Netty5源码分析(三) -- Channel如何注册OP_ACCEPT, OP_READ, OP_WRITE

抛开Netty,一个典型的Java NIO服务端开发需要做几件事:

1.  创建ServerSocketChannel,设置为非阻塞,并绑定端口

2. 创建Selector对象

3. 给ServerSocketChannel注册SelectionKey.OP_ACCEPT事件

4. 启动一个线程循环,调用Selector的select方法来检查IO就绪事件,一旦有IO就绪事件,就通知用户线程去处理IO事件

5. 如果有Accept事件,就创建一个SocketChannel,并注册SelectionKey_OP_READ

6.    如果有读事件,判断一下是否全包,如果全包,就交给后端线程处理

7.    写事件比较特殊。isWriteable表示的是本机的写缓冲区是否可写。这个在绝大多少情况下都是为真的。在Netty中只有写半包的时候才需要注册写事件,如果一次写就完全把数据写入了缓冲区就不需要注册写事件. 具体关于写事件的处理看这篇 http://blog.csdn.net/iter_zc/article/details/39291129


一个典型的Java NIO服务端的代码如下:


public NIOServer(int port) throws IOException {  
        ServerSocketChannel serverSocketChannel = ServerSocketChannel.open();  
        serverSocketChannel.configureBlocking(false);   
        ServerSocket serverSocket = serverSocketChannel.socket();  
        serverSocket.bind(port);  
        selector = Selector.open();  
        // 注册到selector,等待连接  
        serverSocketChannel.register(selector, SelectionKey.OP_ACCEPT);  
    }  



private void eventloop() throws IOException {  
        while (true) {   
            selector.select();   
            Set<SelectionKey> selectionKeys = selector.selectedKeys();  
            Iterator<SelectionKey> iterator = selectionKeys.iterator();  
            while (iterator.hasNext()) {          
                SelectionKey selectionKey = iterator.next();  
                iterator.remove();  
                handleSelectedKey(selectionKey);  
            }  
        }  
    }  

private void handleKey(SelectionKey selectionKey) throws IOException {  
        ServerSocketChannel server = null;  
        SocketChannel client = null;  
 
        if (selectionKey.isAcceptable()) {  
            server = (ServerSocketChannel) selectionKey.channel();    
            client = server.accept();  
            // 配置为非阻塞  
            client.configureBlocking(false);  
            // 注册到selector,等待连接  
            client.register(selector, SelectionKey.OP_READ);  
        } else if (selectionKey.isReadable()) {  
            client = (SocketChannel) selectionKey.channel();  
            client.read(receivebuffer);   
        } else if (selectionKey.isWritable()) {   
            sendbuffer.clear();    
            client = (SocketChannel) selectionKey.channel();  
            client.write(sendbuffer);  
        }  
    }  

那么Netty是如何来注册Accept和读写事件的呢? 首先来看如何注册的OP_ACCPET事件,是在服务器端绑定端口的过程中的,具体的绑定细节看这篇 http://blog.csdn.net/iter_zc/article/details/39349177 这里只分析注册OP_ACCEPT的过程。

在ServerBootstrap调用createChannel来创建NioServerSocketChannel的时候,会给NioServerSocketChannel传入要绑定的OP_ACCEPT事件,然后依次调用父类的构造函数,把readInterestOp往上传递,直到传递到AbstractNioChannel,作为AbstractNioChannel的一个实例属性readInterestOp.

public NioServerSocketChannel(EventLoop eventLoop, EventLoopGroup childGroup) {
        super(null, eventLoop, childGroup, newSocket(), SelectionKey.OP_ACCEPT);
        config = new DefaultServerSocketChannelConfig(this, javaChannel().socket());
    }

protected AbstractNioMessageServerChannel(
            Channel parent, EventLoop eventLoop, EventLoopGroup childGroup, SelectableChannel ch, int readInterestOp) {
        super(parent, eventLoop, ch, readInterestOp);
        this.childGroup = childGroup;
    }

protected AbstractNioMessageChannel(
            Channel parent, EventLoop eventLoop, SelectableChannel ch, int readInterestOp) {
        super(parent, eventLoop, ch, readInterestOp);
    }

protected AbstractNioChannel(Channel parent, EventLoop eventLoop, SelectableChannel ch, int readInterestOp) {
        super(parent, eventLoop);
        this.ch = ch;
        this.readInterestOp = readInterestOp;
        try {
            ch.configureBlocking(false);
        } catch (IOException e) {
            try {
                ch.close();
            } catch (IOException e2) {
                if (logger.isWarnEnabled()) {
                    logger.warn(
                            "Failed to close a partially initialized socket.", e2);
                }
            }

            throw new ChannelException("Failed to enter non-blocking mode.", e);
        }
    }

可以看到在AbstractNioChannel的构造函数里面并没有把Channel注册到selector之上,只是把OP_ACCEPT赋值给了readInterestOp。在服务器绑定的过程中有一个register的过程,最后调用到了AbstractNioChannel的doRegister方法。这里可以看到调用的ServerSocketChannel得register方法,但是并没有注册具体的事件,传了一个0,表示不注册任何事件。这里做了一件事情就是把Channel作为Attachment绑定到了SelectionKey之上。

protected void doRegister() throws Exception {
        boolean selected = false;
        for (;;) {
            try {
                selectionKey = javaChannel().register(eventLoop().selector, 0, this);
                return;
            } catch (CancelledKeyException e) {
                if (!selected) {
                    // Force the Selector to select now as the "canceled" SelectionKey may still be
                    // cached and not removed because no Select.select(..) operation was called yet.
                    eventLoop().selectNow();
                    selected = true;
                } else {
                    // We forced a select operation on the selector before but the SelectionKey is still cached
                    // for whatever reason. JDK bug ?
                    throw e;
                }
            }
        }
    }

// SelectionKey的事件的值

public static final int OP_READ = 1 << 0;  // 1
public static final int OP_WRITE = 1 << 2  // 4
public static final int OP_CONNECT = 1 << 3; // 8
public static final int OP_ACCEPT = 1 << 4;  // 16



Channel注册到Selector有两种方式,一种是调用Channel的register方法,第二种是设置SelectionKey的interestOps的值。Netty是用了第二种方式,通过设置SelectionKey的interestOps来注册Channel关心的事件,把实际的注册延迟了。

在服务器端绑定的过程中,在bind完之后会fireChannelActive事件,在fireChannelActive执行之后,会让Channel主动发起一个outbound的读事件,通过Pipeline最后传递到HeadHandler,HeadHandler会实际使用Unsafe来进行读写。HeadHandler处理读时间时,调用了Unsafe的beginRead方法,最后到了AbstractNioChannel来设置SelectionKey的interestOps属性,来最终把Channel绑定到Selector。

 public final void bind(final SocketAddress localAddress, final ChannelPromise promise) {
            if (!ensureOpen(promise)) {
                return;
            }

            // See: https://github.com/netty/netty/issues/576
            if (!PlatformDependent.isWindows() && !PlatformDependent.isRoot() &&
                Boolean.TRUE.equals(config().getOption(ChannelOption.SO_BROADCAST)) &&
                localAddress instanceof InetSocketAddress &&
                !((InetSocketAddress) localAddress).getAddress().isAnyLocalAddress()) {
                // Warn a user about the fact that a non-root user can't receive a
                // broadcast packet on *nix if the socket is bound on non-wildcard address.
                logger.warn(
                        "A non-root user can't receive a broadcast packet if the socket " +
                        "is not bound to a wildcard address; binding to a non-wildcard " +
                        "address (" + localAddress + ") anyway as requested.");
            }

            boolean wasActive = isActive();
            try {
                doBind(localAddress);
            } catch (Throwable t) {
                promise.setFailure(t);
                closeIfClosed();
                return;
            }
            if (!wasActive && isActive()) {
                invokeLater(new Runnable() {
                    @Override
                    public void run() {
                        pipeline.fireChannelActive();
                    }
                });
            }
            promise.setSuccess();
        }

public ChannelPipeline fireChannelActive() {
        head.fireChannelActive();

        if (channel.config().isAutoRead()) {
        //发出outbound读事件
            channel.read();
        }

        return this;
    }

       // HeadHandler.read
 public void read(ChannelHandlerContext ctx) {
            unsafe.beginRead();
        }

// 最后调用到AbstractNioChannel的doBeginRead方法来设置SelectionKey,此时的readInterestOps是OP_ACCEPT = 16,selectionKey.interestOps=0
// 所以进入到 selectionKey.interestOps(interestOps | readInterestOp); 使用或操作设置selectionKey的interestOps为OP_ACCEPT

@Override
    protected void doBeginRead() throws Exception {
        if (inputShutdown) {
            return;
        }

        final SelectionKey selectionKey = this.selectionKey;
        if (!selectionKey.isValid()) {
            return;
        }

        final int interestOps = selectionKey.interestOps();
        if ((interestOps & readInterestOp) == 0) {
            selectionKey.interestOps(interestOps | readInterestOp);
        }
    }

再看OP_READ事件的注册过程。

1. 当有客户端连接请求过来时,前端的NioEventLoop会通过Selector找到响应的SelectionKey,Netty把OP_READ和OP_ACCEPT统一使用Unsafe的read来处理。

2. 对于OP_ACCEPT事件,调用的是NioMessageUnsafe, 先会执行到NioServerSocketChannel的doReadMessage来处理连接请求

3. 当NioServerSocketChannel通过调用ServerSocketChannel的accept来创建客户端SocketChannel时,会传递OP_READ事件到NioSocketChannel的构造函数,最后到了AbstractNioChannel的构造函数,和OP_ACCEPT一样,成为AbstractNioChannel的readInterestOps属性。

4. 处理完之后NioMessageUnsafe会fireChannelRead来使用Pipeline传递事件,调用到ServerBootstrapAcceptor的channelRead方法来把Channel注册到selector,也只是注册了一个0到selector,把Channel作为Attachment绑定到SelectionKey。

RE5. 和OP_ACCEPT一样,注册完之后会发出fireChannelActive事件,最后调用HeadHandler的read方法,在这个方法里面调用Unsafe的beginRead方法。这里的Unsafe的实例是NioByteUnsafe,最后又进入到AbstractNioChanel的doReadBegin来设置Selectionkey的interestOps为OP_READ


// NioEventLoop处理事件分发

 private static void processSelectedKey(SelectionKey k, AbstractNioChannel ch) {
        final NioUnsafe unsafe = ch.unsafe();
        if (!k.isValid()) {
            // close the channel if the key is not valid anymore
            unsafe.close(unsafe.voidPromise());
            return;
        }

        try {
            int readyOps = k.readyOps();
            // Also check for readOps of 0 to workaround possible JDK bug which may otherwise lead
            // to a spin loop
            if ((readyOps & (SelectionKey.OP_READ | SelectionKey.OP_ACCEPT)) != 0 || readyOps == 0) {
                unsafe.read();
                if (!ch.isOpen()) {
                    // Connection already closed - no need to handle write.
                    return;
                }
            }
            if ((readyOps & SelectionKey.OP_WRITE) != 0) {
                // Call forceFlush which will also take care of clear the OP_WRITE once there is nothing left to write
                ch.unsafe().forceFlush();
            }
            if ((readyOps & SelectionKey.OP_CONNECT) != 0) {
                // remove OP_CONNECT as otherwise Selector.select(..) will always return without blocking
                // See https://github.com/netty/netty/issues/924
                int ops = k.interestOps();
                ops &= ~SelectionKey.OP_CONNECT;
                k.interestOps(ops);

                unsafe.finishConnect();
            }
        } catch (CancelledKeyException e) {
            unsafe.close(unsafe.voidPromise());
        }
    }


public void read() {
            assert eventLoop().inEventLoop();
            if (!config().isAutoRead()) {
                removeReadOp();
            }

            final ChannelConfig config = config();
            final int maxMessagesPerRead = config.getMaxMessagesPerRead();
            final boolean autoRead = config.isAutoRead();
            final ChannelPipeline pipeline = pipeline();
            boolean closed = false;
            Throwable exception = null;
            try {
                for (;;) {
                    int localRead = doReadMessages(readBuf);
                    if (localRead == 0) {
                        break;
                    }
                    if (localRead < 0) {
                        closed = true;
                        break;
                    }

                    if (readBuf.size() >= maxMessagesPerRead | !autoRead) {
                        break;
                    }
                }
            } catch (Throwable t) {
                exception = t;
            }

            int size = readBuf.size();
            for (int i = 0; i < size; i ++) {
                pipeline.fireChannelRead(readBuf.get(i));
            }
            readBuf.clear();
            pipeline.fireChannelReadComplete();

            if (exception != null) {
                if (exception instanceof IOException) {
                    // ServerChannel should not be closed even on IOException because it can often continue
                    // accepting incoming connections. (e.g. too many open files)
                    closed = !(AbstractNioMessageChannel.this instanceof ServerChannel);
                }

                pipeline.fireExceptionCaught(exception);
            }

            if (closed) {
                if (isOpen()) {
                    close(voidPromise());
                }
            }
        }
    }

// NioServerSocketChannel.doReadMessage方法
protected int doReadMessages(List<Object> buf) throws Exception {
        SocketChannel ch = javaChannel().accept();

        try {
            if (ch != null) {
                buf.add(new NioSocketChannel(this, childEventLoopGroup().next(), ch));
                return 1;
            }
        } catch (Throwable t) {
            logger.warn("Failed to create a new channel from an accepted socket.", t);

            try {
                ch.close();
            } catch (Throwable t2) {
                logger.warn("Failed to close a socket.", t2);
            }
        }

        return 0;
    }

public NioSocketChannel(Channel parent, EventLoop eventLoop, SocketChannel socket) {
        super(parent, eventLoop, socket);
        config = new DefaultSocketChannelConfig(this, socket.socket());
    }

protected AbstractNioByteChannel(Channel parent, EventLoop eventLoop, SelectableChannel ch) {
        super(parent, eventLoop, ch, SelectionKey.OP_READ);
    }

protected AbstractNioChannel(Channel parent, EventLoop eventLoop, SelectableChannel ch, int readInterestOp) {
        super(parent, eventLoop);
        this.ch = ch;
        this.readInterestOp = readInterestOp;
        try {
            ch.configureBlocking(false);
        } catch (IOException e) {
            try {
                ch.close();
            } catch (IOException e2) {
                if (logger.isWarnEnabled()) {
                    logger.warn(
                            "Failed to close a partially initialized socket.", e2);
                }
            }

            throw new ChannelException("Failed to enter non-blocking mode.", e);
        }
    }

// ServerBootstrapAcceptor.channelRead方法来注册Channel
public void channelRead(ChannelHandlerContext ctx, Object msg) {
            Channel child = (Channel) msg;

            child.pipeline().addLast(childHandler);

            for (Entry<ChannelOption<?>, Object> e: childOptions) {
                try {
                    if (!child.config().setOption((ChannelOption<Object>) e.getKey(), e.getValue())) {
                        logger.warn("Unknown channel option: " + e);
                    }
                } catch (Throwable t) {
                    logger.warn("Failed to set a channel option: " + child, t);
                }
            }

            for (Entry<AttributeKey<?>, Object> e: childAttrs) {
                child.attr((AttributeKey<Object>) e.getKey()).set(e.getValue());
            }

            child.unsafe().register(child.newPromise());
        }

// AbstractChannel的register0方法,先注册,后fireChannelActive来调用HeadHandler设置SelectionKey的interestOps
private void register0(ChannelPromise promise) {
            try {
                // check if the channel is still open as it could be closed in the mean time when the register
                // call was outside of the eventLoop
                if (!ensureOpen(promise)) {
                    return;
                }
                doRegister();
                registered = true;
                promise.setSuccess();
                pipeline.fireChannelRegistered();
                if (isActive()) {
                    pipeline.fireChannelActive();
                }
            } catch (Throwable t) {
                // Close the channel directly to avoid FD leak.
                closeForcibly();
                closeFuture.setClosed();
                if (!promise.tryFailure(t)) {
                    logger.warn(
                            "Tried to fail the registration promise, but it is complete already. " +
                                    "Swallowing the cause of the registration failure:", t);
                }
            }
        }



// AbstractNioChannel 注册Channel到selector,只注册空事件,具体OP_READ事件延迟注册
protected void doRegister() throws Exception {
        boolean selected = false;
        for (;;) {
            try {
                selectionKey = javaChannel().register(eventLoop().selector, 0, this);
                return;
            } catch (CancelledKeyException e) {
                if (!selected) {
                    // Force the Selector to select now as the "canceled" SelectionKey may still be
                    // cached and not removed because no Select.select(..) operation was called yet.
                    eventLoop().selectNow();
                    selected = true;
                } else {
                    // We forced a select operation on the selector before but the SelectionKey is still cached
                    // for whatever reason. JDK bug ?
                    throw e;
                }
            }
        }
    }


 protected void doBeginRead() throws Exception {
        if (inputShutdown) {
            return;
        }

        final SelectionKey selectionKey = this.selectionKey;
        if (!selectionKey.isValid()) {
            return;
        }
// interestOps = 0, readInterestOps = OP_READ = 1
        final int interestOps = selectionKey.interestOps();
        if ((interestOps & readInterestOp) == 0) {
            selectionKey.interestOps(interestOps | readInterestOp);
        }
    }

最后看OP_WRITE如何注册。OP_WRITE比较特殊,表示本地的写缓冲区可用,一般只有在一次写没有把数据写完的情况下需要注册OP_WRITE,写完后要及时关闭,否则每次循环都有可能被调用,因为写缓冲区在大多数情况下是始终可用的。

Netty只有在flush的时候才会实际调用底层的写方法把数据写到网络中去。ChannelHandlerContext提供了flush和writeAndFlush方法来作flush。

// AbstractNioByteChannel的打开,关闭写事件的方法

protected final void setOpWrite() {
        final SelectionKey key = selectionKey();
        final int interestOps = key.interestOps();
        if ((interestOps & SelectionKey.OP_WRITE) == 0) {
            key.interestOps(interestOps | SelectionKey.OP_WRITE);
        }
    }

    protected final void clearOpWrite() {
        final SelectionKey key = selectionKey();
        final int interestOps = key.interestOps();
        if ((interestOps & SelectionKey.OP_WRITE) != 0) {
            key.interestOps(interestOps & ~SelectionKey.OP_WRITE);
        }
    }

//在写半包的情况下会设置OP_WRITE
protected final void incompleteWrite(boolean setOpWrite) {
        // Did not write completely.
        if (setOpWrite) {
            setOpWrite();
        } else {
            // Schedule flush again later so other tasks can be picked up in the meantime
            Runnable flushTask = this.flushTask;
            if (flushTask == null) {
                flushTask = this.flushTask = new Runnable() {
                    @Override
                    public void run() {
                        flush();
                    }
                };
            }
            eventLoop().execute(flushTask);
        }
    }

protected void doWrite(ChannelOutboundBuffer in) throws Exception {
        int writeSpinCount = -1;

        for (;;) {
            Object msg = in.current(true);
            if (msg == null) {
                // Wrote all messages.
                clearOpWrite();
                break;
            }

            if (msg instanceof ByteBuf) {
                ByteBuf buf = (ByteBuf) msg;
                int readableBytes = buf.readableBytes();
                if (readableBytes == 0) {
                    in.remove();
                    continue;
                }

                boolean setOpWrite = false;
                boolean done = false;
                long flushedAmount = 0;
                if (writeSpinCount == -1) {
                    writeSpinCount = config().getWriteSpinCount();
                }
                for (int i = writeSpinCount - 1; i >= 0; i --) {
                    int localFlushedAmount = doWriteBytes(buf);
                    if (localFlushedAmount == 0) {
                        setOpWrite = true;
                        break;
                    }

                    flushedAmount += localFlushedAmount;
                    if (!buf.isReadable()) {
                        done = true;
                        break;
                    }
                }

                in.progress(flushedAmount);

                if (done) {
                    in.remove();
                } else {
                    incompleteWrite(setOpWrite);
                    break;
                }
            }

// AbstractUnsafe的flush0方法调用了doWrite

protected void flush0() {
            if (inFlush0) {
                // Avoid re-entrance
                return;
            }

            final ChannelOutboundBuffer outboundBuffer = this.outboundBuffer;
            if (outboundBuffer == null || outboundBuffer.isEmpty()) {
                return;
            }

            inFlush0 = true;

            // Mark all pending write requests as failure if the channel is inactive.
            if (!isActive()) {
                try {
                    if (isOpen()) {
                        outboundBuffer.failFlushed(NOT_YET_CONNECTED_EXCEPTION);
                    } else {
                        outboundBuffer.failFlushed(CLOSED_CHANNEL_EXCEPTION);
                    }
                } finally {
                    inFlush0 = false;
                }
                return;
            }

            try {
                doWrite(outboundBuffer);
            } catch (Throwable t) {
                outboundBuffer.failFlushed(t);
            } finally {
                inFlush0 = false;
            }
        }




通过分析Netty注册OP_ACCEPT, OP_READ, OP_WRITE事件的过程,再比照使用原生的Java NIO API开发的过程,能够更深入的理解Netty的封装过程


























你可能感兴趣的:(netty)