialized")
rpcEnv.endpointRef(this)
}
//接收消息并处理,不回复客户端
def receive: PartialFunction[Any, Unit] = {
case _ => throw new SparkException(self + " does not implement 'receive'")
}
//接收消息并处理,通过RpcCallContext回复客户端
def receiveAndReply(context: RpcCallContext): PartialFunction[Any, Unit] = {
case _ => context.sendFailure(new SparkException(self + " won't reply anything"))
}
//onError、onConnected、onDisconnected、onNetworkError、onStart、onStop顾名思义
//用于停止当前RpcEndpoint,注意onStop是trait定义的抽象方法,在停止RpcEndpoint时调用,做一些收尾工作
final def stop(): Unit = {
val _self = self
if (_self != null) {
rpcEnv.stop(_self)
}
}
}
//线程安全的、串行处理消息的ThreadSafeRpcEndpoint
private[spark] trait ThreadSafeRpcEndpoint extends RpcEndpoint
trait ThreadSafeRpcEndpoint/... extends RpcEndpoint
ThreadSafeRpcEndpoint主要用于消息的串行处理,必须是线程安全的
Master/Worker/HeartbeatReceiver/... extends ThreadSafeRpcEndpoint
RpcEndpointRef
要向一个远端RpcEndpoint发送请求,就必须持有这个RpcEndpoint的远程引用RpcEndpointRef,它是线程安全的。
private[spark] abstract class RpcEndpointRef(conf: SparkConf)
extends Serializable with Logging {
//rpc最大重连次数,默认3,可使用spark.rpc.numRetries属性配置
private[this] val maxRetries = RpcUtils.numRetries(conf)
//rpc每次重连等待的毫秒数,默认3s,可使用spark.rpc.retry.wait属性配置
private[this] val retryWaitMs = RpcUtils.retryWaitMs(conf)
//rpc的ask操作默认超时时间,默认120s,可使用spark.rpc.askTimeout(优先级高)/spark.network.timeout属性配置
private[this] val defaultAskTimeout = RpcUtils.askRpcTimeout(conf)
//返回当前RpcEndpointRef对应的RpcEndpoint的RPC地址
def address: RpcAddress
//返回当前RpcEndpointRef对应的RpcEndpoint的名称
def name: String
//发送单向异步的消息到相应的RpcEndpoint.receive。
def send(message: Any): Unit
//发送一条消息到相应的RpcEndpoint.receiveAndReply,并在指定的超时内接收处理结果。此方法只发送消息一次,从不重试。
def ask[T: ClassTag](message: Any, timeout: RpcTimeout): Future[T]
def ask[T: ClassTag](message: Any): Future[T] = ask(message, defaultAskTimeout)
//发送同步请求到相应的RpcEndpoint.receiveAndReply,并在超时时间内等待处理结果,当抛出异常时会请求重试次数以内的重连。
def askSync[T: ClassTag](message: Any): T = askSync(message, defaultAskTimeout)
def askSync[T: ClassTag](message: Any, timeout: RpcTimeout): T = {
val future = ask[T](message, timeout)
timeout.awaitResult(future)
}
}
消息投递规则:
at-most-once:投递0或1此,消息可能会丢失
at-least-once:潜在地多次投递并保证至少成功一次,消息可能会重复
exactly-once:准确发送一次,消息不会丢失也不会重复
1 TransportConf
RPC传输上下文配置类,用于创建TransportClientFactory和TransportServer。
//通过SparkTransportConf的fromSparkConf方法来构建TransportConf需要三个参数:sparkConf、模块名module和可用内核数
private[netty] val transportConf = SparkTransportConf.fromSparkConf(
//先克隆SparkConf并设置节点间取数据的连接数
conf.clone.set("spark.rpc.io.numConnectionsPerPeer", "1"),
//设置模块名
"rpc",
//Netty传输线程数,如果小于或等于0,线程数就是系统可用处理器的数量,最多为8线程。
conf.getInt("spark.rpc.io.threads", numUsableCores))
2 Dispatcher
Dispatcher负责将消息路由到应该对此消息处理的RpcEndpoint,可以提高NettyRpcEnv对消息的异步处理和并行处理能力。
private val dispatcher: Dispatcher = new Dispatcher(this, numUsableCores)
基本概念:
InboxMessage:Inbox盒子内的消息,是一个trait,所有类型的RPC消息都要继承自InboxMessage。
Inbox:端点内的盒子,每个RpcEndpoint都有一个对应的盒子,这个盒子有存储InboxMessage的列表messages,所有的消息都缓存在messages并由RpcEndpoint异步处理。
EndpointData:RPC端点数据,包括RpcEndpoint、NettyRpcEndpointRef和Inbox等属于同一个端点的实例。
endpoints:端点实例RpcEndpoint与EndpointData之间映射关系的缓存。
endpointRefs:端点实例RpcEndpoint与RpcEndpointRef之间映射关系的缓存.
receivers:存储EndpointData的阻塞队列,只有Inbox中有消息的EndpointData才会被放入此队列。
stopped:Dispatcher是否停止的状态。
threadPool:用于对消息进行调度的线程池,里面运行的任务都是MessageLoop。
2.1 Dispatcher注册RpcEndpoint
def registerRpcEndpoint(name: String, endpoint: RpcEndpoint): NettyRpcEndpointRef = {
//使用当前RpcEndpoint所在的NettyRpcEnv的地址和RpcEndpoint的名称创建RpcEndpointAddress对象
val addr = RpcEndpointAddress(nettyEnv.address, name)
//创建RpcEndpoint的引用对象
val endpointRef = new NettyRpcEndpointRef(nettyEnv.conf, addr, nettyEnv)
synchronized {
if (stopped) {
throw new IllegalStateException("RpcEnv has been stopped")
}
if (endpoints.putIfAbsent(name, new EndpointData(name, endpoint, endpointRef)) != null) {
throw new IllegalArgumentException(s"There is already an RpcEndpoint called $name")
}
//创建EndpointData并放入endpoints缓存
val data = endpoints.get(name)
//将RpcEndpoint与NettyRpcEndpointRef映射关系放入endpointRefs缓存
endpointRefs.put(data.endpoint, data.ref)
//将EndpointData放入阻塞队列receivers,由于EndpointData是新建的,内部会新建Inbox并执行Inbox的主构造函数,
//向Inbox自身的messages列表中放入OnStart消息,MessageLoop线程会取出此EndpointData并调用当前Inbox的process方法
//处理OnStart消息,启动与此Inbox相关联的Endpoint。
receivers.offer(data) // for the OnStart message
}
endpointRef
}
2.2 Dispatcher的调度原理
private val threadpool: ThreadPoolExecutor = {
//获取可用处理器数,numUsableCores是NettyRpcEnv的入参,如果大于0则等于numUsableCores,否则为当前系统可用处理器
val availableCores =
if (numUsableCores > 0) numUsableCores else Runtime.getRuntime.availableProcessors()
//获取当前线程池的大小,默认为2和可用处理器之间的最大值,可用spark.rpc.netty.dispatcher.numThreads属性配置
val numThreads = nettyEnv.conf.getInt("spark.rpc.netty.dispatcher.numThreads",
math.max(2, availableCores))
//创建线程池
val pool = ThreadUtils.newDaemonFixedThreadPool(numThreads, "dispatcher-event-loop")
//启动多个线程运行MessageLoop任务
for (i <- 0 until numThreads) {
pool.execute(new MessageLoop)
}
//返回线程池引用
pool
}
/** Message loop used for dispatching messages. */
private class MessageLoop extends Runnable {
override def run(): Unit = {
try {
while (true) {
try {
//在阻塞队列中获取EndpointData
val data = receivers.take()
//如果EndpointData是空数据,则将它重新放回队列并直接返回,这样可以让其他MessageLoop获取到这个空EndpointData并结束线程 //private val PoisonPill = new EndpointData(null,null,null)
if (data == PoisonPill) {
// Put PoisonPill back so that other MessageLoops can see it.
receivers.offer(PoisonPill)
return
}
//调用inbox的process方法对消息进行处理
data.inbox.process(Dispatcher.this)
} catch {
case NonFatal(e) => logError(e.getMessage, e)
}
}
} catch {
case ie: InterruptedException => // exit
}
}
}
Inbox的process方法:
def process(dispatcher: Dispatcher): Unit = {
var message: InboxMessage = null
//线程并发检查,如果不允许多线程执行且当前激活线程不为0,直接返回
inbox.synchronized {
if (!enableConcurrent && numActiveThreads != 0) {
return
}
//获取消息,如果消息不为空,则当前激活线程+1,否则return返回
message = messages.poll()
if (message != null) {
numActiveThreads += 1
} else {
return
}
}
while (true) {
//对匹配执行时可能发生的错误,使用Endpoint的onError方法处理
safelyCall(endpoint) {
//匹配不同类型的消息进行处理
message match{...}
}
//对激活进程数量的控制,如果不允许多线程处理且当前激活进程不为1,当前线程退出,numActiveThreads - 1
//如果message为空,没有消息需要处理,当前线程退出,numActiveThreads - 1
inbox.synchronized {
// "enableConcurrent" will be set to false after `onStop` is called, so we should check it
// every time.
if (!enableConcurrent && numActiveThreads != 1) {
// If we are not the only one worker, exit
numActiveThreads -= 1
return
}
message = messages.poll()
if (message == null) {
numActiveThreads -= 1
return
}
}
}
}
2.3 Dispatcher对RpcEndpoint去注册
def stop(rpcEndpointRef: RpcEndpointRef): Unit = {
synchronized {
if (stopped) {
// This endpoint will be stopped by Dispatcher.stop() method.
return
}
unregisterRpcEndpoint(rpcEndpointRef.name)
}
}
private def unregisterRpcEndpoint(name: String): Unit = {
//取出EndpointData
val data = endpoints.remove(name)
if (data != null) {
//调用Inbox的stop方法
data.inbox.stop()
//将EndpointData重新放入receivers队列,让MessageLoop线程能读取到Stop状态,进行相应的处理
receivers.offer(data) // for the OnStop message
}
// Don't clean `endpointRefs` here because it's possible that some messages are being processed
// now and they can use `getRpcEndpointRef`. So `endpointRefs` will be cleaned in Inbox via
// `removeRpcEndpointRef`.
}
/*
* 当要移除一个EndpointData时,其Inbox可能正在对消息进行处理,所以调用Inbox的stop方法平滑过渡处理;
* 将允许并发运行设置为false,并设置当前Inbox为stopped状态,将当前Inbox所属的EndpointData重新放入receivers,
* Inbox.process方法会匹配执行相应的处理,调用Dispatcher.removeRpcEndpointRef方法从endpointRefs缓存中移除当前RpcEndpointRef的映射; * 在匹配执行OnStop消息的最后,会调用RpcEndpoint的OnStop方法停止RpcEndpoint。
*/
def stop(): Unit = inbox.synchronized {
// The following codes should be in `synchronized` so that we can make sure "OnStop" is the last
// message
if (!stopped) {
// We should disable concurrent here. Then when RpcEndpoint.onStop is called, it's the only
// thread that is processing messages. So `RpcEndpoint.onStop` can release its resources
// safely.
enableConcurrent = false
stopped = true
messages.add(OnStop)
// Note: The concurrent events in messages will be processed one by one.
}
}
Dispatcher.stop()方法用来停止Dispatcher,之前的stop(rpcEndpointRef:RpcEndpointRef)用于对RpcEndpoint的去注册。
def stop(): Unit = {
synchronized {
if (stopped) {
return
}
stopped = true
}
// Stop all endpoints. This will queue all endpoints for processing by the message loops.
//调用unregisterRpcEndpoint方法,对Dispatcher中的所有EndpointData进行去注册,会向endpoints中每个EndpointData中的Inbox中放置
//OnStop消息;最后向receivers中投放PoisonPill,即空EndpointData,以停止所有的MessageLoop线程
endpoints.keySet().asScala.foreach(unregisterRpcEndpoint)
// Enqueue a message that tells the message loops to stop.
receivers.offer(PoisonPill)
threadpool.shutdown()
}
2.4 Dispatcher提交消息
/**
* 将消息提交给指定的RpcEndpoint
* @param endpointName endpoint名称
* @param message 消息类型
* @param callbackIfStopped endpoint为stop状态时的回调函数
*/
private def postMessage(
endpointName: String,
message: InboxMessage,
callbackIfStopped: (Exception) => Unit): Unit = {
val error = synchronized {
//从endpoints缓存获取EndpointData
val data = endpoints.get(endpointName)
if (stopped) {
Some(new RpcEnvStoppedException())
} else if (data == null) {
Some(new SparkException(s"Could not find $endpointName."))
} else {
//如果endpointData不是停止状态且endpoints缓存中确实有这个EndpointData
//调用对应的Inbox.post将消息加入Inbox的消息列表中
data.inbox.post(message)
//将EndpointData加入receivers队列,以便MessageLoop线程处理此Inbox中的消息
receivers.offer(data)
None
}
}
// We don't need to call `onStop` in the `synchronized` block
error.foreach(callbackIfStopped)
}
//在Inbox未停止时,将message加入messages缓存
def post(message: InboxMessage): Unit = inbox.synchronized {
if (stopped) {
// We already put "OnStop" into "messages", so we should drop further messages
onDrop(message)
} else {
messages.add(message)
false
}
}
3 NettyStreamManager
基于ConcurrentHashMap提供NettyRpcEnv的文件流服务,支持普通文件、jar文件及目录的添加缓存和文件流读取,各个Excutor节点可以使用Driver端提供的NettyStreamManager从Driver端下载jar包或文件支持任务的运行。
4 TransportContext
TransportContext内部包含TransportConf和RpcHandler,封装了用于创建TransportClientFactory和TransportServer的上下文信息;TransportClientFactory是创建TransportClient的工厂类,用于创建RPC框架的客户端,transportServer是RPC框架的服务端。
private val transportContext = new TransportContext(transportConf,
new NettyRpcHandler(dispatcher, this, streamManager))
创建TransportContext需要两个参数:transportConf和NettyRpcHandler,主要看一下NettyRpcHandler。
private[netty] class NettyRpcHandler(
dispatcher: Dispatcher,
nettyEnv: NettyRpcEnv,
streamManager: StreamManager) extends RpcHandler with Logging {
// A variable to track the remote RpcEnv addresses of all clients
private val remoteAddresses = new ConcurrentHashMap[RpcAddress, RpcAddress]()
//带回调函数的receive方法,调用internalReceive方法将将ByteBuffer类型的消息转化为RequestMessage
//最后调用dispatcher.postRemoteMessage将消息投递到Inbox,由RpcEndpoint处理消息并回复客户端
override def receive(
client: TransportClient,
message: ByteBuffer,
callback: RpcResponseCallback): Unit = {
val messageToDispatch = internalReceive(client, message)
dispatcher.postRemoteMessage(messageToDispatch, callback)
}
//方法重载,RpcEndpoint处理完消息不会回复客户端
override def receive(
client: TransportClient,
message: ByteBuffer): Unit = {
val messageToDispatch = internalReceive(client, message)
dispatcher.postOneWayMessage(messageToDispatch)
}
//将ByteBuffer类型的消息转化为RequestMessage
private def internalReceive(client: TransportClient, message: ByteBuffer): RequestMessage = {
//从TransportClient中获取远端地址RpcAddress
val addr = client.getChannel().remoteAddress().asInstanceOf[InetSocketAddress]
assert(addr != null)
val clientAddr = RpcAddress(addr.getHostString, addr.getPort)
//封装消息
val requestMessage = RequestMessage(nettyEnv, client, message)
//如果没有发送者地址信息,使用从TransportClient获取的远端地址RpcAddress、消息的接收者(RpcEndpoint)、消息内容构造新的消息
if (requestMessage.senderAddress == null) {
// Create a new message with the socket address of the client as the sender.
new RequestMessage(clientAddr, requestMessage.receiver, requestMessage.content)
} else {
// The remote RpcEnv listens to some port, we should also fire a RemoteProcessConnected for
// the listening address
//获取发送者地址信息,将远端地址RpcAddress和发送者地址信息映射关系放入缓存remoteAddresses
val remoteEnvAddress = requestMessage.senderAddress
if (remoteAddresses.putIfAbsent(clientAddr, remoteEnvAddress) == null) {
//向endpoints缓存中的所有EndpointData的Inbox中放入RemoteProcessConnected类型的消息
dispatcher.postToAll(RemoteProcessConnected(remoteEnvAddress))
}
requestMessage
}
}
... //其他类型消息的处理,与receive类似
}
5 客户端发送请求
//用于处理请求超时的调度器
val timeoutScheduler = ThreadUtils.newDaemonSingleThreadScheduledExecutor("netty-rpc-env-timeout")
//用于异步处理客户端创建的线程池
private[netty] val clientConnectionExecutor = ThreadUtils.newDaemonCachedThreadPool(
"netty-rpc-connection",
conf.getInt("spark.rpc.connect.threads", 64))
/**
* 缓存远端RPC地址与OutBox的关系
* OutBox与之前的Inbox类似,Outbox是在客户端使用,通过OutboxMessage封装对外发送的消息
* Inbox在服务端使用,通过InboxMessage封装接收的消息。
* outbox内部有messgaes列表存放消息,通过drainOutbox方法循环取出消息并调用sendWith方法处理
*
*/
private val outboxes = new ConcurrentHashMap[RpcAddress, Outbox]()
篇幅原因到此为止,很多东西还停留在代码层面,有点云里雾里,后面研究其他组件的时候有机会再重读RPC环境的代码吧==!
请求的发送与接收处理流程:
1、通过NettyRpcEndpointRef的send/ask方法向远端节点的RpcEndpoint发送消息,消息会先被封装为OutboxMessage,然后放入远端RpcEndpoint的地址对应的Outbox的messages列表中。
2、Outbox的drainOutbox方法不断从messages列表取出OutboxMessage,并使用内部的TransportClient向远端NettyRpcEnv发送OutboxMessage。
3、发送的请求与在远端RpcEndpoint的TransportServer建立连接,请求先经过RPC管道的处理后由NettyRpcHandler处理,NettyRpcHandler的receive方法会调用Dispatcher的post...方法将消息放入EndpointData内部的Inbox的messges中,最后MessageLoop线程会读取消息并将消息发送给对应的RpcEndpoint处理。