大纲
1.Pipeline和Handler的作用和构成
2.ChannelHandler的分类
3.几个特殊的ChannelHandler
4.ChannelHandler的生命周期
5.ChannelPipeline的事件处理
6.关于ChannelPipeline的问题整理
7.ChannelPipeline主要包括三部分内容
8.ChannelPipeline的初始化
9.ChannelPipeline添加ChannelHandler
10.ChannelPipeline删除ChannelHandler
11.Inbound事件的传播
12.Outbound事件的传播
13.ChannelPipeline中异常的传播
14.ChannelPipeline总结
10.ChannelPipeline删除ChannelHandler
Netty最大的特征之一就是ChannelHandler是可插拔的，可以动态编织ChannelPipeline。比如在客户端首次连接服务端时，需要进行权限认证，认证通过后就可以不用再认证了。下面的AuthHandler便实现了只对第一个传来的数据包进行认证校验。如果通过验证则删除此AuthHandler，这样后续传来的数据包便不会再校验了。
public class AuthHandler extends SimpleChannelInboundHandler<ByteBuf> {...protected void channelRead0(ChannelHandlerContext ctx, ByteBuf data) throw Exception {if (verify(data)) {ctx.pipeline().remove(this); } else {ctx.close();}}
}DefaultChannelPipeline的remove()方法如下：
public class DefaultChannelPipeline implements ChannelPipeline {...@Overridepublic final ChannelPipeline remove(ChannelHandler handler) {remove(getContextOrDie(handler));return this;}private AbstractChannelHandlerContext getContextOrDie(ChannelHandler handler) {AbstractChannelHandlerContext ctx = (AbstractChannelHandlerContext) context(handler);if (ctx == null) {throw new NoSuchElementException(handler.getClass().getName());} else {return ctx;}}@Overridepublic final ChannelHandlerContext context(ChannelHandler handler) {...AbstractChannelHandlerContext ctx = head.next;//遍历双向链表for (;;) {...if (ctx.handler() == handler) return ctx;ctx = ctx.next;}}private AbstractChannelHandlerContext remove(final AbstractChannelHandlerContext ctx) { //Pipeline中的head和tail结点不能被删除assert ctx != head && ctx != tail;synchronized (this) {//调整链表指针并删除remove0(ctx);...}//回调用户在这个要删除的ChannelHandler实现的handlerRemoved()方法callHandlerRemoved0(ctx);return ctx;}private static void remove0(AbstractChannelHandlerContext ctx) {AbstractChannelHandlerContext prev = ctx.prev;AbstractChannelHandlerContext next = ctx.next;prev.next = next;next.prev = prev;}private void callHandlerRemoved0(final AbstractChannelHandlerContext ctx) {...ctx.handler().handlerRemoved(ctx);...}...
}ChannelPipeline删除ChannelHandler的步骤：
一.遍历双向链表，根据ChannelHandler找到对应的ChannelHandlerContext结点。
二.通过调整ChannelPipeline中双向链表的指针来删除对应的ChannelHandlerContext结点。
三.回调用户在这个要删除的ChannelHandler实现的handlerRemoved()方法，比如进行资源清理。
11.Inbound事件的传播
(1)Unsafe的介绍
(2)Unsafe的继承结构
(3)Unsafe的分类
(4)ChannelPipeline中Inbound事件传播
(5)ChannelPipeline中的头结点和尾结点
(6)Inbound事件的传播总结
(1)Unsafe的介绍
Unsafe和ChannelPipeline密切相关，ChannelPipeline中有关IO的操作最终都会落地到Unsafe的。Unsafe是不安全的意思，即不要在应用程序里直接使用Unsafe及它的衍生类对象。Unsafe是在Channel中定义的，是属于Channel的内部类。Unsafe中的接口操作都和JDK底层相关，包括：分配内存、Socket四元组信息、注册事件循环、绑定端口、Socket的连接和关闭、Socket的读写。
//A nexus to a network socket or a component which is capable of I/O operations such as read, write, connect, and bind.
public interface Channel extends AttributeMap, ChannelOutboundInvoker, Comparable<Channel> {//Returns the globally unique identifier of this Channel.ChannelId id();//Return the EventLoop this Channel was registered to.EventLoop eventLoop();//Returns the parent of this channel.Channel parent();//Returns the configuration of this channel.ChannelConfig config();//Returns true if the Channel is open and may get active laterboolean isOpen();//Returns true if the Channel is registered with an EventLoop.boolean isRegistered();//Return true if the Channel is active and so connected.boolean isActive();//Return the ChannelMetadata of the Channel which describe the nature of the Channel.ChannelMetadata metadata();//Returns the local address where this channel is bound to. //The returned SocketAddress is supposed to be down-cast into more concrete type such as InetSocketAddress to retrieve the detailed information. SocketAddress localAddress();//Returns the remote address where this channel is connected to. //The returned SocketAddress is supposed to be down-cast into more concrete type such as InetSocketAddress to retrieve the detailed information.SocketAddress remoteAddress();//Returns the ChannelFuture which will be notified when this channel is closed. //This method always returns the same future instance.ChannelFuture closeFuture();//Returns true if and only if the I/O thread will perform the requested write operation immediately. //Any write requests made when this method returns false are queued until the I/O thread is ready to process the queued write requests.boolean isWritable();//Get how many bytes can be written until #isWritable() returns false.//This quantity will always be non-negative. If #isWritable() is false then 0.long bytesBeforeUnwritable();//Get how many bytes must be drained from underlying buffers until #isWritable() returns true.//This quantity will always be non-negative. If #isWritable() is true then 0.long bytesBeforeWritable();//Returns an <em>internal-use-only</em> object that provides unsafe operations.Unsafe unsafe();//Return the assigned ChannelPipeline.ChannelPipeline pipeline();//Return the assigned ByteBufAllocator which will be used to allocate ByteBufs.ByteBufAllocator alloc();@OverrideChannel read();@OverrideChannel flush();//Unsafe operations that should never be called from user-code. //These methods are only provided to implement the actual transport, and must be invoked from an I/O thread except for the following methods://#invoker()//#localAddress()//#remoteAddress()//#closeForcibly()//#register(EventLoop, ChannelPromise)//#deregister(ChannelPromise)//#voidPromise()interface Unsafe {//Return the assigned RecvByteBufAllocator.Handle which will be used to allocate ByteBuf's when receiving data.RecvByteBufAllocator.Handle recvBufAllocHandle();//Return the SocketAddress to which is bound local or null if none.SocketAddress localAddress();//Return the SocketAddress to which is bound remote or null if none is bound yet.SocketAddress remoteAddress();//Register the Channel of the ChannelPromise and notify the ChannelFuture once the registration was complete.void register(EventLoop eventLoop, ChannelPromise promise);//Bind the SocketAddress to the Channel of the ChannelPromise and notify it once its done.void bind(SocketAddress localAddress, ChannelPromise promise);//Connect the Channel of the given ChannelFuture with the given remote SocketAddress.//If a specific local SocketAddress should be used it need to be given as argument. Otherwise just pass null to it.//he ChannelPromise will get notified once the connect operation was complete.void connect(SocketAddress remoteAddress, SocketAddress localAddress, ChannelPromise promise);//Disconnect the Channel of the ChannelFuture and notify the ChannelPromise once the operation was complete.void disconnect(ChannelPromise promise);//Close the Channel of the ChannelPromise and notify the ChannelPromise once the operation was complete.void close(ChannelPromise promise);//Closes the Channel immediately without firing any events. Probably only useful when registration attempt failed.void closeForcibly();//Deregister the Channel of the ChannelPromise from EventLoop and notify the ChannelPromise once the operation was complete.void deregister(ChannelPromise promise);//Schedules a read operation that fills the inbound buffer of the first ChannelInboundHandler in the ChannelPipeline.//If there's already a pending read operation, this method does nothing.void beginRead();//Schedules a write operation.void write(Object msg, ChannelPromise promise);//Flush out all write operations scheduled via #write(Object, ChannelPromise).void flush();//Return a special ChannelPromise which can be reused and passed to the operations in Unsafe.//It will never be notified of a success or error and so is only a placeholder for operations//that take a ChannelPromise as argument but for which you not want to get notified.ChannelPromise voidPromise();//Returns the ChannelOutboundBuffer of the Channel where the pending write requests are stored.ChannelOutboundBuffer outboundBuffer();}
}public abstract class AbstractNioChannel extends AbstractChannel {...public interface NioUnsafe extends Unsafe {//Return underlying SelectableChannelSelectableChannel ch();//Finish connectvoid finishConnect();//Read from underlying SelectableChannelvoid read();void forceFlush();}...
}(2)Unsafe的继承结构
一.NioUnsafe增加了可以访问底层JDK的SelectableChannel的功能，定义了从SelectableChannel读取数据的read()方法。
二.AbstractUnsafe实现了大部分Unsafe的功能。
三.AbstractNioUnsafe主要是通过代理到其外部类AbstractNioChannel获得与JDK NIO相关的一些信息，比如SelectableChannel、SelectionKey等。
四.NioMessageUnsafe和NioByteUnsafe是处在同一层次的抽象，Netty将一个新连接的建立也当作一个IO操作来处理，这里Message的含义可以当作一个SelectableChannel，读的意思就是接收一个SelectableChannel。
(3)Unsafe的分类
有两种类型的Unsafe：一种是与连接的字节数据读写相关的NioByteUnsafe，另一种是与新连接建立操作相关的NioMessageUnsafe。
一.NioByteUnsafe的读和写
NioByteUnsafe的读会被委托到NioByteChannel的doReadBytes()方法进行读取处理，doReadBytes()方法会将JDK的SelectableChannel的字节数据读取到Netty的ByteBuf中。
NioByteUnsafe中的写有两个方法，一个是write()方法，一个是flush()方法。write()方法是将数据添加到Netty的缓冲区，flush()方法是将Netty缓冲区的字节流写到TCP缓冲区，并最终委托到NioSocketChannel的doWrite()方法通过JDK底层Channel的write()方法写数据。
//AbstractNioChannel base class for Channels that operate on bytes.
public abstract class AbstractNioByteChannel extends AbstractNioChannel {...protected class NioByteUnsafe extends AbstractNioUnsafe {...//NioByteUnsafe的读@Overridepublic final void read() {...doReadBytes(byteBuf);...}}
}public class NioSocketChannel extends AbstractNioByteChannel implements SocketChannel {...@Overrideprotected int doReadBytes(ByteBuf byteBuf) throws Exception {final RecvByteBufAllocator.Handle allocHandle = unsafe().recvBufAllocHandle();allocHandle.attemptedBytesRead(byteBuf.writableBytes());return byteBuf.writeBytes(javaChannel(), allocHandle.attemptedBytesRead());}@Overrideprotected void doWrite(ChannelOutboundBuffer in) throws Exception {...ByteBuffer[] nioBuffers = in.nioBuffers();SocketChannel ch = javaChannel();...ByteBuffer nioBuffer = nioBuffers[0];...ch.write(nioBuffer)...}
}public abstract class AbstractChannel extends DefaultAttributeMap implements Channel {...protected abstract class AbstractUnsafe implements Unsafe {private volatile ChannelOutboundBuffer outboundBuffer = new ChannelOutboundBuffer(AbstractChannel.this);...//NioByteUnsafe的写@Overridepublic final void write(Object msg, ChannelPromise promise) {assertEventLoop();ChannelOutboundBuffer outboundBuffer = this.outboundBuffer;...outboundBuffer.addMessage(msg, size, promise);}//NioByteUnsafe的写@Overridepublic final void flush() {assertEventLoop();ChannelOutboundBuffer outboundBuffer = this.outboundBuffer;...outboundBuffer.addFlush();flush0();}@SuppressWarnings("deprecation")protected void flush0() {...doWrite(outboundBuffer);...}}
}二.NioMessageUnsafe的读
NioMessageUnsafe的读会委托到NioServerSocketChannel的doReadMessages()方法进行处理。doReadMessages()方法会调用JDK的accept()方法新建立一个连接，并将这个连接放到一个List里以方便后续进行批量处理。
public abstract class AbstractNioMessageChannel extends AbstractNioChannel {...private final class NioMessageUnsafe extends AbstractNioUnsafe {private final List<Object> readBuf = new ArrayList<Object>();//NioMessageUnsafe的读@Overridepublic void read() {...doReadMessages(readBuf)...}}
}public class NioServerSocketChannel extends AbstractNioMessageChannel implements ServerSocketChannel {...@Overrideprotected int doReadMessages(List<Object> buf) throws Exception {SocketChannel ch = javaChannel().accept();if (ch != null) {buf.add(new NioSocketChannel(this, ch));return 1;}return 0;}
}(4)ChannelPipeline中Inbound事件传播
当新连接已准备接入或者已经存在的连接有数据可读时，会在NioEventLoop的processSelectedKey()方法中执行unsafe.read()。
如果是新连接已准备接入，执行的是NioMessageUnsafe的read()方法。如果是已经存在的连接有数据可读，执行的是NioByteUnsafe的read()方法。
最后都会执行pipeline.fireChannelRead()引发ChannelPipeline的读事件传播。首先会从HeadContext结点开始，也就是调用HeadContext的channelRead()方法。然后触发调用AbstractChannelHandlerContext的fireChannelRead()方法，接着通过findContextInbound()方法找到HeadContext的下一个结点，然后通过invokeChannelRead()方法继续调用该结点的channelRead()方法，直到最后一个结点TailContext。
public final class NioEventLoop extends SingleThreadEventLoop {...private void processSelectedKey(SelectionKey k, AbstractNioChannel ch) {final AbstractNioChannel.NioUnsafe unsafe = ch.unsafe();...//新连接已准备接入或者已经存在的连接有数据可读if ((readyOps & (SelectionKey.OP_READ | SelectionKey.OP_ACCEPT)) != 0 || readyOps == 0) {//如果是新连接已准备接入，则调用NioMessageUnsafe的read()方法//如果是已经存在的连接有数据可读，执行的是NioByteUnsafe的read()方法unsafe.read();if (!ch.isOpen()) {return;}}}...
}public abstract class AbstractNioMessageChannel extends AbstractNioChannel {...private final class NioMessageUnsafe extends AbstractNioUnsafe {private final List<Object> readBuf = new ArrayList<Object>();@Overridepublic void read() {assert eventLoop().inEventLoop();final ChannelConfig config = config();final ChannelPipeline pipeline = pipeline();//创建ByteBuf分配器final RecvByteBufAllocator.Handle allocHandle = unsafe().recvBufAllocHandle();allocHandle.reset(config);...do {int localRead = doReadMessages(readBuf);...allocHandle.incMessagesRead(localRead);} while (allocHandle.continueReading());int size = readBuf.size();for (int i = 0; i < size; i ++) {readPending = false;//调用DefaultChannelPipeline的fireChannelRead()方法从Head结点开始传播事件pipeline.fireChannelRead(readBuf.get(i));}readBuf.clear();allocHandle.readComplete();//调用DefaultChannelPipeline的fireChannelReadComplete()方法从Head结点开始传播事件pipeline.fireChannelReadComplete();...}...}
}public abstract class AbstractNioByteChannel extends AbstractNioChannel {...protected class NioByteUnsafe extends AbstractNioUnsafe {...@Overridepublic final void read() {final ChannelConfig config = config();final ChannelPipeline pipeline = pipeline();//创建ByteBuf分配器final ByteBufAllocator allocator = config.getAllocator();final RecvByteBufAllocator.Handle allocHandle = recvBufAllocHandle();allocHandle.reset(config);ByteBuf byteBuf = null;do {//1.分配一个ByteBufbyteBuf = allocHandle.allocate(allocator);//2.将数据读取到分配的ByteBuf中allocHandle.lastBytesRead(doReadBytes(byteBuf));if (allocHandle.lastBytesRead() <= 0) {byteBuf.release();byteBuf = null;close = allocHandle.lastBytesRead() < 0;break;}...//3.调用DefaultChannelPipeline的fireChannelRead()方法从Head结点开始传播事件pipeline.fireChannelRead(byteBuf);byteBuf = null;} while (allocHandle.continueReading());allocHandle.readComplete();//4.调用DefaultChannelPipeline的fireChannelReadComplete()方法从Head结点开始传播事件pipeline.fireChannelReadComplete();...}}
}public class DefaultChannelPipeline implements ChannelPipeline {//ChannelPipeline的头结点final AbstractChannelHandlerContext head;...@Overridepublic final ChannelPipeline fireChannelRead(Object msg) {AbstractChannelHandlerContext.invokeChannelRead(head, msg);return this;}final class HeadContext extends AbstractChannelHandlerContext implements ChannelOutboundHandler, ChannelInboundHandler {...@Overridepublic void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {//调用AbstractChannelHandlerContext的fireChannelRead()方法ctx.fireChannelRead(msg);}}
}abstract class AbstractChannelHandlerContext extends DefaultAttributeMap implements ChannelHandlerContext, ResourceLeakHint {volatile AbstractChannelHandlerContext next;volatile AbstractChannelHandlerContext prev;...static void invokeChannelRead(final AbstractChannelHandlerContext next, Object msg) {final Object m = next.pipeline.touch(ObjectUtil.checkNotNull(msg, "msg"), next);EventExecutor executor = next.executor();if (executor.inEventLoop()) {//调用AbstractChannelHandlerContext的invokeChannelRead()方法next.invokeChannelRead(m);} else {executor.execute(new Runnable() {@Overridepublic void run() {next.invokeChannelRead(m);}});}}private void invokeChannelRead(Object msg) {if (invokeHandler()) {try {//比如调用HeadContext的channelRead()方法((ChannelInboundHandler) handler()).channelRead(this, msg);} catch (Throwable t) {notifyHandlerException(t);}} else {fireChannelRead(msg);}}@Overridepublic ChannelHandlerContext fireChannelRead(final Object msg) {invokeChannelRead(findContextInbound(), msg);return this;}//寻找下一个结点private AbstractChannelHandlerContext findContextInbound() {AbstractChannelHandlerContext ctx = this;do {ctx = ctx.next;} while (!ctx.inbound);return ctx;}
}(5)ChannelPipeline中的头结点和尾结点
HeadContext是一个同时属于Inbound类型和Outbound类型的ChannelHandler，TailContext则只是一个属于Inbound类型的ChannelHandler。
HeadContext结点的作用就是作为头结点开始传递读写事件并调用unsafe进行实际的读写操作。比如Channel读完一次数据后，HeadContext的channelReadComplete()方法会被调用。然后继续执行如下的调用流程：readIfAutoRead() -> channel.read() -> pipeline.read() -> HeadContext.read() -> unsafe.beginRead() -> 再次注册读事件。所以Channel读完一次数据后，会继续向Selector注册读事件。这样只要Channel活跃就可以连续不断地读取数据，然后数据又会通过ChannelPipeline传递到HeadContext结点。
TailContext结点的作用是通过让方法体为空来终止大部分事件的传播，它的exceptionCaugh()方法和channelRead()方法分别会发出告警日志以及释放到达该结点的对象。
public class DefaultChannelPipeline implements ChannelPipeline {//ChannelPipeline的头结点final AbstractChannelHandlerContext head;//ChannelPipeline的尾结点final AbstractChannelHandlerContext tail;...@Overridepublic final ChannelPipeline fireChannelRead(Object msg) {AbstractChannelHandlerContext.invokeChannelRead(head, msg);return this;}final class HeadContext extends AbstractChannelHandlerContext implements ChannelOutboundHandler, ChannelInboundHandler {private final Unsafe unsafe;HeadContext(DefaultChannelPipeline pipeline) {super(pipeline, null, HEAD_NAME, false, true);unsafe = pipeline.channel().unsafe();setAddComplete();}@Overridepublic void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {//调用AbstractChannelHandlerContext的fireChannelRead()方法ctx.fireChannelRead(msg);}@Overridepublic void channelReadComplete(ChannelHandlerContext ctx) throws Exception {ctx.fireChannelReadComplete();readIfIsAutoRead();}private void readIfIsAutoRead() {if (channel.config().isAutoRead()) {channel.read();}}}final class TailContext extends AbstractChannelHandlerContext implements ChannelInboundHandler {TailContext(DefaultChannelPipeline pipeline) {super(pipeline, null, TAIL_NAME, true, false);setAddComplete();}@Overridepublic void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {onUnhandledInboundMessage(msg);}...}//Called once a message hit the end of the ChannelPipeline without been handled by the user in ChannelInboundHandler#channelRead(ChannelHandlerContext, Object).//This method is responsible to call ReferenceCountUtil#release(Object) on the given msg at some point.protected void onUnhandledInboundMessage(Object msg) {try {logger.debug("Discarded inbound message {} that reached at the tail of the pipeline. " + "Please check your pipeline configuration.", msg);} finally {ReferenceCountUtil.release(msg);}}
}(6)Inbound事件的传播总结
一般用户自定义的ChannelInboundHandler都继承自ChannelInboundHandlerAdapter。如果用户代码没有覆盖ChannelInboundHandlerAdapter的channelXXX()方法，那么Inbound事件会从HeadContext开始遍历ChannelPipeline的双向链表进行传播，并默认情况下传播到TailContext结点。
如果用户代码覆盖了ChannelInboundHandlerAdapter的channelXXX()方法，那么事件传播就会在当前结点结束。所以如果此时这个ChannelHandler又忘记了手动释放业务对象ByteBuf，则可能会造成内存泄露，而SimpleChannelInboundHandler则可以帮用户自动释放业务对象。
如果用户代码调用了ChannelHandlerContext的fireXXX()方法来传播事件，那么该事件就从当前结点开始往下传播。
public class ChannelInboundHandlerAdapter extends ChannelHandlerAdapter implements ChannelInboundHandler {//Calls ChannelHandlerContext#fireChannelRegistered() to forward to the next ChannelInboundHandler in the ChannelPipeline. @Overridepublic void channelRegistered(ChannelHandlerContext ctx) throws Exception {ctx.fireChannelRegistered();}//Calls ChannelHandlerContext#fireChannelUnregistered() to forward to the next ChannelInboundHandler in the ChannelPipeline. @Overridepublic void channelUnregistered(ChannelHandlerContext ctx) throws Exception {ctx.fireChannelUnregistered();}//Calls ChannelHandlerContext#fireChannelActive() to forward to the next ChannelInboundHandler in the ChannelPipeline.@Overridepublic void channelActive(ChannelHandlerContext ctx) throws Exception {ctx.fireChannelActive();}//Calls ChannelHandlerContext#fireChannelInactive() to forward to the next ChannelInboundHandler in the ChannelPipeline.@Overridepublic void channelInactive(ChannelHandlerContext ctx) throws Exception {ctx.fireChannelInactive();}//Calls ChannelHandlerContext#fireChannelRead(Object) to forward to the next ChannelInboundHandler in the ChannelPipeline.@Overridepublic void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {ctx.fireChannelRead(msg);}//Calls ChannelHandlerContext#fireChannelReadComplete() to forward to the next ChannelInboundHandler in the ChannelPipeline. @Overridepublic void channelReadComplete(ChannelHandlerContext ctx) throws Exception {ctx.fireChannelReadComplete();}//Calls ChannelHandlerContext#fireUserEventTriggered(Object) to forward to the next ChannelInboundHandler in the ChannelPipeline. @Overridepublic void userEventTriggered(ChannelHandlerContext ctx, Object evt) throws Exception {ctx.fireUserEventTriggered(evt);}//Calls ChannelHandlerContext#fireChannelWritabilityChanged() to forward to the next ChannelInboundHandler in the ChannelPipeline. @Overridepublic void channelWritabilityChanged(ChannelHandlerContext ctx) throws Exception {ctx.fireChannelWritabilityChanged();}//Calls ChannelHandlerContext#fireExceptionCaught(Throwable) to forward to the next ChannelHandler in the ChannelPipeline.@Overridepublic void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {ctx.fireExceptionCaught(cause);}
}12.Outbound事件的传播
(1)触发Outbound事件传播的入口
(2)Outbound事件传播的源码
(3)总结
(1)触发Outbound事件传播的入口
在消息推送系统中，可能会有如下代码，意思是根据用户ID获得对应的Channel，然后向用户推送消息。
Channel channel = ChannelManager.getChannel(userId);
channel.writeAndFlush(response);(2)Outbound事件传播的源码
如果通过Channel来传播Outbound事件，则是从TailContext开始传播的。
和Inbound事件一样，Netty为了保证程序的高效执行，所有核心操作都要在Reactor线程中处理。如果业务线程调用了Channel的方法，那么Netty会将该操作封装成一个Task任务添加到任务队列中，随后在Reactor线程的事件循环中执行。
findContextOutbound()方法找Outbound结点的过程和findContextInbound()方法找Inbound结点类似，需要反向遍历ChannelPipeline中的双向链表，一直遍历到第一个Outbound结点HeadCountext。
如果用户的ChannelHandler覆盖了Outbound类型的方法，但没有把事件在方法中继续传播下去，那么会导致该事件的传播中断。
最后一个Inbound结点是TailContext，最后一个Outbound结点是HeadContext，而数据最终会落到HeadContext的write()方法上。
下面是channel.writeAndFlush()方法的源码：
public interface ChannelOutboundInvoker {...//Shortcut for call #write(Object) and #flush().ChannelFuture writeAndFlush(Object msg);...
}public abstract class AbstractChannel extends DefaultAttributeMap implements Channel {private final Channel parent;private final ChannelId id;private final Unsafe unsafe;private final DefaultChannelPipeline pipeline;...protected AbstractChannel(Channel parent) {this.parent = parent;id = newId();unsafe = newUnsafe();pipeline = newChannelPipeline();}protected DefaultChannelPipeline newChannelPipeline() {return new DefaultChannelPipeline(this);}@Overridepublic ChannelFuture writeAndFlush(Object msg) {return pipeline.writeAndFlush(msg);}...
}public class DefaultChannelPipeline implements ChannelPipeline {final AbstractChannelHandlerContext head;final AbstractChannelHandlerContext tail;private final Channel channel;protected DefaultChannelPipeline(Channel channel) {this.channel = ObjectUtil.checkNotNull(channel, "channel");tail = new TailContext(this);head = new HeadContext(this);head.next = tail;tail.prev = head;}@Overridepublic final ChannelFuture writeAndFlush(Object msg) {//从TailContext开始传播//但TailContext没有重写writeAndFlush()方法//所以会调用AbstractChannelHandlerContext的writeAndFlush()方法return tail.writeAndFlush(msg);}...
}abstract class AbstractChannelHandlerContext extends DefaultAttributeMap implements ChannelHandlerContext, ResourceLeakHint {volatile AbstractChannelHandlerContext next;volatile AbstractChannelHandlerContext prev;...@Overridepublic ChannelFuture writeAndFlush(Object msg) {return writeAndFlush(msg, newPromise());}@Overridepublic ChannelFuture writeAndFlush(Object msg, ChannelPromise promise) {if (msg == null) throw new NullPointerException("msg");if (!validatePromise(promise, true)) {ReferenceCountUtil.release(msg);return promise;}write(msg, true, promise);return promise;}private void write(Object msg, boolean flush, ChannelPromise promise) {//反向遍历链表进行查找AbstractChannelHandlerContext next = findContextOutbound();final Object m = pipeline.touch(msg, next);EventExecutor executor = next.executor();//最终都会由Reactor线程处理Channel的数据读写if (executor.inEventLoop()) {if (flush) {//调用结点的invokeWriteAndFlush()方法next.invokeWriteAndFlush(m, promise);} else {next.invokeWrite(m, promise);}} else {AbstractWriteTask task;if (flush) {task = WriteAndFlushTask.newInstance(next, m, promise);} else {task = WriteTask.newInstance(next, m, promise);}safeExecute(executor, task, promise, m);}}private void invokeWriteAndFlush(Object msg, ChannelPromise promise) {if (invokeHandler()) {//逐个调用ChannelHandler结点的write()方法，但前提是当前ChannelHandler可以往下传//即write()方法在最后也像ChannelOutboundHandlerAdapter那样，调用了ctx.write()往下传播invokeWrite0(msg, promise);//逐个调用ChannelHandler结点的flush()方法，但前提是当前ChannelHandler可以往下传//即flush()方法在最后也像ChannelOutboundHandlerAdapter那样，调用了ctx.flush()往下传播invokeFlush0();} else {writeAndFlush(msg, promise);}}private void invokeWrite0(Object msg, ChannelPromise promise) {try {//逐个调用，最终回到HeadContext的write()方法((ChannelOutboundHandler) handler()).write(this, msg, promise);} catch (Throwable t) {notifyOutboundHandlerException(t, promise);}}private void invokeFlush0() {try {//逐个调用，最终回到HeadContext的flush()方法((ChannelOutboundHandler) handler()).flush(this);} catch (Throwable t) {notifyHandlerException(t);}}...
}public class DefaultChannelPipeline implements ChannelPipeline {...final class HeadContext extends AbstractChannelHandlerContext implements ChannelOutboundHandler, ChannelInboundHandler {private final Unsafe unsafe;HeadContext(DefaultChannelPipeline pipeline) {super(pipeline, null, HEAD_NAME, false, true);unsafe = pipeline.channel().unsafe();setAddComplete();}...@Overridepublic void write(ChannelHandlerContext ctx, Object msg, ChannelPromise promise) throws Exception {unsafe.write(msg, promise);}@Overridepublic void flush(ChannelHandlerContext ctx) throws Exception {unsafe.flush();}}...
}//Skeleton implementation of a ChannelOutboundHandler. This implementation just forwards each method call via the ChannelHandlerContext.
public class ChannelOutboundHandlerAdapter extends ChannelHandlerAdapter implements ChannelOutboundHandler {//Calls ChannelHandlerContext#bind(SocketAddress, ChannelPromise) to forward to the next ChannelOutboundHandler in the ChannelPipeline.@Overridepublic void bind(ChannelHandlerContext ctx, SocketAddress localAddress, ChannelPromise promise) throws Exception {ctx.bind(localAddress, promise);}//Calls ChannelHandlerContext#connect(SocketAddress, SocketAddress, ChannelPromise) to forward to the next ChannelOutboundHandler in the ChannelPipeline. @Overridepublic void connect(ChannelHandlerContext ctx, SocketAddress remoteAddress, SocketAddress localAddress, ChannelPromise promise) throws Exception {ctx.connect(remoteAddress, localAddress, promise);}//Calls ChannelHandlerContext#disconnect(ChannelPromise) to forward to the next ChannelOutboundHandler in the ChannelPipeline.@Overridepublic void disconnect(ChannelHandlerContext ctx, ChannelPromise promise) throws Exception {ctx.disconnect(promise);}//Calls ChannelHandlerContext#close(ChannelPromise) to forward to the next ChannelOutboundHandler in the ChannelPipeline.@Overridepublic void close(ChannelHandlerContext ctx, ChannelPromise promise) throws Exception {ctx.close(promise);}//Calls ChannelHandlerContext#deregister(ChannelPromise) to forward to the next ChannelOutboundHandler in the ChannelPipeline.@Overridepublic void deregister(ChannelHandlerContext ctx, ChannelPromise promise) throws Exception {ctx.deregister(promise);}//Calls ChannelHandlerContext#read() to forward to the next ChannelOutboundHandler in the ChannelPipeline.@Overridepublic void read(ChannelHandlerContext ctx) throws Exception {ctx.read();}//Calls ChannelHandlerContext#write(Object, ChannelPromise)} to forward to the next ChannelOutboundHandler in the ChannelPipeline.@Overridepublic void write(ChannelHandlerContext ctx, Object msg, ChannelPromise promise) throws Exception {ctx.write(msg, promise);}//Calls ChannelHandlerContext#flush() to forward to the next ChannelOutboundHandler in the ChannelPipeline.@Overridepublic void flush(ChannelHandlerContext ctx) throws Exception {ctx.flush();}
}(3)总结
Outbound事件的传播机制和Inbound事件的传播机制类似。但Outbound事件是从链表尾部开始向前传播，而Inbound事件是从链表头部开始向后传播。Outbound事件传播中的写数据，最终都会落到HeadContext结点中的unsafe进行处理。
13.ChannelPipeline中异常的传播
Inbound事件和Outbound事件在传播时发生异常都会调用notifyHandlerExecption()方法，该方法会按Inbound事件的传播顺序找每个结点的异常处理方法exceptionCaught()进行处理。
我们通常在自定义的ChannelHandler中实现一个处理异常的方法exceptionCaught()，统一处理ChannelPipeline过程中的所有异常。这个自定义ChannelHandler一般继承自ChannelDuplexHandler，表示该结点既是一个Inbound结点，又是一个Outbound结点。
如果我们在自定义的ChannelHandler中没有处理异常，由于ChannelHandler通常都继承了ChannelInboundHandlerAdapter，通过其默认实现的exceptionCaught()方法可知异常会一直往下传递，直到最后一个结点的异常处理方法exceptionCaught()中结束。因此如果异常处理方法exceptionCaught()在ChannelPipeline中间的结点实现，则该结点后面的ChannelHandler抛出的异常就没法处理了。所以一般会在ChannelHandler链表的末尾结点实现处理异常的方法exceptionCaught()。
需要注意的是：在任何结点中发生的异常都会向下一个结点进行传递。
public class DefaultChannelPipeline implements ChannelPipeline {...@Overridepublic final ChannelPipeline fireChannelRead(Object msg) {AbstractChannelHandlerContext.invokeChannelRead(head, msg);return this;}...
}abstract class AbstractChannelHandlerContext extends DefaultAttributeMap implements ChannelHandlerContext, ResourceLeakHint { ...@Overridepublic ChannelHandlerContext fireChannelRead(final Object msg) {invokeChannelRead(findContextInbound(), msg);return this;}static void invokeChannelRead(final AbstractChannelHandlerContext next, Object msg) {final Object m = next.pipeline.touch(ObjectUtil.checkNotNull(msg, "msg"), next);EventExecutor executor = next.executor();if (executor.inEventLoop()) {next.invokeChannelRead(m);} else {executor.execute(new Runnable() {@Overridepublic void run() {next.invokeChannelRead(m);}});}}private void invokeChannelRead(Object msg) {if (invokeHandler()) {try {((ChannelInboundHandler) handler()).channelRead(this, msg);} catch (Throwable t) {notifyHandlerException(t);}} else {fireChannelRead(msg);}}private void notifyHandlerException(Throwable cause) {if (inExceptionCaught(cause)) {if (logger.isWarnEnabled()) {logger.warn("...", cause);}return;}invokeExceptionCaught(cause);}private void invokeExceptionCaught(final Throwable cause) {if (invokeHandler()) {try {//调用ChannelHandler的exceptionCaught()handler().exceptionCaught(this, cause);} catch (Throwable error) {if (logger.isDebugEnabled()) {logger.debug("...",ThrowableUtil.stackTraceToString(error), cause);} else if (logger.isWarnEnabled()) {logger.warn("...", error, cause);}}} else {fireExceptionCaught(cause);}}@Overridepublic ChannelHandlerContext fireExceptionCaught(final Throwable cause) {//调用下一个结点next的exceptionCaught()方法invokeExceptionCaught(next, cause);return this;}static void invokeExceptionCaught(final AbstractChannelHandlerContext next, final Throwable cause) {ObjectUtil.checkNotNull(cause, "cause");EventExecutor executor = next.executor();if (executor.inEventLoop()) {next.invokeExceptionCaught(cause);} else {try {executor.execute(new Runnable() {@Overridepublic void run() {next.invokeExceptionCaught(cause);}});} catch (Throwable t) {if (logger.isWarnEnabled()) {logger.warn("Failed to submit an exceptionCaught() event.", t);logger.warn("The exceptionCaught() event that was failed to submit was:", cause);}}}}...
}public class ChannelInboundHandlerAdapter extends ChannelHandlerAdapter implements ChannelInboundHandler {...//Calls ChannelHandlerContext#fireExceptionCaught(Throwable) to forward to the next ChannelHandler in the ChannelPipeline. @Overridepublic void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {ctx.fireExceptionCaught(cause);}
}14.ChannelPipeline总结
(1)ChannelPipeline的初始化
(2)ChannelPipeline的数据结构
(3)ChannelHandler类型的判断
(4)ChannelPipeline的头尾结点
(5)Channel与Unsafe
(6)ChannelPipeline的事件传播机制
(1)ChannelPipeline的初始化
ChannelPipeline在服务端Channel和客户端Channel被创建时创建，创建ChannelPipeline的类是服务端Channel和客户端Channel的共同父类AbstractChannel。
(2)ChannelPipeline的数据结构
ChannelPipeline中的数据结构是双向链表结构，每一个结点都是一个ChannelHandlerContext对象。ChannelHandlerContext里包装了用户自定义的ChannelHandler，即前者会保存后者的引用到其成员变量handler中。ChannelHandlerContext中拥有ChannelPipeline和Channel的所有上下文信息。添加和删除ChannelHandler最终都是在ChannelPipeline的链表结构中添加和删除对应的ChannelHandlerContext结点。
(3)ChannelHandler类型的判断
在旧版Netty中，会使用instanceof关键字来判断ChannelHandler的类型，并使用两个成员变量inbound和outbound来标识。在新版Netty中，会使用一个16位的二进制数executionMask来表示ChannelHandler具体实现的事件类型，若实现则给对应的位标1。
(4)ChannelPipeline的头尾结点
创建ChannelPipeline时会默认添加两个结点：HeadContext结点和TailContext结点。HeadContext结点的作用是作为头结点，开始传播读写事件，并且通过它的unsafe变量实现具体的读写操作。TailContext结点的作用是起到终止事件传播(方法体为空)以及异常和对象未处理的告警。
(5)Channel与Unsafe
一个Channel对应一个Unsafe，Unsafe用于处理底层IO操作。NioServerSocketChannel对应NioMessageUnsafe，NioSocketChannel对应NioByteUnsafe。
(6)ChannelPipeline的事件传播机制
ChannelPipeline中的事件传播机制分为3种：Inbound事件的传播、Outbound事件的传播、异常事件的传播。
一.Inbound事件的传播
如果通过Channel的Pipeline触发这类事件(默认情况下)，那么触发的规则是从head结点开始不断寻找下一个InboundHandler，最终落到tail结点。如果在当前ChannelHandlerContext上触发这类事件，那么事件只会从当前结点开始向下传播。
二.Outbound事件的传播
如果通过Channel的Pipeline触发这类事件(默认情况下)，那么触发的规则是从tail结点开始不断寻找上一个InboundHandler，最终落到head结点。如果在当前ChannelHandlerContext上触发这类事件，那么事件只会从当前结点开始向上传播。
三.异常事件的传播
异常在ChannelPipeline中的双向链表传播时，无论Inbound结点还是Outbound结点，都是向下一个结点传播，直到tail结点为止。TailContext结点会打印这些异常信息，最佳实践是在ChannelPipeline的最后实现异常处理方法exceptionCaught()。
