ChannelPipeline在Netty中是用来处理请求的责任链，默认实现是DefaultChannelPipeline，其构造方法如下：
1 private final Channel channel;
2 private final ChannelFuture succeededFuture;
3 private final VoidChannelPromise voidPromise;
4 final AbstractChannelHandlerContext head;
5 final AbstractChannelHandlerContext tail;
6
7 protected DefaultChannelPipeline(Channel channel) {
8 this.channel = (Channel)ObjectUtil.checkNotNull(channel, "channel");
9 this.succeededFuture = new SucceededChannelFuture(channel, (EventExecutor)null);
10 this.voidPromise = new VoidChannelPromise(channel, true);
11 this.tail = new DefaultChannelPipeline.TailContext(this);
12 this.head = new DefaultChannelPipeline.HeadContext(this);
13 this.head.next = this.tail;
14 this.tail.prev = this.head;
15 }
ChannelPipeline和Channel是一一对应关系，一个Channel绑定一条ChannelPipeline责任链
succeededFuture 和voidPromise用来处理异步操作
AbstractChannelHandlerContext 是持有请求的上下文对象，其和ChannelHandler是对应关系（在使用Sharable注解的情况下，不同的AbstractChannelHandlerContext 还可以对应同一个ChannelHandler），ChannelPipeline责任链
处理的就AbstractChannelHandlerContext ，再将最后的AbstractChannelHandlerContext 交给ChannelHandler去做正真的逻辑处理
AbstractChannelHandlerContext构造方法如下：
1 private final String name;
2 private final DefaultChannelPipeline pipeline;
3 final EventExecutor executor;
4 private final boolean inbound;
5 private final boolean outbound;
6 private final boolean ordered;
7 volatile AbstractChannelHandlerContext next;
8 volatile AbstractChannelHandlerContext prev;
9
10 AbstractChannelHandlerContext(DefaultChannelPipeline pipeline, EventExecutor executor, String name, boolean inbound, boolean outbound) {
11 this.name = (String)ObjectUtil.checkNotNull(name, "name");
12 this.pipeline = pipeline;
13 this.executor = executor;
14 this.inbound = inbound;
15 this.outbound = outbound;
16 this.ordered = executor == null || executor instanceof OrderedEventExecutor;
17 }
name是AbstractChannelHandlerContext的名称，pipeline就是上面说的ChannelPipeline；executor是用来进行异步操作的，默认使用的是在前面博客中说过的NioEventLoop （Netty中NioEventLoopGroup的创建源码分析）
inbound 和outbound 代表两种请求处理方式，对应Netty中的I/O操作，若是inbound则处理Input操作，由ChannelPipeline从head 开始向后遍历链表，并且只处理ChannelInboundHandler类型的AbstractChannelHandlerContext；若是outbound 则处理Output操作，由ChannelPipeline从tail开始向前遍历链表，并且只处理ChannelOutboundHandler类型的AbstractChannelHandlerContext；
ordered 是判断是否需要提供executor。
由next和prev成员可以知道，ChannelPipeline维护的是一条AbstractChannelHandlerContext的双向链表
其头节点head和尾节点tail分别默认初始化了HeadContext和TailContext
HeadContext的构造：
1 final class HeadContext extends AbstractChannelHandlerContext implements ChannelOutboundHandler, ChannelInboundHandler {
2 private final Unsafe unsafe;
3
4 HeadContext(DefaultChannelPipeline pipeline) {
5 super(pipeline, (EventExecutor)null, DefaultChannelPipeline.HEAD_NAME, false, true);
6 this.unsafe = pipeline.channel().unsafe();
7 this.setAddComplete();
8 }
9 }
其中setAddComplete是由AbstractChannelHandlerContext实现的：
1 final void setAddComplete() {
2 int oldState;
3 do {
4 oldState = this.handlerState;
5 } while(oldState != 3 && !HANDLER_STATE_UPDATER.compareAndSet(this, oldState, 2));
6
7 }
handlerState表示AbstractChannelHandlerContext对应的ChannelHandler的状态，有一下几种：
1 private static final int ADD_PENDING = 1; 2 private static final int ADD_COMPLETE = 2; 3 private static final int REMOVE_COMPLETE = 3; 4 private static final int INIT = 0; 5 private volatile int handlerState = 0;
handlerState初始化默认是INIT状态。
HANDLER_STATE_UPDATER是一个原子更新器：
1 private static final AtomicIntegerFieldUpdater<AbstractChannelHandlerContext> HANDLER_STATE_UPDATER = AtomicIntegerFieldUpdater.newUpdater(AbstractChannelHandlerContext.class, "handlerState");
所以setAddComplete方法，就是通过CAS操作，将handlerState状态更新为ADD_COMPLETE
TailContext的构造：
1 final class TailContext extends AbstractChannelHandlerContext implements ChannelInboundHandler {
2 TailContext(DefaultChannelPipeline pipeline) {
3 super(pipeline, (EventExecutor)null, DefaultChannelPipeline.TAIL_NAME, true, false);
4 this.setAddComplete();
5 }
6 }
和HeadContext一样，将handlerState状态更新为ADD_COMPLETE
结合官方给出的ChannelPipeline的图示更容易理解：
1 I/O Request 2 via Channel or 3 ChannelHandlerContext 4 | 5 +---------------------------------------------------+---------------+ 6 | ChannelPipeline | | 7 | \|/ | 8 | +---------------------+ +-----------+----------+ | 9 | | Inbound Handler N | | Outbound Handler 1 | | 10 | +----------+----------+ +-----------+----------+ | 11 | /|\ | | 12 | | \|/ | 13 | +----------+----------+ +-----------+----------+ | 14 | | Inbound Handler N-1 | | Outbound Handler 2 | | 15 | +----------+----------+ +-----------+----------+ | 16 | /|\ . | 17 | . . | 18 | ChannelHandlerContext.fireIN_EVT() ChannelHandlerContext.OUT_EVT()| 19 | [ method call] [method call] | 20 | . . | 21 | . \|/ | 22 | +----------+----------+ +-----------+----------+ | 23 | | Inbound Handler 2 | | Outbound Handler M-1 | | 24 | +----------+----------+ +-----------+----------+ | 25 | /|\ | | 26 | | \|/ | 27 | +----------+----------+ +-----------+----------+ | 28 | | Inbound Handler 1 | | Outbound Handler M | | 29 | +----------+----------+ +-----------+----------+ | 30 | /|\ | | 31 +---------------+-----------------------------------+---------------+ 32 | \|/ 33 +---------------+-----------------------------------+---------------+ 34 | | | | 35 | [ Socket.read() ] [ Socket.write() ] | 36 | | 37 | Netty Internal I/O Threads (Transport Implementation) | 38 +-------------------------------------------------------------------+
下面对一些主要方法分析：
addFirst方法，有如下几种重载：
1 public final ChannelPipeline addFirst(ChannelHandler handler) {
2 return this.addFirst((String)null, (ChannelHandler)handler);
3 }
4
5 public final ChannelPipeline addFirst(String name, ChannelHandler handler) {
6 return this.addFirst((EventExecutorGroup)null, name, handler);
7 }
8
9 public final ChannelPipeline addFirst(ChannelHandler... handlers) {
10 return this.addFirst((EventExecutorGroup)null, (ChannelHandler[])handlers);
11 }
12
13 public final ChannelPipeline addFirst(EventExecutorGroup executor, ChannelHandler... handlers) {
14 if (handlers == null) {
15 throw new NullPointerException("handlers");
16 } else if (handlers.length != 0 && handlers[0] != null) {
17 int size;
18 for(size = 1; size < handlers.length && handlers[size] != null; ++size) {
19 ;
20 }
21
22 for(int i = size - 1; i >= 0; --i) {
23 ChannelHandler h = handlers[i];
24 this.addFirst(executor, (String)null, h);
25 }
26
27 return this;
28 } else {
29 return this;
30 }
31 }
32
33 public final ChannelPipeline addFirst(EventExecutorGroup group, String name, ChannelHandler handler) {
34 final AbstractChannelHandlerContext newCtx;
35 synchronized(this) {
36 checkMultiplicity(handler);
37 name = this.filterName(name, handler);
38 newCtx = this.newContext(group, name, handler);
39 this.addFirst0(newCtx);
40 if (!this.registered) {
41 newCtx.setAddPending();
42 this.callHandlerCallbackLater(newCtx, true);
43 return this;
44 }
45
46 EventExecutor executor = newCtx.executor();
47 if (!executor.inEventLoop()) {
48 newCtx.setAddPending();
49 executor.execute(new Runnable() {
50 public void run() {
51 DefaultChannelPipeline.this.callHandlerAdded0(newCtx);
52 }
53 });
54 return this;
55 }
56 }
57
58 this.callHandlerAdded0(newCtx);
59 return this;
60 }
前面几种都是间接调用的第四种没什么好说的，直接看第四种addFirst
首先调用checkMultiplicity，检查ChannelHandlerAdapter在不共享的情况下是否重复：
1 private static void checkMultiplicity(ChannelHandler handler) {
2 if (handler instanceof ChannelHandlerAdapter) {
3 ChannelHandlerAdapter h = (ChannelHandlerAdapter)handler;
4 if (!h.isSharable() && h.added) {
5 throw new ChannelPipelineException(h.getClass().getName() + " is not a @Sharable handler, so can‘t be added or removed multiple times.");
6 }
7
8 h.added = true;
9 }
10
11 }
isSharable方法：
1 public boolean isSharable() {
2 Class<?> clazz = this.getClass();
3 Map<Class<?>, Boolean> cache = InternalThreadLocalMap.get().handlerSharableCache();
4 Boolean sharable = (Boolean)cache.get(clazz);
5 if (sharable == null) {
6 sharable = clazz.isAnnotationPresent(Sharable.class);
7 cache.put(clazz, sharable);
8 }
9
10 return sharable;
11 }
首先尝试从当前线程的InternalThreadLocalMap中获取handlerSharableCache，（InternalThreadLocalMap是在Netty中使用高效的FastThreadLocal替代JDK的ThreadLocal使用的 Netty中FastThreadLocal源码分析）
InternalThreadLocalMap的handlerSharableCache方法：
1 public Map<Class<?>, Boolean> handlerSharableCache() {
2 Map<Class<?>, Boolean> cache = this.handlerSharableCache;
3 if (cache == null) {
4 this.handlerSharableCache = (Map)(cache = new WeakHashMap(4));
5 }
6
7 return (Map)cache;
8 }
当当前线程的InternalThreadLocalMap中没有handlerSharableCache时，直接创建一个大小为4的WeakHashMap弱引用Map；
根据clazz从map中get，若是没有，需要检测当前clazz是否有Sharable注解，添加了Sharable注解的ChannelHandlerAdapter可以在不同Channel中共享使用一个单例，前提是确保线程安全；
之后会将该clazz以及是否实现Sharable注解的情况添加在cache缓存中；
其中ChannelHandler的added是用来标识是否添加过；
回到addFirst方法：
checkMultiplicity成功结束后，调用filterName方法，给当前要产生的AbstractChannelHandlerContext对象产生一个名称，
然后调用newContext方法，产生AbstractChannelHandlerContext对象：
1 private AbstractChannelHandlerContext newContext(EventExecutorGroup group, String name, ChannelHandler handler) {
2 return new DefaultChannelHandlerContext(this, this.childExecutor(group), name, handler);
3 }
这里实际上产生了一个DefaultChannelHandlerContext对象：
1 final class DefaultChannelHandlerContext extends AbstractChannelHandlerContext {
2 private final ChannelHandler handler;
3
4 DefaultChannelHandlerContext(DefaultChannelPipeline pipeline, EventExecutor executor, String name, ChannelHandler handler) {
5 super(pipeline, executor, name, isInbound(handler), isOutbound(handler));
6 if (handler == null) {
7 throw new NullPointerException("handler");
8 } else {
9 this.handler = handler;
10 }
11 }
12
13 public ChannelHandler handler() {
14 return this.handler;
15 }
16
17 private static boolean isInbound(ChannelHandler handler) {
18 return handler instanceof ChannelInboundHandler;
19 }
20
21 private static boolean isOutbound(ChannelHandler handler) {
22 return handler instanceof ChannelOutboundHandler;
23 }
24 }
可以看到DefaultChannelHandlerContext 仅仅是将AbstractChannelHandlerContext和ChannelHandler封装了
在产生了DefaultChannelHandlerContext 对象后，调用addFirst0方法：
1 private void addFirst0(AbstractChannelHandlerContext newCtx) {
2 AbstractChannelHandlerContext nextCtx = this.head.next;
3 newCtx.prev = this.head;
4 newCtx.next = nextCtx;
5 this.head.next = newCtx;
6 nextCtx.prev = newCtx;
7 }
这里就是一个简单的双向链表的操作，将newCtx节点插入到了head后面
然后判断registered成员的状态：
1 private boolean registered;
在初始化时是false
registered若是false，首先调用AbstractChannelHandlerContext的setAddPending方法：
1 final void setAddPending() {
2 boolean updated = HANDLER_STATE_UPDATER.compareAndSet(this, 0, 1);
3
4 assert updated;
5
6 }
和前面说过的setAddComplete方法同理，通过CAS操作，将handlerState状态设置为ADD_PENDING
接着调用callHandlerCallbackLater方法：
1 private void callHandlerCallbackLater(AbstractChannelHandlerContext ctx, boolean added) {
2 assert !this.registered;
3
4 DefaultChannelPipeline.PendingHandlerCallback task = added ? new DefaultChannelPipeline.PendingHandlerAddedTask(ctx) : new DefaultChannelPipeline.PendingHandlerRemovedTask(ctx);
5 DefaultChannelPipeline.PendingHandlerCallback pending = this.pendingHandlerCallbackHead;
6 if (pending == null) {
7 this.pendingHandlerCallbackHead = (DefaultChannelPipeline.PendingHandlerCallback)task;
8 } else {
9 while(pending.next != null) {
10 pending = pending.next;
11 }
12
13 pending.next = (DefaultChannelPipeline.PendingHandlerCallback)task;
14 }
15
16 }
首先断言判断registered可能存在的多线程改变，然后根据added判断产生何种类型的PendingHandlerCallback
PendingHandlerCallback是用来处理ChannelHandler的两种回调，定义如下：
1 private abstract static class PendingHandlerCallback implements Runnable {
2 final AbstractChannelHandlerContext ctx;
3 DefaultChannelPipeline.PendingHandlerCallback next;
4
5 PendingHandlerCallback(AbstractChannelHandlerContext ctx) {
6 this.ctx = ctx;
7 }
8
9 abstract void execute();
10 }
PendingHandlerAddedTask定义如下：
1 private final class PendingHandlerAddedTask extends DefaultChannelPipeline.PendingHandlerCallback {
2 PendingHandlerAddedTask(AbstractChannelHandlerContext ctx) {
3 super(ctx);
4 }
5
6 public void run() {
7 DefaultChannelPipeline.this.callHandlerAdded0(this.ctx);
8 }
9
10 void execute() {
11 EventExecutor executor = this.ctx.executor();
12 if (executor.inEventLoop()) {
13 DefaultChannelPipeline.this.callHandlerAdded0(this.ctx);
14 } else {
15 try {
16 executor.execute(this);
17 } catch (RejectedExecutionException var3) {
18 if (DefaultChannelPipeline.logger.isWarnEnabled()) {
19 DefaultChannelPipeline.logger.warn("Can‘t invoke handlerAdded() as the EventExecutor {} rejected it, removing handler {}.", new Object[]{executor, this.ctx.name(), var3});
20 }
21
22 DefaultChannelPipeline.remove0(this.ctx);
23 this.ctx.setRemoved();
24 }
25 }
26
27 }
28 }
除去异常处理，无论是在execute方法还是在run方法中，主要核心是异步执行callHandlerAdded0方法：
1 private void callHandlerAdded0(AbstractChannelHandlerContext ctx) {
2 try {
3 ctx.setAddComplete();
4 ctx.handler().handlerAdded(ctx);
5 } catch (Throwable var10) {
6 boolean removed = false;
7
8 try {
9 remove0(ctx);
10
11 try {
12 ctx.handler().handlerRemoved(ctx);
13 } finally {
14 ctx.setRemoved();
15 }
16
17 removed = true;
18 } catch (Throwable var9) {
19 if (logger.isWarnEnabled()) {
20 logger.warn("Failed to remove a handler: " + ctx.name(), var9);
21 }
22 }
23
24 if (removed) {
25 this.fireExceptionCaught(new ChannelPipelineException(ctx.handler().getClass().getName() + ".handlerAdded() has thrown an exception; removed.", var10));
26 } else {
27 this.fireExceptionCaught(new ChannelPipelineException(ctx.handler().getClass().getName() + ".handlerAdded() has thrown an exception; also failed to remove.", var10));
28 }
29 }
30
31 }
除去异常处理，主要核心就两行代码，首先通过setAddComplete方法，设置handlerState状态为ADD_COMPLETE，然后回调ChannelHandler的handlerAdded方法，这个handlerAdded方法就很熟悉了，在使用Netty处理业务逻辑时，会覆盖这个方法。
PendingHandlerRemovedTask定义如下：
1 private final class PendingHandlerRemovedTask extends DefaultChannelPipeline.PendingHandlerCallback {
2 PendingHandlerRemovedTask(AbstractChannelHandlerContext ctx) {
3 super(ctx);
4 }
5
6 public void run() {
7 DefaultChannelPipeline.this.callHandlerRemoved0(this.ctx);
8 }
9
10 void execute() {
11 EventExecutor executor = this.ctx.executor();
12 if (executor.inEventLoop()) {
13 DefaultChannelPipeline.this.callHandlerRemoved0(this.ctx);
14 } else {
15 try {
16 executor.execute(this);
17 } catch (RejectedExecutionException var3) {
18 if (DefaultChannelPipeline.logger.isWarnEnabled()) {
19 DefaultChannelPipeline.logger.warn("Can‘t invoke handlerRemoved() as the EventExecutor {} rejected it, removing handler {}.", new Object[]{executor, this.ctx.name(), var3});
20 }
21
22 this.ctx.setRemoved();
23 }
24 }
25
26 }
27 }
和PendingHandlerAddedTask一样，主要还是异步调用callHandlerRemoved0方法：
1 private void callHandlerRemoved0(AbstractChannelHandlerContext ctx) {
2 try {
3 try {
4 ctx.handler().handlerRemoved(ctx);
5 } finally {
6 ctx.setRemoved();
7 }
8 } catch (Throwable var6) {
9 this.fireExceptionCaught(new ChannelPipelineException(ctx.handler().getClass().getName() + ".handlerRemoved() has thrown an exception.", var6));
10 }
11
12 }
首先直接回调ChannelHandler的handlerRemoved方法，然后通过setRemoved方法将handlerState状态设置为REMOVE_COMPLETE
回到callHandlerCallbackLater，其中成员pendingHandlerCallbackHead定义：
1 private DefaultChannelPipeline.PendingHandlerCallback pendingHandlerCallbackHead;
结合PendingHandlerCallback 可知，这个pendingHandlerCallbackHead是 DefaultChannelPipeline存储的一条PendingHandlerCallback单链表，用来处理ChannelHandler的handlerAdded和handlerRemoved的回调，在add的这些方法里调用callHandlerCallbackLater时，added参数都为true，所以add的ChannelHandler只向pendingHandlerCallbackHead添加了handlerAdded的回调。
回到addFirst方法，若是registered为true，先获取EventExecutor，判断是否处于轮询中，若不是，则需要开启轮询线程直接异步执行callHandlerAdded0方法，若处于轮询，由于ChannelPipeline的调用是发生在轮询时的，所以还是直接异步执行callHandlerAdded0方法。
addFirst方法到此结束，再来看addLast方法，同样有好几种重载：
1 public final ChannelPipeline addLast(ChannelHandler handler) {
2 return this.addLast((String)null, (ChannelHandler)handler);
3 }
4
5 public final ChannelPipeline addLast(String name, ChannelHandler handler) {
6 return this.addLast((EventExecutorGroup)null, name, handler);
7 }
8
9 public final ChannelPipeline addLast(ChannelHandler... handlers) {
10 return this.addLast((EventExecutorGroup)null, (ChannelHandler[])handlers);
11 }
12
13 public final ChannelPipeline addLast(EventExecutorGroup executor, ChannelHandler... handlers) {
14 if (handlers == null) {
15 throw new NullPointerException("handlers");
16 } else {
17 ChannelHandler[] var3 = handlers;
18 int var4 = handlers.length;
19
20 for(int var5 = 0; var5 < var4; ++var5) {
21 ChannelHandler h = var3[var5];
22 if (h == null) {
23 break;
24 }
25
26 this.addLast(executor, (String)null, h);
27 }
28
29 return this;
30 }
31 }
32
33 public final ChannelPipeline addLast(EventExecutorGroup group, String name, ChannelHandler handler) {
34 final AbstractChannelHandlerContext newCtx;
35 synchronized(this) {
36 checkMultiplicity(handler);
37 newCtx = this.newContext(group, this.filterName(name, handler), handler);
38 this.addLast0(newCtx);
39 if (!this.registered) {
40 newCtx.setAddPending();
41 this.callHandlerCallbackLater(newCtx, true);
42 return this;
43 }
44
45 EventExecutor executor = newCtx.executor();
46 if (!executor.inEventLoop()) {
47 newCtx.setAddPending();
48 executor.execute(new Runnable() {
49 public void run() {
50 DefaultChannelPipeline.this.callHandlerAdded0(newCtx);
51 }
52 });
53 return this;
54 }
55 }
56
57 this.callHandlerAdded0(newCtx);
58 return this;
59 }
还是间接调用最后一种：
对比addFirst来看，只有addLast0不一样：
1 private void addLast0(AbstractChannelHandlerContext newCtx) {
2 AbstractChannelHandlerContext prev = this.tail.prev;
3 newCtx.prev = prev;
4 newCtx.next = this.tail;
5 prev.next = newCtx;
6 this.tail.prev = newCtx;
7 }
还是非常简单的双向链表基本操作，只不过这次，是将AbstractChannelHandlerContext插入到了tail之前
还有两个，addBefore和addAfter方法，和上述方法类似，就不再累赘
接下来看看ChannelPipeline是如何完成请求的传递的：
invokeHandlerAddedIfNeeded方法：
1 final void invokeHandlerAddedIfNeeded() {
2 assert this.channel.eventLoop().inEventLoop();
3
4 if (this.firstRegistration) {
5 this.firstRegistration = false;
6 this.callHandlerAddedForAllHandlers();
7 }
8
9 }
断言判断是否处于轮询线程（ChannelPipeline处理请求都是在轮询线程中，都需要异步处理）
其中firstRegistration成员在DefaultChannelPipeline初始化时为true：
1 private boolean firstRegistration = true;
此时设置为false，表示第一次调用，以后都不再调用后面的callHandlerAddedForAllHandlers：
1 private void callHandlerAddedForAllHandlers() {
2 DefaultChannelPipeline.PendingHandlerCallback pendingHandlerCallbackHead;
3 synchronized(this) {
4 assert !this.registered;
5
6 this.registered = true;
7 pendingHandlerCallbackHead = this.pendingHandlerCallbackHead;
8 this.pendingHandlerCallbackHead = null;
9 }
10
11 for(DefaultChannelPipeline.PendingHandlerCallback task = pendingHandlerCallbackHead; task != null; task = task.next) {
12 task.execute();
13 }
14
15 }
刚才说过registered初始是false，在这里判断符合，之后就令其为true，然后获取处理ChannelHandler的回调链表pendingHandlerCallbackHead，并且将pendingHandlerCallbackHead置为null
然后遍历这个单链表，处理ChannelHandler的handlerAdded和handlerRemoved的回调
fireChannelRegistered方法，当Channel完成了向Selector的注册后，会由channel的Unsafe进行回调，异步处理：
1 public final ChannelPipeline fireChannelRegistered() {
2 AbstractChannelHandlerContext.invokeChannelRegistered(this.head);
3 return this;
4 }
实际上的处理由AbstractChannelHandlerContext的静态方法invokeChannelRegistered完成，这里传递的参数head就是DefaultChannelPipeline初始化时创建的HeadContext：
1 static void invokeChannelRegistered(final AbstractChannelHandlerContext next) {
2 EventExecutor executor = next.executor();
3 if (executor.inEventLoop()) {
4 next.invokeChannelRegistered();
5 } else {
6 executor.execute(new Runnable() {
7 public void run() {
8 next.invokeChannelRegistered();
9 }
10 });
11 }
12
13 }
可以看到实际上是异步执行head对象的invokeChannelRegistered方法：
1 private void invokeChannelRegistered() {
2 if (this.invokeHandler()) {
3 try {
4 ((ChannelInboundHandler)this.handler()).channelRegistered(this);
5 } catch (Throwable var2) {
6 this.notifyHandlerException(var2);
7 }
8 } else {
9 this.fireChannelRegistered();
10 }
11
12 }
其中invokeHandler是用来判断当前的handlerState状态：
1 private boolean invokeHandler() {
2 int handlerState = this.handlerState;
3 return handlerState == 2 || !this.ordered && handlerState == 1;
4 }
若是当前handlerState状态为ADD_COMPLETE，或者不需要提供EventExecutor并且状态为ADD_PENDING时返回true，否则返回false
在成立的情况下，调用ChannelInboundHandler的channelRegistered方法，由于当前是head，所以由HeadContext实现了：
1 public void channelRegistered(ChannelHandlerContext ctx) throws Exception {
2 DefaultChannelPipeline.this.invokeHandlerAddedIfNeeded();
3 ctx.fireChannelRegistered();
4 }
首先调用invokeHandlerAddedIfNeeded，处理ChannelHandler的handlerAdded和handlerRemoved的回调
然后调用ctx的fireChannelRegistered方法：
1 public ChannelHandlerContext fireChannelRegistered() {
2 invokeChannelRegistered(this.findContextInbound());
3 return this;
4 }
findContextInbound方法，用来找出下一个ChannelInboundInvoker：
1 private AbstractChannelHandlerContext findContextInbound() {
2 AbstractChannelHandlerContext ctx = this;
3
4 do {
5 ctx = ctx.next;
6 } while(!ctx.inbound);
7
8 return ctx;
9 }
从当前节点向后遍历，inbound之前说过，该方法就是找到下一个ChannelInboundInvoker的类型的AbstractChannelHandlerContext，然后调用静态方法invokeChannelRegistered，重复上述操作，若是在ChannelInboundHandler中没有重写channelRegistered方法，会一直执直到完所有ChannelHandler的channelRegistered方法。
ChannelInboundHandlerAdapter中的默认channelRegistered方法：
1 public void channelRegistered(ChannelHandlerContext ctx) throws Exception {
2 ctx.fireChannelRegistered();
3 }
比HeadContext中的实现还简单，直接调用fireChannelRegistered向后传递
fireChannelRead方法，是在Selector轮循到读事件就绪，会由channel的Unsafe进行回调，异步处理：
1 public final ChannelPipeline fireChannelRead(Object msg) {
2 AbstractChannelHandlerContext.invokeChannelRead(this.head, msg);
3 return this;
4 }
还是从head开始调用AbstractChannelHandlerContext的静态方法invokeChannelRead：
1 static void invokeChannelRead(final AbstractChannelHandlerContext next, Object msg) {
2 final Object m = next.pipeline.touch(ObjectUtil.checkNotNull(msg, "msg"), next);
3 EventExecutor executor = next.executor();
4 if (executor.inEventLoop()) {
5 next.invokeChannelRead(m);
6 } else {
7 executor.execute(new Runnable() {
8 public void run() {
9 next.invokeChannelRead(m);
10 }
11 });
12 }
13
14 }
和上面一个逻辑异步调用AbstractChannelHandlerContext对象的invokeChannelRead方法：
1 private void invokeChannelRead(Object msg) {
2 if (this.invokeHandler()) {
3 try {
4 ((ChannelInboundHandler)this.handler()).channelRead(this, msg);
5 } catch (Throwable var3) {
6 this.notifyHandlerException(var3);
7 }
8 } else {
9 this.fireChannelRead(msg);
10 }
11
12 }
这里也和上面一样，调用了HeadContext的channelRead方法：
1 public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
2 ctx.fireChannelRead(msg);
3 }
这里直接不处理，调用ChannelHandlerContext 的fireChannelRead方法：
1 public ChannelHandlerContext fireChannelRead(Object msg) {
2 invokeChannelRead(this.findContextInbound(), msg);
3 return this;
4 }
和之前注册一样，选择下一个ChannelInboundHandler，重复执行上述操作。
再来看到writeAndFlush方法，和上面的就不太一样，这个发生在轮询前，用户通过channel来间接调用，在AbstractChannel中实现：
1 public ChannelFuture writeAndFlush(Object msg) {
2 return this.pipeline.writeAndFlush(msg);
3 }
实际上直接调用了DefaultChannelPipeline的writeAndFlush方法：
1 public final ChannelFuture writeAndFlush(Object msg) {
2 return this.tail.writeAndFlush(msg);
3 }
这里又有些不一样了，调用了tail的writeAndFlush方法，即TailContext的writeAndFlush，在AbstractChannelHandlerContext中实现：
1 public ChannelFuture writeAndFlush(Object msg) {
2 return this.writeAndFlush(msg, this.newPromise());
3 }
newPromise产生了一个ChannelPromise，用来处理异步事件的；实际上调用了writeAndFlush的重载：
1 public ChannelFuture writeAndFlush(Object msg, ChannelPromise promise) {
2 if (msg == null) {
3 throw new NullPointerException("msg");
4 } else if (this.isNotValidPromise(promise, true)) {
5 ReferenceCountUtil.release(msg);
6 return promise;
7 } else {
8 this.write(msg, true, promise);
9 return promise;
10 }
11 }
继续调用write方法：
1 private void write(Object msg, boolean flush, ChannelPromise promise) {
2 AbstractChannelHandlerContext next = this.findContextOutbound();
3 Object m = this.pipeline.touch(msg, next);
4 EventExecutor executor = next.executor();
5 if (executor.inEventLoop()) {
6 if (flush) {
7 next.invokeWriteAndFlush(m, promise);
8 } else {
9 next.invokeWrite(m, promise);
10 }
11 } else {
12 Object task;
13 if (flush) {
14 task = AbstractChannelHandlerContext.WriteAndFlushTask.newInstance(next, m, promise);
15 } else {
16 task = AbstractChannelHandlerContext.WriteTask.newInstance(next, m, promise);
17 }
18
19 safeExecute(executor, (Runnable)task, promise, m);
20 }
21
22 }
还是很相似，只不过先调用findContextOutbound找到下一个ChannelOutboundInvoker类型的ChannelHandlerContext，而且这里是从尾部往前遍历的，这样来看前面所给的图是没有任何问题的
在找到ChannelOutboundInvoker后，调用invokeWriteAndFlush或者invokeWrite方法：
invokeWriteAndFlush方法：
1 private void invokeWriteAndFlush(Object msg, ChannelPromise promise) {
2 if (this.invokeHandler()) {
3 this.invokeWrite0(msg, promise);
4 this.invokeFlush0();
5 } else {
6 this.writeAndFlush(msg, promise);
7 }
8
9 }
10
11 private void invokeWrite0(Object msg, ChannelPromise promise) {
12 try {
13 ((ChannelOutboundHandler)this.handler()).write(this, msg, promise);
14 } catch (Throwable var4) {
15 notifyOutboundHandlerException(var4, promise);
16 }
17
18 }
19
20 private void invokeFlush0() {
21 try {
22 ((ChannelOutboundHandler)this.handler()).flush(this);
23 } catch (Throwable var2) {
24 this.notifyHandlerException(var2);
25 }
26
27 }
可以看到invokeWriteAndFlush回调了ChannelOutboundHandler的write和flush方法
最终会调用HeadContext的write和flush方法：
1 public void write(ChannelHandlerContext ctx, Object msg, ChannelPromise promise) throws Exception {
2 this.unsafe.write(msg, promise);
3 }
4
5 public void flush(ChannelHandlerContext ctx) throws Exception {
6 this.unsafe.flush();
7 }
可以看到调用了unsafe的write和flush方法，向unsafe缓冲区写入了消息，当Selector轮询到写事件就绪时，就会通过unsafe将刚才写入的内容交由JDK的SocketChannel完成最终的write操作。
ChannelPipeline的分析到此全部结束。
原文地址：https://www.cnblogs.com/a526583280/p/10961855.html