在 Netty 中创建 1 个 NioServerSocketChannel 在指定的端口监听客户端连接，这个过程主要有以下 个步骤：
- 创建 NioServerSocketChannel
- 初始化并注册 NioServerSocketChannel
- 绑定指定端口
首先列出一个简易服务端的启动代码：
1 public void start() { 2 ????EventLoopGroup bossGroup = new NioEventLoopGroup(1); 3 ????EventLoopGroup workerGroup = new NioEventLoopGroup(); 4 ????try { 5 ????????ServerBootstrap sbs = new ServerBootstrap() 6 ????????????????//添加 group 7 ????????????????.group(bossGroup, workerGroup) 8 ????????????????//指定服务端 Channel 类型 9 ????????????????.channel(NioServerSocketChannel.class)10 ????????????????//添加服务端 Channel 的 Handler11 ????????????????.handler(new HelloWorldServerHandler())12 ????????????????//添加客户端 Channel 的 Handler13 ????????????????.childHandler(new ChannelInitializer<NioSocketChannel>() {14 ????????????????????@Override15 ????????????????????protected void initChannel(NioSocketChannel ch) throws Exception {16 ????????????????????????//为后续接入的客户端 Channel 准备的字符串编解码 Handler 17 ????????????????????????ch.pipeline().addLast(new StringDecoder());18 ????????????????????????ch.pipeline().addLast(new StringEncoder());19 ????????????????????}20 ????????????????});21 ????????//监听指定的端口22 ????????ChannelFuture future = sbs.bind(port).sync();23 ????????System.out.println("Server start listen at " + port);24 ????????future.channel().closeFuture().sync();25 ????} catch (Exception e) {26 ????????bossGroup.shutdownGracefully();27 ????????workerGroup.shutdownGracefully();28 ????}29 }下面就从 ServerBootstrap 的 bind(int port)方法开始分析服务端的 NioServerSocketChannel 的创建过程。
1. 创建 NioServerSocketChannel
跟随 bind 方法的调用，最终在 AbstractBootstrap 类的 doBind()方法找到了初始化，注册和绑定方法调用：
1 private ChannelFuture doBind(final SocketAddress localAddress) { 2 ????//初始化并注册 3 ????final ChannelFuture regFuture = initAndRegister(); 4 ????final Channel channel = regFuture.channel(); 5 ????if (regFuture.cause() != null) { 6 ????????return regFuture; 7 ????} 8 ?9 ????if (regFuture.isDone()) {10 ????????// At this point we know that the registration was complete and successful.11 ????????ChannelPromise promise = channel.newPromise();12 ????????//绑定本地端口13 ????????doBind0(regFuture, channel, localAddress, promise);14 ????????return promise;15 ????} else {16 ????????//....17 ????}18 }2.
- 初始化并注册 NioServerSocketChannel
首先来看一下这个 initAndRegister()方法：
1 final ChannelFuture initAndRegister() { 2 ????Channel channel = null; 3 ????try { 4 ????????//创建 Channel 5 ????????channel = channelFactory.newChannel(); 6 ????????//初始化 Channel 7 ????????init(channel); 8 ????} catch (Throwable t) { 9 ????????//...10 ????}11 12 ????//注册13 ????ChannelFuture regFuture = config().group().register(channel);14 ????if (regFuture.cause() != null) {15 ????????if (channel.isRegistered()) {16 ????????????channel.close();17 ????????} else {18 ????????????channel.unsafe().closeForcibly();19 ????????}20 ????}21 ????//...22 }Channel 也是通过工厂类来创建的，这个工厂默认是 ReflectiveChannelFactory，是在前面启动代码中，设置服务端 Channel 类型时创建的。通过名字可以知道，是用反射的方式创建了 Channel 对象。
init()方法有两种实现，这里分析的是 ServerBootstrap 的实现：
1 @Override 2 void init(Channel channel) throws Exception { 3 ????//... option 的设置省略掉 4 ????//pipeline 的创建，默认使用的 DefaultPipeline 5 ????ChannelPipeline p = channel.pipeline(); 6 ?7 ????//... 客户端 Channel 相关配置的保存 8 ?9 ????p.addLast(new ChannelInitializer<Channel>() {10 ????????@Override11 ????????public void initChannel(Channel ch) throws Exception {12 ????????????final ChannelPipeline pipeline = ch.pipeline();13 ????????????//这里添加的是启动代码中，服务端的 Handler14 ????????????ChannelHandler handler = config.handler();15 ????????????if (handler != null) {16 ????????????????pipeline.addLast(handler);17 ????????????}18 19 ????????????// We add this handler via the EventLoop as the user may have used a ChannelInitializer as handler.20 ????????????// In this case the initChannel(...) method will only be called after this method returns. Because21 ????????????// of this we need to ensure we add our handler in a delayed fashion so all the users handler are22 ????????????// placed in front of the ServerBootstrapAcceptor.23 ????????????ch.eventLoop().execute(new Runnable() {24 ????????????????@Override25 ????????????????public void run() {26 ????????????????????//这里添加了一个 Accepter，用来处理新连接的接入27 ????????????????????pipeline.addLast(new ServerBootstrapAcceptor(28 ????????????????????????????currentChildGroup, currentChildHandler, currentChildOptions, currentChildAttrs));29 ????????????????}30 ????????????});31 ????????}32 ????});33 }初始化 Channel 这个动作，主要做了 4 件事：
- 创建 pipeline
- 为 Channel 添加用户创建的 Handler
- 添加 Accepter
- 其他属性的设置
接下来分析 Channel 的注册，需要关注的是这行代码：
1 ChannelFuture regFuture = config().group().register(channel);
config()方法获取了启动时创建的 config 对象，这个对象的 group()方法就返回了启动时传入的 bossGroup。启动代码中传入了两个 group，返回的为什么是 boosGroup 呢？查看启动代码中的 group(EventLoopGroup parentGroup, EventLoopGroup childGroup)方法，在它第一行就调用了 super.group(parentGroup)，将第一个 group 对象传给了父类 AbstractBootstrap。而此处 config 调用的 group()方法返回的正是父类中的 group。
因为这里是一个 NioEventLoopGroup 对象，所以使用的 register(channel)方法是 MultithreadEventLoopGroup 中的。
1 @Override2 public ChannelFuture register(Channel channel) {3 ????return next().register(channel);4 }查看 next()方法可以发现，最终是调用之前创建 group 时创建的 chooser 的 next()方法，该方法会返回一个 NioEventLooop 对象(EventLoop 是在这里分配的)，它的 register()方法是在父类 SingleThreadEventLoop 中实现的。最终调用了 AbstractChannel 中的注册方法。
1 @Override 2 public final void register(EventLoop eventLoop, final ChannelPromise promise) { 3 ????//... 4 ????//将前面返回的 eventLoop 保存起来 5 ????AbstractChannel.this.eventLoop = eventLoop; 6 ????//判断 eventLoop 中的 thread 是否是当前线程 7 ????//初次启动时，eventLoop 中的 thread 为 null 8 ????if (eventLoop.inEventLoop()) { 9 ????????register0(promise);10 ????} else {11 ????????try {12 ????????????//将注册任务传进去13 ????????????eventLoop.execute(new Runnable() {14 ????????????????@Override15 ????????????????public void run() {16 ????????????????????//注册17 ????????????????????register0(promise);18 ????????????????}19 ????????????});20 ????????} catch (Throwable t) {21 ????????????//...22 ????????}23 ????}24 }将注册动作封装成一个任务，然后交给 eventLoop 对象处理。
@Overridepublic void execute(Runnable task) { ???//... ???//这里通用是 false ???boolean inEventLoop = inEventLoop(); ???if (inEventLoop) { ???????addTask(task); ???} else { ???????//启动线程 ???????startThread(); ???????addTask(task);//将前面传进来的注册任务添加进队列 ???????if (isShutdown() && removeTask(task)) { ???????????reject(); ???????} ???} ???if (!addTaskWakesUp && wakesUpForTask(task)) { ???????wakeup(inEventLoop); ???}}private void startThread() { ???//判断是否需要启动线程 ???if (STATE_UPDATER.get(this) == ST_NOT_STARTED) { ???????if (STATE_UPDATER.compareAndSet(this, ST_NOT_STARTED, ST_STARTED)) { ???????????//启动线程 ???????????doStartThread(); ???????} ???}}上面代码中的 startThread()方法有个 STATE_UPDATER，它是用来更新该对象的 state 属性，是一个线程安全的操作。state 默认值为 ST_NOT_STARTED，所以第一次进入该方法，条件判断为 true，接下来进行 CAS 操作，将 state 设置为 ST_STARTED，然后调用 doStartThread()方法。当 group 中的线程都启用之后，下一次 chooser 再选中这个线程，startThread()方法中的第一个 if 的条件判断就是 false 了，不会再创建新的线程。
1 private void doStartThread() { 2 ????assert thread == null; 3 ????//这个 executor 就是构建 group 时，创建出来的 executor 4 ????executor.execute(new Runnable() { 5 ????????@Override 6 ????????public void run() { 7 ????????????thread = Thread.currentThread(); 8 ????????????if (interrupted) { 9 ????????????????thread.interrupt();10 ????????????}11 12 ????????????boolean success = false;13 ????????????updateLastExecutionTime();14 ????????????try {15 ????????????????//前面创建的是 NioEventLoop16 ????????????????SingleThreadEventExecutor.this.run();17 ????????????????success = true;18 ????????????} catch (Throwable t) {19 ????????????????logger.warn("Unexpected exception from an event executor: ", t);20 ????????????} finally {21 ????????????????for (;;) {22 ????????????????????//更新 state23 ????????????????????int oldState = STATE_UPDATER.get(SingleThreadEventExecutor.this);24 ????????????????????if (oldState >= ST_SHUTTING_DOWN || STATE_UPDATER.compareAndSet(25 ????????????????????????????SingleThreadEventExecutor.this, oldState, ST_SHUTTING_DOWN)) {26 ????????????????????????break;27 ????????????????????}28 ????????????????}29 ????????????????//...30 ????????????}31 ????????}32 ????});33 }前一篇分析 EventLoopGroup 创建时说过，会在 EventLoop 保存一个 executor 对象的引用，最终个任务就是交给这个 executor 来处理的。executor 的 execute(Runnable task) 方法会创建新线程，并执行传入的 task。接下来看一下 NioEventLoop 中的 run() 方法。
1 protected void run() { 2 ????for (;;) { 3 ????????try { 4 ????????????//计算 select 策略，当前有任务时，会进行一次 selectNow(NIO)，返回就绪的 key 个数 5 ????????????//显然 switch 中没有匹配项，直接跳出 switch 6 ????????????//无任务时，则直接返回 SelectStrategy.SELECT 7 ????????????//这里的 SelectStrategy.CONTINUE 感觉不会匹配到 8 ????????????switch (selectStrategy.calculateStrategy(selectNowSupplier, hasTasks())) { 9 ????????????????case SelectStrategy.CONTINUE:10 ????????????????????continue;11 ????????????????case SelectStrategy.SELECT:12 ????????????????????//当没有可处理的任务时，直接进行 select 操作13 ????????????????????// wakenUp.getAndSet(false) 返回的是 oldValue，由于默认值是 false14 ????????????????????// 所以第一次返回的是 false15 ????????????????????select(wakenUp.getAndSet(false));16 17 ????????????????????// ‘wakenUp.compareAndSet(false, true)‘ is always evaluated18 ????????????????????// before calling ‘selector.wakeup()‘ to reduce the wake-up19 ????????????????????// overhead. (Selector.wakeup() is an expensive operation.)20 ????????????????????//21 ????????????????????// However, there is a race condition in this approach.22 ????????????????????// The race condition is triggered when ‘wakenUp‘ is set to23 ????????????????????// true too early.24 ????????????????????//25 ????????????????????// ‘wakenUp‘ is set to true too early if:26 ????????????????????// 1) Selector is waken up between ‘wakenUp.set(false)‘ and27 ????????????????????// ???‘selector.select(...)‘. (BAD)28 ????????????????????// 2) Selector is waken up between ‘selector.select(...)‘ and29 ????????????????????// ???‘if (wakenUp.get()) { ... }‘. (OK)30 ????????????????????//31 ????????????????????// In the first case, ‘wakenUp‘ is set to true and the32 ????????????????????// following ‘selector.select(...)‘ will wake up immediately.33 ????????????????????// Until ‘wakenUp‘ is set to false again in the next round,34 ????????????????????// ‘wakenUp.compareAndSet(false, true)‘ will fail, and therefore35 ????????????????????// any attempt to wake up the Selector will fail, too, causing36 ????????????????????// the following ‘selector.select(...)‘ call to block37 ????????????????????// unnecessarily.38 ????????????????????//39 ????????????????????// To fix this problem, we wake up the selector again if wakenUp40 ????????????????????// is true immediately after selector.select(...).41 ????????????????????// It is inefficient in that it wakes up the selector for both42 ????????????????????// the first case (BAD - wake-up required) and the second case43 ????????????????????// (OK - no wake-up required).44 45 ????????????????????if (wakenUp.get()) {46 ????????????????????????selector.wakeup();47 ????????????????????}48 ????????????????default:49 ????????????????????// fallthrough50 ????????????}51 52 ????????????cancelledKeys = 0;53 ????????????needsToSelectAgain = false;54 ????????????final int ioRatio = this.ioRatio;55 ????????????//根据比例来处理 IO 事件和任务56 ????????????if (ioRatio == 100) {57 ????????????????try {58 ????????????????????//处理就绪的 key59 ????????????????????processSelectedKeys();60 ????????????????} finally {61 ????????????????????// Ensure we always run tasks.62 ????????????????????//执行任务63 ????????????????????runAllTasks();64 ????????????????}65 ????????????} else {66 ????????????????final long ioStartTime = System.nanoTime();67 ????????????????try {68 ????????????????????processSelectedKeys();69 ????????????????} finally {70 ????????????????????// Ensure we always run tasks.71 ????????????????????// 计算出处理 IO 事件的时间，然后根据比例算出执行任务的时间72 ????????????????????final long ioTime = System.nanoTime() - ioStartTime;73 ????????????????????runAllTasks(ioTime * (100 - ioRatio) / ioRatio);74 ????????????????}75 ????????????}76 ????????} catch (Throwable t) {77 ????????????handleLoopException(t);78 ????????}79 ????????// Always handle shutdown even if the loop processing threw an exception.80 ????????try {81 ????????????if (isShuttingDown()) {82 ????????????????closeAll();83 ????????????????if (confirmShutdown()) {84 ????????????????????return;85 ????????????????}86 ????????????}87 ????????} catch (Throwable t) {88 ????????????handleLoopException(t);89 ????????}90 ????}91 }run()方法主要是做 select 操作，和处理 IO 事件和任务队列中的任务，这部分内容下一篇文章再分析。从 executor 执行 execute()方法开始，由 Netyy 管理的线程就开始启动运行了。实际上此时的 NioServerSocketChannel 对象还没有注册到 Netty 线程的 Selector 上，Debug 结果如下图：
上图中的 startThread()方法实际上是给 executor 提交了一个任务，紧接着 main 线程就调用了 addTask()方法，将 task 添加到 EventLoop 对象的任务队列中，而这个 task 的内容就是执行注册操作。在添加了注册任务之后，Netty 线程就会在 select 完成后，执行队列中的任务，将 NioServerSocketChannel 注册到该线程的 Selector 上。接下来分析一下 AbstractChannel 的 register0()方法：
1 private void register0(ChannelPromise promise) { 2 ????try { 3 ????????// check if the channel is still open as it could be closed in the mean time when the register 4 ????????// call was outside of the eventLoop 5 ????????if (!promise.setUncancellable() || !ensureOpen(promise)) { 6 ????????????return; 7 ????????} 8 ????????boolean firstRegistration = neverRegistered; 9 ????????//注册通道10 ????????doRegister();11 ????????neverRegistered = false;12 ????????registered = true;13 14 ????????// Ensure we call handlerAdded(...) before we actually notify the promise. This is needed as the15 ????????// user may already fire events through the pipeline in the ChannelFutureListener.16 ????????//添加服务端 Channel 的 Handler17 ????????pipeline.invokeHandlerAddedIfNeeded();18 19 ????????safeSetSuccess(promise);20 ????????//触发通道注册事件在 pipeline 上传播21 ????????pipeline.fireChannelRegistered();22 ????????// Only fire a channelActive if the channel has never been registered. This prevents firing23 ????????// multiple channel actives if the channel is deregistered and re-registered.24 ????????if (isActive()) {//第一次运行到这儿时，结果为 false,因为此时还没有 bind25 ????????????if (firstRegistration) {26 ????????????????pipeline.fireChannelActive();27 ????????????} else if (config().isAutoRead()) {28 ????????????????// This channel was registered before and autoRead() is set. This means we need to begin read29 ????????????????// again so that we process inbound data.30 ????????????????//31 ????????????????// See https://github.com/netty/netty/issues/480532 ????????????????beginRead();33 ????????????}34 ????????}35 ????} catch (Throwable t) {36 ????????// Close the channel directly to avoid FD leak.37 ????????closeForcibly();38 ????????closeFuture.setClosed();39 ????????safeSetFailure(promise, t);40 ????}41 }doRegister()方法实际上就是 Java NIO 中将通道注册到 Selector 上的操作：
1 selectionKey = javaChannel().register(eventLoop().selector, 0, this);//这里感兴趣的事件传入的是 0
pipeline.invokeHandlerAddedIfNeeded() 和 pipeline.fireChannelRegistered() 则是用来添加 Handler 并触发 Handler 别添加的事件的动作。
在 isActive()这个方法，由于当前是 NioServerSocketChannel，所以实际上是判断当前通道是否成功绑定到一个地址，很显然到目前为止，只是创建了通道并注册到 Selector 上，还没由绑定。
3. 绑定指定端口
在 initAndRegister()方法结束后，main 线程开始调用 doBind0()方法，该方法将绑定操作封装成任务交给 Netty 线程去执行。最后，调用 DefaultPipeline 中的 HeadContext 的 bind()方法，然后通过 unsafe.bind(localAddress,promise)完成绑定：
1 @Override 2 public final void bind(final SocketAddress localAddress, final ChannelPromise promise) { 3 ????//... 4 ????//显然这里返回的是 false 5 ????boolean wasActive = isActive(); 6 ????try { 7 ????????//绑定操作 8 ????????doBind(localAddress); 9 ????} catch (Throwable t) {10 ????????safeSetFailure(promise, t);11 ????????closeIfClosed();12 ????????return;13 ????}14 15 ????if (!wasActive && isActive()) {16 ????????invokeLater(new Runnable() {17 ????????????@Override18 ????????????public void run() {19 ????????????????//这里才是触发服务端 Channel 激活事件的地方20 ????????????????pipeline.fireChannelActive();21 ????????????}22 ????????});23 ????}24 25 ????safeSetSuccess(promise);26 }这个过程，建议 Debug 跟一下代码，比较清楚代码是如何一步一步到 HeadContext 中来的。接下来分析一下 doBind()方法：
1 @Override2 protected void doBind(SocketAddress localAddress) throws Exception {3 ????if (PlatformDependent.javaVersion() >= 7) {4 ????????javaChannel().bind(localAddress, config.getBacklog());5 ????} else {6 ????????javaChannel().socket().bind(localAddress, config.getBacklog());7 ????}8 }最终是根据平台及其 Java 版本来调用 JDK 中的绑定方法。在绑定完成后，会触发通道激活事件，在 HeadContext 中经过时，发现它里面有这么一行代码：
1 readIfIsAutoRead();
Debug 一下，发现这个方法最终会调用到 HeadContext 的 read()方法，该方法是调用了 unsafe.beginRead()，紧接着就到了 AbstractNioChannel 的 doBeginRead()方法：
1 @Override 2 protected void doBeginRead() throws Exception { 3 ????// Channel.read() or ChannelHandlerContext.read() was called 4 ????final SelectionKey selectionKey = this.selectionKey; 5 ????if (!selectionKey.isValid()) { 6 ????????return; 7 ????} 8 ?9 ????readPending = true;10 11 ????final int interestOps = selectionKey.interestOps();12 ????if ((interestOps & readInterestOp) == 0) {//说明对 OP_ACCEPT 不感兴趣13 ????????selectionKey.interestOps(interestOps | readInterestOp);//通过 | 修改感兴趣的事件14 ????}15 }前面通过反射创建 NioServerSocketChannel 对象时，调用了父类也就是 AbstractNioChannel 的构造方法，将 readInterestOp 设置为 16 了，在 NIO 中就是 OP_ACCEPT。从此，该 NioServerSocketChannel 就可以接收客户端连接了。
4. 总结
在 Netty 服务端启动过程中，主线程仅仅是创建了 EventLoopGroup 和启动引导对象，然后发起绑定操作。这个过程中的绑定，注册等操作都是主线程封装成任务交给 Netty 线程去执行的。
由于 Netty 代码中抽象类和接口都比较多，所以某些地方调用的方法有很多种实现，不熟悉的时候可以通过 Debug 来确定。
Netty 服务端启动过程
原文地址：https://www.cnblogs.com/magexi/p/10228780.html