Spark - Inbox & InboxMessage & Dispatcher&RequestMessage&Outbox&OutboxMessage源码解析
Spark-Inbox & InboxMessage&Outbox&OutboxMessage 源码解析
- trait InboxMessage
-
- case class OneWayMessage
- case class RpcMessage
- case object OnStart
- case object OnStop
- case class RemoteProcessConnected
- case class RemoteProcessDisconnected
- case class RemoteProcessConnectionError
- Inbox class
- EndpointData
- Dispatcher class
- NettyRpcHandler class
- RequestMessage class & object
-
- RequestMessage object
- RequestMessage class
- 一个Rpc消息的处理的流程解读
- trait OutboxMessage
-
- OneWayOutboxMessage
- RpcOutboxMessage
- Outbox Class
trait InboxMessage
这个接口是各种消息的统一接口，用来封装各种类型的消息。比如Rpc消息，start消息，stop消息，remote消息，方便在 Inbox Process 方法执行的时候，进行识别是 那类消息。
case class OneWayMessage
OneWay消息
case class RpcMessage
Rpc消息
case object OnStart
启动消息
case object OnStop
stop消息
case class RemoteProcessConnected
远程连接消息
case class RemoteProcessDisconnected
远程断开消息
case class RemoteProcessConnectionError
远程连接Error消息
Inbox class
这个就是 spark的收件箱，上面的各种类型的消息先存在在这个类的message这个LinkedList中。
属性：
protected val messages = new java.util.LinkedList[InboxMessage]() //保存message消息
private var stopped = false
private var enableConcurrent = false //是否运行多线程 同时处理Inbox message
private var numActiveThreads = 0 //多少个线程正在处理这个Inbox message
构造方法：
private[netty] class Inbox(
val endpointRef: NettyRpcEndpointRef,
val endpoint: RpcEndpoint)
extends Logging{
//启动的时候 在messages中加入 OnStart 的Message
inbox.synchronized {
messages.add(OnStart)
}
}
方法：
//message LinkedList 是否为 null
def isEmpty: Boolean = inbox.synchronized {
messages.isEmpty }
//丢弃消息
protected def onDrop(message: InboxMessage): Unit = {
logWarning(s"Drop $message because $endpointRef is stopped")
}
//停止 执行 case object OnStop message，属性参数复原
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.
}
}
//message LinkedList 投递消息，如果已经stop 则丢弃消息
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
}
}
//处理消息的方法，根据消息类型分别处理消息，尽可能多的一次处理所有保存的消息
//这个传进来的dispatcher变量，只有在OnStop 消息的时候，才会使用到，从dispatcher的属性中 通过endPoint remove endPointRef
def process(dispatcher: Dispatcher): Unit = {
var message: InboxMessage = null
//没有处理OnStart消息的时候始终只能有一个线程处理消息
//当处理了OnStart消息的时候且这个endPoint是ThreadSafeRpcEndpoint的话，就可以多个线程处理消息了
inbox.synchronized {
if (!enableConcurrent && numActiveThreads != 0) {
return
}
message = messages.poll()
if (message != null) {
numActiveThreads += 1
} else {
return
}
}
while (true) {
safelyCall(endpoint) {
message match {
//RpcMessage 需要等待replay
case RpcMessage(_sender, content, context) =>
try {
endpoint.receiveAndReply(context).applyOrElse[Any, Unit](content, {
msg =>
throw new SparkException(s"Unsupported message $message from ${_sender}")
})
} catch {
case e: Throwable =>
context.sendFailure(e)
// Throw the exception -- this exception will be caught by the safelyCall function.
// The endpoint's onError function will be called.
throw e
}
case OneWayMessage(_sender, content) =>
endpoint.receive.applyOrElse[Any, Unit](content, {
msg =>
throw new SparkException(s"Unsupported message $message from ${_sender}")
})
case OnStart =>
endpoint.onStart()
if (!endpoint.isInstanceOf[ThreadSafeRpcEndpoint]) {
inbox.synchronized {
if (!stopped) {
enableConcurrent = true
}
}
}
case OnStop =>
val activeThreads = inbox.synchronized {
inbox.numActiveThreads }
assert(activeThreads == 1,
s"There should be only a single active thread but found $activeThreads threads.")
dispatcher.removeRpcEndpointRef(endpoint)//stop的时候dispatcher remove这个endPoint
endpoint.onStop()
assert(isEmpty, "OnStop should be the last message")
case RemoteProcessConnected(remoteAddress) =>
endpoint.onConnected(remoteAddress)
case RemoteProcessDisconnected(remoteAddress) =>
endpoint.onDisconnected(remoteAddress)
case RemoteProcessConnectionError(cause, remoteAddress) =>
endpoint.onNetworkError(cause, remoteAddress)
}
}
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
}
}
}
}
EndpointData
这个是一个内部类，一个有一个属性就是 上面的Inbox，即每一个EndpointData的实例都会有一个Inbox的属性。
private class EndpointData(
val name: String,
val endpoint: RpcEndpoint,
val ref: NettyRpcEndpointRef) {
val inbox = new Inbox(ref, endpoint)
}
Dispatcher class
这个类负责保存已经注册的RpcEndpoint和投递消息到对应的EndpointData的Inbox中，所以其内部属性有2个ConcurrentHashMap用来保存
EndpointName-》EndpointData，EndpointData-》RpcEndpointRef这2个 实例。
属性：
//保存已经注册的endPointName和EndpointData，EndpointData是一个内部类
//可以得出 一个唯一name的 RpcEndpoint 会有唯一的 EndpointData，也会有唯一的 Inbox
private val endpoints: ConcurrentMap[String, EndpointData] =
new ConcurrentHashMap[String, EndpointData]
//保存RpcEndpoint-》RpcEndpointRef键值对
private val endpointRefs: ConcurrentMap[RpcEndpoint, RpcEndpointRef] =
new ConcurrentHashMap[RpcEndpoint, RpcEndpointRef]
//保存收到的EndpointData信息到 LinkedBlockingQueue，先从endpoints 通过name标识 get
//出 EndpointData ，再把真正的message post到 其内部（Inbox）的发件箱的消息队列属性messages中，等待处理。
//待处理的消息队列
private val receivers = new LinkedBlockingQueue[EndpointData]
//标示 运行状态
private var stopped = false
//这里会根据配置的 CPU核数和配置的核数 算出核数，启动线程池，来处理 为处理的消息队列。
//这些 线程的名称以 dispatcher-event-loop 开头
private val threadpool: ThreadPoolExecutor = {
}
方法：
//注册一个别称为name的endPoint
def registerRpcEndpoint(name: String, endpoint: RpcEndpoint): NettyRpcEndpointRef = {
val addr = RpcEndpointAddress(nettyEnv.address, name)//创建一个RpcEndpointAddress地址
val endpointRef = new NettyRpcEndpointRef(nettyEnv.conf, addr, nettyEnv)//创建一个对应的Ref NettyRpcEndpointRef，使用的是 dispather内部的nettyRpcEnv 的 host 和 port
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")
}
val data = endpoints.get(name)
endpointRefs.put(data.endpoint, data.ref)
receivers.offer(data) // for the OnStart message
}
endpointRef
}
//通过endpoint获取对应的endpointRef
def getRpcEndpointRef(endpoint: RpcEndpoint): RpcEndpointRef = endpointRefs.get(endpoint)
//通过endpoint移除对应的endpointRef
def removeRpcEndpointRef(endpoint: RpcEndpoint): Unit = endpointRefs.remove(endpoint)
//通过name注销rpcEndpoint
private def unregisterRpcEndpoint(name: String): Unit = {
val data = endpoints.remove(name)
if (data != null) {
data.inbox.stop()
receivers.offer(data) // for the OnStop message
}
}
//通过rpcEndpointRef stop对应的rpcEndpoint
def stop(rpcEndpointRef: RpcEndpointRef): Unit = {
synchronized {
if (stopped) {
// This endpoint will be stopped by Dispatcher.stop() method.
return
}
unregisterRpcEndpoint(rpcEndpointRef.name)
}
}
//投递消息到对应的 name 的 endpointData 的 Inbox中，再发送endpointData到 待处理的消息队列中
private def postMessage(
endpointName: String,
message: InboxMessage,
callbackIfStopped: (Exception) => Unit): Unit = {
val error = synchronized {
val data = endpoints.get(endpointName)
if (stopped) {
Some(new RpcEnvStoppedException())
} else if (data == null) {
Some(new SparkException(s"Could not find $endpointName."))
} else {
data.inbox.post(message)
receivers.offer(data)
None
}
}
// We don't need to call `onStop` in the `synchronized` block
error.foreach(callbackIfStopped)
}
//验证是否存在这个name的endPoint
def verify(name: String): Boolean = {
endpoints.containsKey(name)
}
NettyRpcHandler class
这个类是处理 socket 消息处理类。
里面有2个receiver方法，用来接受消息；channelActive 用来是发送消息。
这2个receive方法是重载的，区别在于一个需要返回响应，即会要回调。
反序列化数据流，拿到 发送方地址，接收方地址，RpcEndpoint name 等信息，包装成RequestMessage的实例， 最后投递到 NettyRpcHandler class 属性 dispatcher 中的属性中的对应的 Inbox中（参考 Dispatcher class）
receiver：
//会要回调 所以是 postRemoteMessage
override def receive(
client: TransportClient,
message: ByteBuffer,
callback: RpcResponseCallback): Unit = {
val messageToDispatch = internalReceive(client, message)
dispatcher.postRemoteMessage(messageToDispatch, callback)
}
//不需要 回调 所以是 postOneWayMessage
override def receive(
client: TransportClient,
message: ByteBuffer): Unit = {
val messageToDispatch = internalReceive(client, message)
dispatcher.postOneWayMessage(messageToDispatch)
}
其中有一个internalReceive的方法，反序列化数据流，拿到 发送方地址，接收方地址，RpcEndpoint name 等信息，包装成RequestMessage的实例。
private def internalReceive(client: TransportClient, message: ByteBuffer): RequestMessage = {
val addr = client.getChannel().remoteAddress().asInstanceOf[InetSocketAddress]
assert(addr != null)
val clientAddr = RpcAddress(addr.getHostString, addr.getPort)
//这里完成后 已经是 反序列化后的 数据了
val requestMessage = RequestMessage(nettyEnv, client, message)
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
val remoteEnvAddress = requestMessage.senderAddress
if (remoteAddresses.putIfAbsent(clientAddr, remoteEnvAddress) == null) {
dispatcher.postToAll(RemoteProcessConnected(remoteEnvAddress))
}
requestMessage
}
}
RequestMessage class & object
RequestMessage object
这个object中一共有2个方法：
//这个方法用来读取 数据流中的 host 和 port 的地址信息
private def readRpcAddress(in: DataInputStream): RpcAddress = {
val hasRpcAddress = in.readBoolean()
if (hasRpcAddress) {
RpcAddress(in.readUTF(), in.readInt())
} else {
null
}
}
//用来处理 socket 接收到的 数据流
def apply(nettyEnv: NettyRpcEnv, client: TransportClient, bytes: ByteBuffer): RequestMessage = {
val bis = new ByteBufferInputStream(bytes)
val in = new DataInputStream(bis)
try {
//获取流头部的 host和port 信息
val senderAddress = readRpcAddress(in) //这个地址是发送这的地址
//这里的地址是 接受者的地址，即接收到这个信息的机器的地址
//后面的in.readUTF() 是获取 这个endPoint 的name
//最后组成RpcEndpointAddress
val endpointAddress = RpcEndpointAddress(readRpcAddress(in), in.readUTF())
//在组成一个 NettyRpcEndpointRef
val ref = new NettyRpcEndpointRef(nettyEnv.conf, endpointAddress, nettyEnv)
ref.client = client
//返回新的 RequestMessage
new RequestMessage(
senderAddress,
ref,
// The remaining bytes in `bytes` are the message content.
nettyEnv.deserialize(client, bytes) //这里是反序列化后的对象
)
} finally {
in.close()
}
}
RequestMessage class
这个类就是包装了消息发送放的RpcAddress，接收方的NettyRpcEndpointRef，还有接收的已经反序列化的内容。所以它的构造方法是下面的样子。
class RequestMessage(
val senderAddress: RpcAddress,
val receiver: NettyRpcEndpointRef,
val content: Any)
方法：
//序列化 发送方，接收方的地址信息 和 方法内容到 数据流
//这个方法是在 发送信息的时候使用
def serialize(nettyEnv: NettyRpcEnv): ByteBuffer = {
val bos = new ByteBufferOutputStream()
val out = new DataOutputStream(bos)
try {
//当然发送消息的时候，这里 的 senderAddress 就是本机的地址了
writeRpcAddress(out, senderAddress)
writeRpcAddress(out, receiver.address)
out.writeUTF(receiver.name)
val s = nettyEnv.serializeStream(out)
try {
s.writeObject(content)
} finally {
s.close()
}
} finally {
out.close()
}
bos.toByteBuffer
}
//这个是序列化 rpcAddress 信息到 数据流
private def writeRpcAddress(out: DataOutputStream, rpcAddress: RpcAddress): Unit = {
if (rpcAddress == null) {
out.writeBoolean(false)
} else {
out.writeBoolean(true)
out.writeUTF(rpcAddress.host)
out.writeInt(rpcAddress.port)
}
}
一个Rpc消息的处理的流程解读
当NettyRpcEnv class 的startServer 方法启动 socket 服务之后。当接受到客户端的一个请求的时候，为委托给NettyRpcHandler class 处理（这一部分的只是 可以学习 netty 网络框架）。
NettyRpcHandler class 里面会 dispacher 和 nettyRpcEnv、streamManager。
dispacher 用来分发（已经反序列化后）消息到对应的endPoint 的Inbox；
nettyRpcEnv 主要是要使用其 反序列化方法，来序列化信息。
经过NettyRpcHandler receive 和 internalReceiver 处理之后，就会返回一个新的 RequestMessage.
最后把这个消息投递到对应的endPoint的Ibox中，再由dispacher 的线程池 调度执行 Inbox的process的方法。
trait OutboxMessage
这个是Outbox 消息的总接口，一共有2个实现类：OneWayOutboxMessage、RpcOutboxMessage。一个是发送即忘记，另一个是需要回调的。
注意，如果消息传递到OutboxMessage的时候，已经是序列化后的消息了。
一共有2个方法:
//发送方法 TransportClient就是发送的客户端
def sendWith(client: TransportClient): Unit
//发送失败处理
def onFailure(e: Throwable): Unit
OneWayOutboxMessage
这个类就是 发送即忘记 的具体实现类。
private[netty] case class OneWayOutboxMessage(content: ByteBuffer) extends OutboxMessage
with Logging {
override def sendWith(client: TransportClient): Unit = {
client.send(content)
}
override def onFailure(e: Throwable): Unit = {
e match {
case e1: RpcEnvStoppedException => logDebug(e1.getMessage)
case e1: Throwable => logWarning(s"Failed to send one-way RPC.", e1)
}
}
}
RpcOutboxMessage
这个类就是 发送获取结果后需要回调的具体实现类，所以这里又实现了RpcResponseCallback。因为需要回调，所以在构造的时候，需要传入 _onSuccess 的函数，自己处理结果。
private[netty] case class RpcOutboxMessage(
content: ByteBuffer,
_onFailure: (Throwable) => Unit,
_onSuccess: (TransportClient, ByteBuffer) => Unit)
extends OutboxMessage with RpcResponseCallback with Logging {
private var client: TransportClient = _
private var requestId: Long = _
override def sendWith(client: TransportClient): Unit = {
this.client = client
this.requestId = client.sendRpc(content, this)
}
def onTimeout(): Unit = {
if (client != null) {
client.removeRpcRequest(requestId)
} else {
logError("Ask timeout before connecting successfully")
}
}
override def onFailure(e: Throwable): Unit = {
_onFailure(e)
}
override def onSuccess(response: ByteBuffer): Unit = {
_onSuccess(client, response)
}
}
Outbox Class
Outbox是发送消息的主类，组合了 nettyRpcEnv和rpcAddress。nettyRpcEnv里面又netty的发送客户端，负责发送和接收消息结果；rpcAddress主要在创建客户端和接收消息结果时使用。
下面来看看构造方法：
//这里的address: RpcAddress 是别的主机的地址，在发送消息的时候需要 指定目标的host 和 port
private[netty] class Outbox(nettyEnv: NettyRpcEnv, val address: RpcAddress) {
//内部存储 OutboxMessage，处理OutboxMessage的时候，从这里获取
private val messages = new java.util.LinkedList[OutboxMessage]
//发送消息的客户端，第一次的时候是null，则会创建一个 client。
private var client: TransportClient = null
private var connectFuture: java.util.concurrent.Future[Unit] = null
private var stopped = false
//标示 是否有线程正在处理 消息
private var draining = false
}
下面是这个类的方法：
//这个方法 主要在nettyRpcEnv 类中调用，用来把OutboxMessage消息put到本类的messages LinkedList 中，再执行drainOutbox 方法，处理 LinkedList中的消息
//发送消息是不用 启动服务的，所以不用后台程序一直存在的
def send(message: OutboxMessage): Unit = {
val dropped = synchronized {
if (stopped) {
true
} else {
messages.add(message)
false
}
}
if (dropped) {
message.onFailure(new SparkException("Message is dropped because Outbox is stopped"))
} else {
drainOutbox()
}
}
//处理消息的主方法，尽可能一次处理多个消息
private def drainOutbox(): Unit = {
var message: OutboxMessage = null
synchronized {
if (stopped) {
return
}
if (connectFuture != null) {
// We are connecting to the remote address, so just exit
return
}
if (client == null) {
// There is no connect task but client is null, so we need to launch the connect task.
//创建一个 发送客户端,递归 调用 drainOutbox 方法，所以先、这里有个 return，防止无限递归
launchConnectTask()
return
}
if (draining) {
// There is some thread draining, so just exit
return
}//获取到一个消息
message = messages.poll()
if (message == null) {
return
}
draining = true
}
while (true) {
try {
val _client = synchronized {
client }
if (_client != null) {
//使用OutboxMessage子类的发送方法发送消息
message.sendWith(_client)
} else {
assert(stopped == true)
}
} catch {
case NonFatal(e) =>
handleNetworkFailure(e)
return
}
synchronized {
if (stopped) {
return
}
message = messages.poll()
if (message == null) {
draining = false
return
}
}
}
}