提醒：本篇适合有一定netty基础的读者阅读

心跳机制

何为心跳

所谓心跳, 即在 TCP 长连接中, 客户端和服务器之间定期发送的一种特殊的数据包, 通知对方自己还在线, 以确保 TCP 连接的有效性.

注：心跳包还有另一个作用，经常被忽略，即：一个连接如果长时间不用，防火墙或者路由器就会断开该连接。

如何实现

核心Handler —— IdleStateHandler

在 Netty 中, 实现心跳机制的关键是 IdleStateHandler, 那么这个 Handler 如何使用呢? 先看下它的构造器：

public IdleStateHandler(int readerIdleTimeSeconds, int writerIdleTimeSeconds, int allIdleTimeSeconds) {
    this((long)readerIdleTimeSeconds, (long)writerIdleTimeSeconds, (long)allIdleTimeSeconds, TimeUnit.SECONDS);
}

这里解释下三个参数的含义：

• readerIdleTimeSeconds: 读超时. 即当在指定的时间间隔内没有从 Channel 读取到数据时, 会触发一个 READER_IDLE 的 IdleStateEvent 事件.
• writerIdleTimeSeconds: 写超时. 即当在指定的时间间隔内没有数据写入到 Channel 时, 会触发一个 WRITER_IDLE 的 IdleStateEvent 事件.
• allIdleTimeSeconds: 读/写超时. 即当在指定的时间间隔内没有读或写操作时, 会触发一个 ALL_IDLE 的 IdleStateEvent 事件.

注：这三个参数默认的时间单位是秒。若需要指定其他时间单位，可以使用另一个构造方法：IdleStateHandler(boolean observeOutput, long readerIdleTime, long writerIdleTime, long allIdleTime, TimeUnit unit)

在看下面的实现之前，建议先了解一下IdleStateHandler的实现原理。相关链接：一起学Netty（十一）之 Netty心跳之IdleStateHandler。

下面直接上代码，需要注意的地方，会在代码中通过注释进行说明。

使用IdleStateHandler实现心跳

下面将使用IdleStateHandler来实现心跳，Client端连接到Server端后，会循环执行一个任务：随机等待几秒，然后ping一下Server端，即发送一个心跳包。当等待的时间超过规定时间，将会发送失败，以为Server端在此之前已经主动断开连接了。代码如下：

Client端

ClientIdleStateTrigger —— 心跳触发器

类ClientIdleStateTrigger也是一个Handler，只是重写了userEventTriggered方法，用于捕获IdleState.WRITER_IDLE事件（未在指定时间内向服务器发送数据），然后向Server端发送一个心跳包。

/**
 * <p>
 *  用于捕获{@link IdleState#WRITER_IDLE}事件（未在指定时间内向服务器发送数据），然后向<code>Server</code>端发送一个心跳包。
 * </p>
 */
public class ClientIdleStateTrigger extends ChannelInboundHandlerAdapter {

    public static final String HEART_BEAT = "heart beat!";

    @Override
    public void userEventTriggered(ChannelHandlerContext ctx, Object evt) throws Exception {
        if (evt instanceof IdleStateEvent) {
            IdleState state = ((IdleStateEvent) evt).state();
            if (state == IdleState.WRITER_IDLE) {
                // write heartbeat to server
                ctx.writeAndFlush(HEART_BEAT);
            }
        } else {
            super.userEventTriggered(ctx, evt);
        }
    }

}

Pinger —— 心跳发射器

/**
 * <p>客户端连接到服务器端后，会循环执行一个任务：随机等待几秒，然后ping一下Server端，即发送一个心跳包。</p>
 */
public class Pinger extends ChannelInboundHandlerAdapter {

    private Random random = new Random();
    private int baseRandom = 8;

    private Channel channel;

    @Override
    public void channelActive(ChannelHandlerContext ctx) throws Exception {
        super.channelActive(ctx);
        this.channel = ctx.channel();

        ping(ctx.channel());
    }

    private void ping(Channel channel) {
        int second = Math.max(1, random.nextInt(baseRandom));
        System.out.println("next heart beat will send after " + second + "s.");
        ScheduledFuture<?> future = channel.eventLoop().schedule(new Runnable() {
            @Override
            public void run() {
                if (channel.isActive()) {
                    System.out.println("sending heart beat to the server...");
                    channel.writeAndFlush(ClientIdleStateTrigger.HEART_BEAT);
                } else {
                    System.err.println("The connection had broken, cancel the task that will send a heart beat.");
                    channel.closeFuture();
                    throw new RuntimeException();
                }
            }
        }, second, TimeUnit.SECONDS);

        future.addListener(new GenericFutureListener() {
            @Override
            public void operationComplete(Future future) throws Exception {
                if (future.isSuccess()) {
                    ping(channel);
                }
            }
        });
    }

    @Override
    public void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {
        // 当Channel已经断开的情况下, 仍然发送数据, 会抛异常, 该方法会被调用.
        cause.printStackTrace();
        ctx.close();
    }
}

ClientHandlersInitializer —— 客户端处理器集合的初始化类

public class ClientHandlersInitializer extends ChannelInitializer<SocketChannel> {

    private ReconnectHandler reconnectHandler;
    private EchoHandler echoHandler;

    public ClientHandlersInitializer(TcpClient tcpClient) {
        Assert.notNull(tcpClient, "TcpClient can not be null.");
        this.reconnectHandler = new ReconnectHandler(tcpClient);
        this.echoHandler = new EchoHandler();
    }

    @Override
    protected void initChannel(SocketChannel ch) throws Exception {
        ChannelPipeline pipeline = ch.pipeline();
        pipeline.addLast(new LengthFieldBasedFrameDecoder(Integer.MAX_VALUE, 0, 4, 0, 4));
        pipeline.addLast(new LengthFieldPrepender(4));
        pipeline.addLast(new StringDecoder(CharsetUtil.UTF_8));
        pipeline.addLast(new StringEncoder(CharsetUtil.UTF_8));
        pipeline.addLast(new Pinger());
    }
}

注： 上面的Handler集合，除了Pinger，其他都是编解码器和解决粘包，可以忽略。

TcpClient —— TCP连接的客户端

public class TcpClient {

    private String host;
    private int port;
    private Bootstrap bootstrap;
    /** 将<code>Channel</code>保存起来, 可用于在其他非handler的地方发送数据 */
    private Channel channel;

    public TcpClient(String host, int port) {
        this(host, port, new ExponentialBackOffRetry(1000, Integer.MAX_VALUE, 60 * 1000));
    }

    public TcpClient(String host, int port, RetryPolicy retryPolicy) {
        this.host = host;
        this.port = port;
        init();
    }

    /**
     * 向远程TCP服务器请求连接
     */
    public void connect() {
        synchronized (bootstrap) {
            ChannelFuture future = bootstrap.connect(host, port);
            this.channel = future.channel();
        }
    }

    private void init() {
        EventLoopGroup group = new NioEventLoopGroup();
        // bootstrap 可重用, 只需在TcpClient实例化的时候初始化即可.
        bootstrap = new Bootstrap();
        bootstrap.group(group)
                .channel(NioSocketChannel.class)
                .handler(new ClientHandlersInitializer(TcpClient.this));
    }

    public static void main(String[] args) {
        TcpClient tcpClient = new TcpClient("localhost", 2222);
        tcpClient.connect();
    }

}

Server端

ServerIdleStateTrigger —— 断连触发器

/**
 * <p>在规定时间内未收到客户端的任何数据包, 将主动断开该连接</p>
 */
public class ServerIdleStateTrigger extends ChannelInboundHandlerAdapter {
    @Override
    public void userEventTriggered(ChannelHandlerContext ctx, Object evt) throws Exception {
        if (evt instanceof IdleStateEvent) {
            IdleState state = ((IdleStateEvent) evt).state();
            if (state == IdleState.READER_IDLE) {
                // 在规定时间内没有收到客户端的上行数据, 主动断开连接
                ctx.disconnect();
            }
        } else {
            super.userEventTriggered(ctx, evt);
        }
    }
}

ServerBizHandler —— 服务器端的业务处理器

/**
 * <p>收到来自客户端的数据包后, 直接在控制台打印出来.</p>
 */
@ChannelHandler.Sharable
public class ServerBizHandler extends SimpleChannelInboundHandler<String> {

    private final String REC_HEART_BEAT = "I had received the heart beat!";

    @Override
    protected void channelRead0(ChannelHandlerContext ctx, String data) throws Exception {
        try {
            System.out.println("receive data: " + data);
//            ctx.writeAndFlush(REC_HEART_BEAT);
        } catch (Exception e) {
            e.printStackTrace();
        }
    }

    @Override
    public void channelActive(ChannelHandlerContext ctx) throws Exception {
        System.out.println("Established connection with the remote client.");

        // do something

        ctx.fireChannelActive();
    }

    @Override
    public void channelInactive(ChannelHandlerContext ctx) throws Exception {
        System.out.println("Disconnected with the remote client.");

        // do something

        ctx.fireChannelInactive();
    }

    @Override
    public void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {
        cause.printStackTrace();
        ctx.close();
    }
}

ServerHandlerInitializer —— 服务器端处理器集合的初始化类

/**
 * <p>用于初始化服务器端涉及到的所有<code>Handler</code></p>
 */
public class ServerHandlerInitializer extends ChannelInitializer<SocketChannel> {

    protected void initChannel(SocketChannel ch) throws Exception {
        ch.pipeline().addLast("idleStateHandler", new IdleStateHandler(5, 0, 0));
        ch.pipeline().addLast("idleStateTrigger", new ServerIdleStateTrigger());
        ch.pipeline().addLast("frameDecoder", new LengthFieldBasedFrameDecoder(Integer.MAX_VALUE, 0, 4, 0, 4));
        ch.pipeline().addLast("frameEncoder", new LengthFieldPrepender(4));
        ch.pipeline().addLast("decoder", new StringDecoder());
        ch.pipeline().addLast("encoder", new StringEncoder());
        ch.pipeline().addLast("bizHandler", new ServerBizHandler());
    }

}

注：new IdleStateHandler(5, 0, 0)该handler代表如果在5秒内没有收到来自客户端的任何数据包（包括但不限于心跳包），将会主动断开与该客户端的连接。

TcpServer —— 服务器端

public class TcpServer {
    private int port;
    private ServerHandlerInitializer serverHandlerInitializer;

    public TcpServer(int port) {
        this.port = port;
        this.serverHandlerInitializer = new ServerHandlerInitializer();
    }

    public void start() {
        EventLoopGroup bossGroup = new NioEventLoopGroup(1);
        EventLoopGroup workerGroup = new NioEventLoopGroup();
        try {
            ServerBootstrap bootstrap = new ServerBootstrap();
            bootstrap.group(bossGroup, workerGroup)
                    .channel(NioServerSocketChannel.class)
                    .childHandler(this.serverHandlerInitializer);
            // 绑定端口，开始接收进来的连接
            ChannelFuture future = bootstrap.bind(port).sync();

            System.out.println("Server start listen at " + port);
            future.channel().closeFuture().sync();
        } catch (Exception e) {
            bossGroup.shutdownGracefully();
            workerGroup.shutdownGracefully();
            e.printStackTrace();
        }
    }

    public static void main(String[] args) throws Exception {
        int port = 2222;
        new TcpServer(port).start();
    }
}

至此，所有代码已经编写完毕。

测试

首先启动客户端，再启动服务器端。启动完成后，在客户端的控制台上，可以看到打印如下类似日志：

客户端控制台输出的日志

在服务器端可以看到控制台输出了类似如下的日志：

服务器端控制台输出的日志

异常情况

在测试过程中，有可能会出现如下情况：

异常情况

目前尚未找到优雅处理这种情况的方案，各位看官如果有好的解决方案，还望不吝赐教。拜谢！！！

断线重连

断线重连这里就不过多介绍，相信各位都知道是怎么回事。这里只说大致思路，然后直接上代码。

实现思路

客户端在监测到与服务器端的连接断开后，或者一开始就无法连接的情况下，使用指定的重连策略进行重连操作，直到重新建立连接或重试次数耗尽。

对于如何监测连接是否断开，则是通过重写ChannelInboundHandler#channelInactive来实现，但连接不可用，该方法会被触发，所以只需要在该方法做好重连工作即可。

代码实现

注：以下代码都是在上面心跳机制的基础上修改/添加的。

因为断线重连是客户端的工作，所以只需对客户端代码进行修改。

重试策略

RetryPolicy —— 重试策略接口

public interface RetryPolicy {

    /**
     * Called when an operation has failed for some reason. This method should return
     * true to make another attempt.
     *
     * @param retryCount the number of times retried so far (0 the first time)
     * @return true/false
     */
    boolean allowRetry(int retryCount);

    /**
     * get sleep time in ms of current retry count.
     *
     * @param retryCount current retry count
     * @return the time to sleep
     */
    long getSleepTimeMs(int retryCount);
}

ExponentialBackOffRetry —— 重连策略的默认实现

/**
 * <p>Retry policy that retries a set number of times with increasing sleep time between retries</p>
 */
public class ExponentialBackOffRetry implements RetryPolicy {

    private static final int MAX_RETRIES_LIMIT = 29;
    private static final int DEFAULT_MAX_SLEEP_MS = Integer.MAX_VALUE;

    private final Random random = new Random();
    private final long baseSleepTimeMs;
    private final int maxRetries;
    private final int maxSleepMs;

    public ExponentialBackOffRetry(int baseSleepTimeMs, int maxRetries) {
        this(baseSleepTimeMs, maxRetries, DEFAULT_MAX_SLEEP_MS);
    }

    public ExponentialBackOffRetry(int baseSleepTimeMs, int maxRetries, int maxSleepMs) {
        this.maxRetries = maxRetries;
        this.baseSleepTimeMs = baseSleepTimeMs;
        this.maxSleepMs = maxSleepMs;
    }

    @Override
    public boolean allowRetry(int retryCount) {
        if (retryCount < maxRetries) {
            return true;
        }
        return false;
    }

    @Override
    public long getSleepTimeMs(int retryCount) {
        if (retryCount < 0) {
            throw new IllegalArgumentException("retries count must greater than 0.");
        }
        if (retryCount > MAX_RETRIES_LIMIT) {
            System.out.println(String.format("maxRetries too large (%d). Pinning to %d", maxRetries, MAX_RETRIES_LIMIT));
            retryCount = MAX_RETRIES_LIMIT;
        }
        long sleepMs = baseSleepTimeMs * Math.max(1, random.nextInt(1 << retryCount));
        if (sleepMs > maxSleepMs) {
            System.out.println(String.format("Sleep extension too large (%d). Pinning to %d", sleepMs, maxSleepMs));
            sleepMs = maxSleepMs;
        }
        return sleepMs;
    }
}

ReconnectHandler—— 重连处理器

@ChannelHandler.Sharable
public class ReconnectHandler extends ChannelInboundHandlerAdapter {

    private int retries = 0;
    private RetryPolicy retryPolicy;

    private TcpClient tcpClient;

    public ReconnectHandler(TcpClient tcpClient) {
        this.tcpClient = tcpClient;
    }

    @Override
    public void channelActive(ChannelHandlerContext ctx) throws Exception {
        System.out.println("Successfully established a connection to the server.");
        retries = 0;
        ctx.fireChannelActive();
    }

    @Override
    public void channelInactive(ChannelHandlerContext ctx) throws Exception {
        if (retries == 0) {
            System.err.println("Lost the TCP connection with the server.");
            ctx.close();
        }

        boolean allowRetry = getRetryPolicy().allowRetry(retries);
        if (allowRetry) {

            long sleepTimeMs = getRetryPolicy().getSleepTimeMs(retries);

            System.out.println(String.format("Try to reconnect to the server after %dms. Retry count: %d.", sleepTimeMs, ++retries));

            final EventLoop eventLoop = ctx.channel().eventLoop();
            eventLoop.schedule(() -> {
                System.out.println("Reconnecting ...");
                tcpClient.connect();
            }, sleepTimeMs, TimeUnit.MILLISECONDS);
        }
        ctx.fireChannelInactive();
    }


    private RetryPolicy getRetryPolicy() {
        if (this.retryPolicy == null) {
            this.retryPolicy = tcpClient.getRetryPolicy();
        }
        return this.retryPolicy;
    }
}

ClientHandlersInitializer

在之前的基础上，添加了重连处理器ReconnectHandler。

public class ClientHandlersInitializer extends ChannelInitializer<SocketChannel> {

    private ReconnectHandler reconnectHandler;
    private EchoHandler echoHandler;

    public ClientHandlersInitializer(TcpClient tcpClient) {
        Assert.notNull(tcpClient, "TcpClient can not be null.");
        this.reconnectHandler = new ReconnectHandler(tcpClient);
        this.echoHandler = new EchoHandler();
    }

    @Override
    protected void initChannel(SocketChannel ch) throws Exception {
        ChannelPipeline pipeline = ch.pipeline();
        pipeline.addLast(this.reconnectHandler);
        pipeline.addLast(new LengthFieldBasedFrameDecoder(Integer.MAX_VALUE, 0, 4, 0, 4));
        pipeline.addLast(new LengthFieldPrepender(4));
        pipeline.addLast(new StringDecoder(CharsetUtil.UTF_8));
        pipeline.addLast(new StringEncoder(CharsetUtil.UTF_8));
        pipeline.addLast(new Pinger());
    }
}

TcpClient

在之前的基础上添加重连、重连策略的支持。

public class TcpClient {

    private String host;
    private int port;
    private Bootstrap bootstrap;
    /** 重连策略 */
    private RetryPolicy retryPolicy;
    /** 将<code>Channel</code>保存起来, 可用于在其他非handler的地方发送数据 */
    private Channel channel;

    public TcpClient(String host, int port) {
        this(host, port, new ExponentialBackOffRetry(1000, Integer.MAX_VALUE, 60 * 1000));
    }

    public TcpClient(String host, int port, RetryPolicy retryPolicy) {
        this.host = host;
        this.port = port;
        this.retryPolicy = retryPolicy;
        init();
    }

    /**
     * 向远程TCP服务器请求连接
     */
    public void connect() {
        synchronized (bootstrap) {
            ChannelFuture future = bootstrap.connect(host, port);
            future.addListener(getConnectionListener());
            this.channel = future.channel();
        }
    }

    public RetryPolicy getRetryPolicy() {
        return retryPolicy;
    }

    private void init() {
        EventLoopGroup group = new NioEventLoopGroup();
        // bootstrap 可重用, 只需在TcpClient实例化的时候初始化即可.
        bootstrap = new Bootstrap();
        bootstrap.group(group)
                .channel(NioSocketChannel.class)
                .handler(new ClientHandlersInitializer(TcpClient.this));
    }

    private ChannelFutureListener getConnectionListener() {
        return new ChannelFutureListener() {
            @Override
            public void operationComplete(ChannelFuture future) throws Exception {
                if (!future.isSuccess()) {
                    future.channel().pipeline().fireChannelInactive();
                }
            }
        };
    }

    public static void main(String[] args) {
        TcpClient tcpClient = new TcpClient("localhost", 2222);
        tcpClient.connect();
    }

}

测试

在测试之前，为了避开 Connection reset by peer 异常，可以稍微修改Pinger的ping()方法，添加if (second == 5)的条件判断。如下：

private void ping(Channel channel) {
        int second = Math.max(1, random.nextInt(baseRandom));
        if (second == 5) {
            second = 6;
        }
        System.out.println("next heart beat will send after " + second + "s.");
        ScheduledFuture<?> future = channel.eventLoop().schedule(new Runnable() {
            @Override
            public void run() {
                if (channel.isActive()) {
                    System.out.println("sending heart beat to the server...");
                    channel.writeAndFlush(ClientIdleStateTrigger.HEART_BEAT);
                } else {
                    System.err.println("The connection had broken, cancel the task that will send a heart beat.");
                    channel.closeFuture();
                    throw new RuntimeException();
                }
            }
        }, second, TimeUnit.SECONDS);

        future.addListener(new GenericFutureListener() {
            @Override
            public void operationComplete(Future future) throws Exception {
                if (future.isSuccess()) {
                    ping(channel);
                }
            }
        });
    }

启动客户端

先只启动客户端，观察控制台输出，可以看到类似如下日志：

断线重连测试——客户端控制台输出

可以看到，当客户端发现无法连接到服务器端，所以一直尝试重连。随着重试次数增加，重试时间间隔越大，但又不想无限增大下去，所以需要定一个阈值，比如60s。如上图所示，当下一次重试时间超过60s时，会打印Sleep extension too large(*). Pinning to 60000，单位为ms。出现这句话的意思是，计算出来的时间超过阈值（60s），所以把真正睡眠的时间重置为阈值（60s）。

启动服务器端

接着启动服务器端，然后继续观察客户端控制台输出。

断线重连测试——服务器端启动后客户端控制台输出

扩展

在不同环境，可能会有不同的重连需求。有不同的重连需求的，只需自己实现RetryPolicy接口，然后在创建TcpClient的时候覆盖默认的重连策略即可。

完！！！