spark源码阅读之network(2)
在上节的解读中发现spark的源码中大量使用netty的buffer部分的api,该节将看到netty核心的一些api,比如channel:
在Netty里,Channel是通讯的载体(网络套接字或组件的连接),而ChannelHandler负责Channel中的逻辑处理,channel支持读,写,绑定本地端口,连接远程等,Channel中所有的操作都是异步的,当发生io操作的时候将会返回一个ChannelFutrue的接口,在ChannelFutrue里面可以处理操作成功、失败、取消后的动作。有了这些理解就可以看client部分的源码了
TransportClientFactory是一个创建TransportClient的工厂类,该类为每一个网络地址都提供了一个连接池,相同的主机返回相同的TransportClient。所有的TransportClient都共享一个EventLoopGroup,该类用于处理channel的上的事件的。
privatestaticclassClientPool{
TransportClient[] clients;
Object[] locks;
publicClientPool(int size){
clients =newTransportClient[size];
locks =newObject[size];
for(int i =0; i < size; i++){
locks[i]=newObject();
}
}
ClientPool表示一个连接池,每一个地址对应一个连接池。连接池的大小spark.shuffle.io.numConnectionsPerPeer来指定。该连接池怎么使用呢。用户传递一个地址进来,该地址作为key到connectionPool中查找该地址对应的连接池,没有就生成一个,获取连接池后需要随机的获取一个连接,这个时候连接池中锁就用到了。
publicTransportClient createClient(String remoteHost,int remotePort)throwsIOException{
// Get connection from the connection pool first.
// If it is not found or not active, create a new one.
finalInetSocketAddress address =newInetSocketAddress(remoteHost, remotePort);
// Create the ClientPool if we don't have it yet.
ClientPool clientPool = connectionPool.get(address);
if(clientPool ==null){
connectionPool.putIfAbsent(address,newClientPool(numConnectionsPerPeer));
clientPool = connectionPool.get(address);
}
int clientIndex = rand.nextInt(numConnectionsPerPeer);
TransportClient cachedClient = clientPool.clients[clientIndex];
if(cachedClient !=null&& cachedClient.isActive()){
logger.trace("Returning cached connection to {}: {}", address, cachedClient);
return cachedClient;
}
// If we reach here, we don't have an existing connection open. Let's create a new one.
// Multiple threads might race here to create new connections. Keep only one of them active.
synchronized(clientPool.locks[clientIndex]){
cachedClient = clientPool.clients[clientIndex];
if(cachedClient !=null){
if(cachedClient.isActive()){
logger.trace("Returning cached connection to {}: {}", address, cachedClient);
return cachedClient;
}else{
logger.info("Found inactive connection to {}, creating a new one.", address);
}
}
clientPool.clients[clientIndex]= createClient(address);
return clientPool.clients[clientIndex];
}
}
怎么创建一个全新的TransportClient,这块要使用netty的bootstrap类帮忙了,主要bootstrap的配置,缓存分配器使用缓存池来管理。调用bootstrap的handler函数给bootstrap添加了一个ChannelHandler,当bootstrap连接成功后回调该ChannelHandler,在ChannelHandler的initchannel的监听方法里面获取了连接通道SocketChannel,使用TransportContext的initializePipeline来初始化通道,就是给通道添加监听器。在这个方法里面我们拿到了
TransportClient和channel。
为什么要使用AtomicReference来保存他们的引用呢?
说说我的理解:内部类只能使用外部类的final变量,局部final变量必须声明的时候初始化,如果不使用AtomicReference就无法保持内部类的一些对象的引用。
*/
publicTransportClient createUnmanagedClient(String remoteHost,int remotePort)
throwsIOException{
finalInetSocketAddress address =newInetSocketAddress(remoteHost, remotePort);
return createClient(address);
}
/** Create a completely new {@link TransportClient} to the remote address. */
privateTransportClient createClient(InetSocketAddress address)throwsIOException{
logger.debug("Creating new connection to "+ address);
Bootstrap bootstrap =newBootstrap();
bootstrap.group(workerGroup)
.channel(socketChannelClass)
// Disable Nagle's Algorithm since we don't want packets to wait
.option(ChannelOption.TCP_NODELAY,true)
.option(ChannelOption.SO_KEEPALIVE,true)
.option(ChannelOption.CONNECT_TIMEOUT_MILLIS, conf.connectionTimeoutMs())
.option(ChannelOption.ALLOCATOR, pooledAllocator);
finalAtomicReference<TransportClient> clientRef =newAtomicReference<TransportClient>();
finalAtomicReference<Channel> channelRef =newAtomicReference<Channel>();
bootstrap.handler(newChannelInitializer<SocketChannel>(){
@Override
publicvoid initChannel(SocketChannel ch){
TransportChannelHandler clientHandler = context.initializePipeline(ch);
clientRef.set(clientHandler.getClient());
channelRef.set(ch);
}
});
// Connect to the remote server
long preConnect =System.nanoTime();
ChannelFuture cf = bootstrap.connect(address);
if(!cf.awaitUninterruptibly(conf.connectionTimeoutMs())){
thrownewIOException(
String.format("Connecting to %s timed out (%s ms)", address, conf.connectionTimeoutMs()));
}elseif(cf.cause()!=null){
thrownewIOException(String.format("Failed to connect to %s", address), cf.cause());
}
TransportClient client = clientRef.get();
Channel channel = channelRef.get();
assert client !=null:"Channel future completed successfully with null client";
// Execute any client bootstraps synchronously before marking the Client as successful.
long preBootstrap =System.nanoTime();
logger.debug("Connection to {} successful, running bootstraps...", address);
try{
for(TransportClientBootstrap clientBootstrap : clientBootstraps){
clientBootstrap.doBootstrap(client, channel);
}
}catch(Exception e){// catch non-RuntimeExceptions too as bootstrap may be written in Scala
long bootstrapTimeMs =(System.nanoTime()- preBootstrap)/1000000;
logger.error("Exception while bootstrapping client after "+ bootstrapTimeMs +" ms", e);
client.close();
throwThrowables.propagate(e);
}
long postBootstrap =System.nanoTime();
logger.debug("Successfully created connection to {} after {} ms ({} ms spent in bootstraps)",
address,(postBootstrap - preConnect)/1000000,(postBootstrap - preBootstrap)/1000000);
return client;
}
下面看下TransportClient,该类有两个作用:获取数据和发送请求,获取数据用来获取预先协议好的数据流,把数据打散成块(大小为KB和MB)便于传输,当TransportClient要从流上获取数据时,流相关的配置不是TCP的传输层做的,而是需要调用TransportClient的sendRPC执行一些配置。具体流程如下:
client.sendRPC(new OpenFile("/foo") 返回一个StreamId = 10
client.fetchChunk(streamId=100,chunkIndex= 0,callback)
client.fetchChunk(streamId=100,chunkIndex= 1,callback)
client.sendRPC(new CloseStream(100))
一个TransportClient可以使用在多个Streams上,但是一个streams只能和一个client绑定,以免响应顺序错乱。
一个client有3个成员变量:channel用于写操作,向服务器端发送请求,TransportResponseHandler用于处理服务器端响应,clientId给client编号。
privatefinalChannel channel;
privatefinalTransportResponseHandler handler;
@NullableprivateString clientId;
client有3个请求函数,一个是请求数据流中的一个数据块,用于数据传输,第二个是请求整个数据流,用于数据传输,第三个是发送控制请求。有点像ftp,一个用于控制,一个用于数据。
publicvoid fetchChunk(
long streamId,
finalint chunkIndex,
finalChunkReceivedCallback callback){
finalString serverAddr =NettyUtils.getRemoteAddress(channel);
finallong startTime =System.currentTimeMillis();
logger.debug("Sending fetch chunk request {} to {}", chunkIndex, serverAddr);
finalStreamChunkId streamChunkId =newStreamChunkId(streamId, chunkIndex);
handler.addFetchRequest(streamChunkId, callback);
channel.writeAndFlush(newChunkFetchRequest(streamChunkId)).addListener(
newChannelFutureListener(){
@Override
publicvoid operationComplete(ChannelFuture future)throwsException{
if(future.isSuccess()){
long timeTaken =System.currentTimeMillis()- startTime;
logger.trace("Sending request {} to {} took {} ms", streamChunkId, serverAddr,
timeTaken);
}else{
String errorMsg =String.format("Failed to send request %s to %s: %s", streamChunkId,
serverAddr, future.cause());
logger.error(errorMsg, future.cause());
handler.removeFetchRequest(streamChunkId);
channel.close();
try{
callback.onFailure(chunkIndex,newIOException(errorMsg, future.cause()));
}catch(Exception e){
logger.error("Uncaught exception in RPC response callback handler!", e);
}
}
}
});
}
callback有两个方法,这里要说明下他们的回调机制,onFailure在channel的IO操作失败后调用,就是ChannelFuture失败时候调用,ChannelFuture是IO操作的结果。
onSuccess调用时在channel的事件处理流程中使用,context.initializePipeline(ch)给channel注册了一个TransportChannelHandler,TransportChannelHandler包含了TransportResponseHandler对象,它把响应结果转发给TransportResponseHandler用于处理服务器端响应,handler.addFetchRequest(streamChunkId, callback)映射每个响应对应的回调接口。在对应响应到来时调用对应回调接口。
channel.writeAndFlush的对象需要实现Encodable接口。该接口的一些方法被MessageDecoder和MessageEncoder调用
sendRpc和上面方法一样,这里就不描述了,看下stream方法
publicvoid stream(finalString streamId,finalStreamCallback callback){
finalString serverAddr =NettyUtils.getRemoteAddress(channel);
finallong startTime =System.currentTimeMillis();
logger.debug("Sending stream request for {} to {}", streamId, serverAddr);
// Need to synchronize here so that the callback is added to the queue and the RPC is
// written to the socket atomically, so that callbacks are called in the right order
// when responses arrive.
synchronized(this){
handler.addStreamCallback(callback);
channel.writeAndFlush(newStreamRequest(streamId)).addListener(
newChannelFutureListener(){
@Override
publicvoid operationComplete(ChannelFuture future)throwsException{
if(future.isSuccess()){
long timeTaken =System.currentTimeMillis()- startTime;
logger.trace("Sending request for {} to {} took {} ms", streamId, serverAddr,
timeTaken);
}else{
String errorMsg =String.format("Failed to send request for %s to %s: %s", streamId,
serverAddr, future.cause());
logger.error(errorMsg, future.cause());
channel.close();
try{
callback.onFailure(streamId,newIOException(errorMsg, future.cause()));
}catch(Exception e){
logger.error("Uncaught exception in RPC response callback handler!", e);
}
}
}
});
}
}
这里加了一个同步块,这样保证回调接口调用和请求的顺序一样。这里一个不明白的地方就是,同时发两个请求,第二个请求可能比第一个请求更快返回。怎么保证顺序一致呢?
sendRpcSysnc是一个非常有意思的方法,这里学习了Future怎么使用了。
publicbyte[] sendRpcSync(byte[] message,long timeoutMs){
finalSettableFuture<byte[]> result =SettableFuture.create();
sendRpc(message,newRpcResponseCallback(){
@Override
publicvoid onSuccess(byte[] response){
result.set(response);
}
@Override
publicvoid onFailure(Throwable e){
result.setException(e);
}
});
try{
return result.get(timeoutMs,TimeUnit.MILLISECONDS);
}catch(ExecutionException e){
throwThrowables.propagate(e.getCause());
}catch(Exception e){
throwThrowables.propagate(e);
}
}
匿名内部类智能使用外部类的final,要异步获取内部类的数据使用了一个SettableFuture。