前言:
公司的加密机调度系统一直使用的是http请求调度的方式去调度,但是会出现网络故障导致某个客户端或者服务端断线的情况,导致很多请求信息以及回执信息丢失的情况,接着我们抛弃了http的方式,改为Tcp的方式去建立客户端和服务器之间的连接,并且要去实现断线重连的功能,经过讨论后决定使用java中成熟的nio框架 – netty去解决这一系列的问题。
1. netty简单介绍:
在百度中对netty的解释是:
Netty是由JBOSS提供的一个java开源框架。Netty提供异步的、事件驱动的网络应用程序框架和工具,用以快速开发高性能、高可靠性的网络服务器和客户端程序。
Netty框架并不只是封装了多路复用的IO模型,也包括提供了传统的阻塞式/非阻塞式 同步IO的模型封,Netty 是一个利用 Java 的高级网络的能力,隐藏其背后的复杂性而提供一个易于使用的 API 的客户端/服务器框架。其并发高、传输快、封装好的特性受到了许多大公司的青睐,在这里我们就不过多的分析netty的原理和特性了,之后我会写一篇文章专门写一下从io到nio,再到netty的整个过程。重点讲一下netty的魅力所在,今天我们已代码实现为主,讲解一下在springboot架构中,用netty实现服务端和客户端之间的通信以及断线重连等机制。
2. 服务端代码:
首先,引入netty的pom依赖
<dependency><groupId>io.netty</groupId><artifactId>netty-all</artifactId><version>5.0.0.Alpha2</version> </dependency>
然后我们在配置文件中写入服务端的ip和端口号,用于连接
在springboot的application启动类中写入服务端的启动start方法,用于在启动项目时自动启动服务端
1 @SpringBootApplication 2 public class Application implements CommandLineRunner { 3 4 @Value("${netty.server.port}") 5 private int port; 6 7 @Value("${netty.server.host}") 8 private String host; 9 10 @Autowired 11 NettyServer server; 12 13 public static void main(String[] args) { 14 SpringApplication.run(Application.class, args); 15 } 16 17 18 @Override 19 public void run(String... strings) throws Exception { 20 this.startServer(); 21 22 } 23 24 //启动service 25 public void startServer(){
//这个类实现一个IP套接字地址(IP地址+端口号) 26 InetSocketAddress address = new InetSocketAddress(host, port); 27 ChannelFuture future = server.start(address); 28 29 Runtime.getRuntime().addShutdownHook(new Thread(){ 30 @Override 31 public void run() { 32 server.destroy(); 33 } 34 }); 35 36 future.channel().closeFuture().syncUninterruptibly(); 37 } 38 } 39 40 41 }
ChannelFuture:
Future最早出现于JDK的java.util.concurrent.Future,它用于表示异步操作的结果.由于Netty的Future都是与异步I/O操作相关的,因此命名为ChannelFuture,代表它与Channel操作相关.由于Netty中的所有I / O操作都是异步的,因此Netty为了解决调用者如何获取异步操作结果的问题而专门设计了ChannelFuture接口.
因此,Channel与ChannelFuture可以说形影不离的.
然后我们要去重点看server.start()
public class NettyServer {private static final Logger logger = LoggerFactory.getLogger(ConnectionWatchdog.class);private final ChannelGroup channelGroup = new DefaultChannelGroup(ImmediateEventExecutor.INSTANCE);private final EventLoopGroup bossGroup = new NioEventLoopGroup();private final EventLoopGroup workGroup = new NioEventLoopGroup();private Channel channel;/*** 开启及服务线程*/public ChannelFuture start(InetSocketAddress address) {//服务端引导类ServerBootstrap bootstrap = new ServerBootstrap();bootstrap.group(bossGroup, workGroup)//通过ServerBootstrap的group方法,设置(1)中初始化的主从"线程池".channel(NioServerSocketChannel.class)//指定通道channel的类型,由于是服务端,故而是NioServerSocketChannel.childHandler(new NettyServerInitializer())//设置ServerSocketChannel的处理器.option(ChannelOption.SO_BACKLOG, 100)// 设置tcp协议的请求等待队列.childOption(ChannelOption.SO_KEEPALIVE, true);//配置子通道也就是SocketChannel的选项ChannelFuture future = bootstrap.bind(address).syncUninterruptibly();logger.info("准备接收——————");channel = future.channel();return future;}public void destroy() {if(channel != null) {channel.close();}channelGroup.close();workGroup.shutdownGracefully();bossGroup.shutdownGracefully();}}
在这里的设置中,.childHandler(new NettyServerInitializer()) 用于设置了服务器管道 NioServerSocketChannel 的处理器handler,
这个handler是我们自定义封装的一些对channel的public class NettyServerInitializer extends ChannelInitializer<Channel>{
@Component
public class TcpMsgHandler extends ChannelInboundHandlerAdapter {
@Overrideprotected void initChannel(Channel ch) throws Exception {ChannelPipeline pipeline = ch.pipeline();//处理日志//pipeline.addLast(new LoggingHandler(LogLevel.INFO));//处理心跳pipeline.addLast(new IdleStateHandler(5, 0, 0, TimeUnit.SECONDS));
//消息编码pipeline.addLast(new MessageEncoder());
//粘包长度控制pipeline.addLast(new LengthFieldBasedFrameDecoder(Integer.MAX_VALUE,0,4));
//消息解码pipeline.addLast(new MessageDecoder());
//自定义handerpipeline.addLast(new TcpMsgHandler());}
}
ChannelPipeline :
Netty 的 Channel 过滤器实现原理与 Servlet Filter 机制一致,它将 Channel 的数据管道抽象为 ChannelPipeline,消息在 ChannelPipeline 中流动和传递。ChannelPipeline 持有 I/O 事件拦截器 ChannelHandler 的链表,由 ChannelHandler 来对 I/O 事件进行具体的拦截和处理,可以方便地通过新增和删除 ChannelHandler 来实现不同业务逻辑的定制,能够实现对修改封闭和对扩展到支持。
我们看到我们添加了idleStateHandler用来处理心跳,那么心跳究竟是什么呢,我们先来介绍一下心跳
心跳机制
- 心跳是在TCP长连接中,客户端和服务端定时向对方发送数据包通知对方自己还在线,保证连接的有效性的一种机制
- 在服务器和客户端之间一定时间内没有数据交互时, 即处于 idle 状态时, 客户端或服务器会发送一个特殊的数据包给对方, 当接收方收到这个数据报文后, 也立即发送一个特殊的数据报文, 回应发送方, 此即一个 PING-PONG 交互. 自然地, 当某一端收到心跳消息后, 就知道了对方仍然在线, 这就确保 TCP 连接的有效性
在我们的服务端中,不会主动发心跳给客户端,只会对对应的心跳消息,进行回应,告诉那些给我发心跳的客户端说:我还活着!
-
服务端添加IdleStateHandler心跳检测处理器,并添加自定义处理Handler类实现userEventTriggered()方法作为超时事件的逻辑处理;
-
设定IdleStateHandler心跳检测每五秒进行一次读检测,如果五秒内ChannelRead()方法未被调用则触发一次userEventTrigger()方法
TcpMsgHandler.java
@Component
public class TcpMsgHandler extends ChannelInboundHandlerAdapter {private final static Logger logger = LoggerFactory.getLogger("");@Overridepublic void userEventTriggered(ChannelHandlerContext ctx, Object evt) throws Exception {if (evt instanceof IdleStateEvent) { }IdleState state = ((IdleStateEvent) evt).state();if (state == IdleState.READER_IDLE) {ctx.close();}} else {super.userEventTriggered(ctx, evt);}}@Overridepublic void channelRead(ChannelHandlerContext ctx, Object obj) throws Exception {TcpMsg msg = (TcpMsg) obj;try {//处理心跳...ctx.writeAndFlush(msg);}}catch(Exception ex){logger.info(ex.getMessage());}}
}
在这里,我们的channelRead比较简单,只是将客户端发来的心跳直接发回去了,实现了响应客户端心跳请求的目的,除了心跳,我们还可以去定义不同的消息类别,比如说是加密请求,还是处理数据的请求,入库的请求等等,
我们可以自己从channel中获取到客户端发过来的信息做处理,记得要即使响应,比如,心跳中,我们将msg又返回给了channel:
ctx.writeAndFlush(msg);
在handler中,decoder用于解码的作用,将客户端发来的ByteBuf流的形式,转为我们需要的格式,可以转为我们要的对象,或者是一个string字符串
MessageDecoder.java
public class MessageDecoder extends ByteToMessageDecoder {private Logger logger = LoggerFactory.getLogger("");@Overrideprotected void decode(ChannelHandlerContext ctx, ByteBuf in, List<Object> out) throws Exception {int len = in.readableBytes();byte[] bytes = new byte[len];
//将ByteBuf转为byte数组in.readBytes(bytes);try {TcpMsg msg = TcpMsg.ByteToObj(bytes);out.add(msg);} catch (Exception ex) {logger.error("MessageDecoder",ex);}}}
encoder负责在我们发送数据的时候,将我们的对象、或者是字符串转为byte数组,然后输出
public class MessageEncoder extends MessageToByteEncoder<TcpMsg>{private Logger logger = LoggerFactory.getLogger("");@Overrideprotected void encode(ChannelHandlerContext ctx, TcpMsg msg, ByteBuf out) throws Exception {try{if (msg.getType() != 0){//logger.info("send: " + msg.getType() + ":" + msg.getGuid() + ":" + msg.getBody());}byte[] src = msg.ToBytes();out.writeBytes(src);}catch (Exception e){logger.error("MessageEncoder",e);}}
}
3. 客户端代码:
在application配置文件中加入服务端的主机名和端口号
netty.server.host = 127.0.0.1
netty.server.port = 9090
启动类Application:
@SpringBootApplication
public class Application{@Autowiredprivate NettyClient client;@Value("${netty.server.port}")private int port;@Value("${netty.server.host}")private String host;public static void main(String[] args) throws Exception {SpringApplication.run(NettyClientApplication.class, args);}@Beanpublic NettyClient nettyClient() {return new NettyClient();}@Overridepublic void run(String... arg0) throws Exception {client.start(host, port);}}
NettyClient.java: 客户端启动类
@Component
public class NettyClient {//日志输出private static final Logger logger = LoggerFactory.getLogger(NettyClient.class);//主要连接地址private static String nettyHost = "";//备用连接地址private static String nettyHostRe = "";private static Integer nettyPort = 0;public boolean start(String host1,String host2,int port){//主要连接地址nettyHost = host1;//备用连接地址nettyHostRe = host2;nettyPort = port;//EventLoopGroup可以理解为是一个线程池,这个线程池用来处理连接、接受数据、发送数据EventLoopGroup nioEventLoopGroup = new NioEventLoopGroup();//NioEventLoop//客户端引导类Bootstrap bootstrap = new Bootstrap();//多线程处理bootstrap.group(nioEventLoopGroup);//指定通道类型为NioServerSocketChannel,一种异步模式bootstrap.channel(NioSocketChannel.class);//指定请求地址bootstrap.remoteAddress(new InetSocketAddress(nettyHost,port));bootstrap.option(ChannelOption.TCP_NODELAY,true);final ConnectionWatchdog watchDog = new ConnectionWatchdog(bootstrap, new HashedWheelTimer(), nettyHost,nettyHostRe, port) {@Overridepublic ChannelHandler[] handlers() {return new ChannelHandler[]{new MessageEncoder(),new LengthFieldBasedFrameDecoder(Integer.MAX_VALUE,0,4),new MessageDecoder(),this,// 每隔5s的时间触发一次userEventTriggered的方法,并且指定IdleState的状态位是WRITER_IDLEnew IdleStateHandler(0, 1, 0, TimeUnit.SECONDS),// 实现userEventTriggered方法,并在state是WRITER_IDLE的时候发送一个心跳包到sever端,告诉server端我还活着new ClientHeartBeatHandler(),};}};final ChannelFuture future;try {synchronized (bootstrap) {bootstrap.handler(new ChannelInitializer<NioSocketChannel>() {@Overrideprotected void initChannel(NioSocketChannel ch) throws Exception {ch.pipeline().addLast(watchDog.handlers());}});future = bootstrap.connect().sync();// 链接服务器.调用sync()方法会同步阻塞//服务端连接ip:logger.info("目前服务端连接ip为" + nettyHost);}if (!future.isSuccess()) {logger.info("---- 连接服务器失败,2秒后重试 ---------port=" + port);future.channel().eventLoop().schedule(new Runnable() {@Overridepublic void run() {start(nettyHost,nettyHostRe,nettyPort);}}, 2L, TimeUnit.SECONDS);}} catch (Exception e) {logger.info("exception happends e {}", e);return false;}return true;}}
ConnectionWatchdog.java :重连检测狗,当发现当前的链路不稳定关闭之后,进行重连
@ChannelHandler.Sharable
public abstract class ConnectionWatchdog extends ChannelInboundHandlerAdapter implements TimerTask,ChannelHandlerHolder{//日志输出private static final Logger logger = LoggerFactory.getLogger(ConnectionWatchdog.class);//客户端引导类private Bootstrap bootstrap;private Timer timer;private final String host;//备用服务端ipprivate final String host2;//使用ipprivate String useHost;private final int port;private volatile boolean reconnect = true;private int attempts;//刷新时间private volatile long refreshTime = 0L;//心跳连接标识private volatile boolean heartBeatCheck = false;//通道private volatile Channel channel;//失败次数private static int failCount;public ConnectionWatchdog(Bootstrap boot, Timer timer, String host,String host2, int port) {this.bootstrap = boot;this.timer = timer;this.host = host;this.host2 = host2;this.port = port;}public boolean isReconnect() {return reconnect;}public void setReconnect(boolean reconnect) {this.reconnect = reconnect;}
//连接成功时调用的方法@Overridepublic void channelActive(final ChannelHandlerContext ctx) throws Exception {channel = ctx.channel();attempts = 0;reconnect =false;refreshTime = new Date().getTime();if (!heartBeatCheck) {heartBeatCheck = true;channel.eventLoop().scheduleAtFixedRate(new Runnable() {@Overridepublic void run() {long time = new Date().getTime() - refreshTime;logger.info(String.valueOf(time));if (time > 5 * 1000L) {channel.close();logger.info("心跳检查失败");} else {logger.info("心跳检查Successs");}}}, 5L, 5L, TimeUnit.SECONDS);}logger.info("Connects with {}.", channel);ctx.fireChannelActive();}/*** 因为链路断掉之后,会触发channelInActive方法,进行重连 2秒重连一次*/@Overridepublic void channelInactive(ChannelHandlerContext ctx) throws Exception {reconnect = true;logger.warn("Disconnects with {}, doReconnect = {},attempts == {}", ctx.channel(), reconnect, attempts);if (reconnect) {/*if (attempts < 12) {attempts++;} else {reconnect = false;}*/long timeout = 2;logger.info("再过 {} 秒客户端将进行重连",timeout);timer.newTimeout(this, timeout, TimeUnit.SECONDS);}}/** run启动方法* */public void run(Timeout timeout) throws Exception {//Future表示异步操作的结果final ChannelFuture future;if(failCount > 2){//使用备用ipif(host.equals(useHost)){useHost = host2;}else{useHost = host;}}else {if(StrUtil.IsNullOrEmpty(useHost)) {//首次重连useHost = host;}}synchronized (bootstrap) {future = bootstrap.connect(useHost, port);}//使用future监听结果,执行异步操作结束后的回调.future.addListener(new ChannelFutureListener() {@Overridepublic void operationComplete(final ChannelFuture f) throws Exception {boolean succeed = f.isSuccess();logger.warn("连接通过 {}, {}.", useHost + ":" + port, succeed ? "成功" : "失败");if (!succeed) {logger.info("重连失败");failCount ++;f.channel().pipeline().fireChannelInactive();}else{failCount = 0;logger.info("重连成功");}}});}@Overridepublic void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {if (msg instanceof TcpMsg) {TcpMsg heartMsg = (TcpMsg) msg;if (heartMsg.getType()>=0) {refreshTime = new Date().getTime();}logger.warn("得到服务器响应,响应内容为"+ ((TcpMsg) msg).getBody());}}@Overridepublic void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {super.exceptionCaught(ctx, cause);Channel channel = ctx.channel();logger.info("客户端:"+channel.remoteAddress()+"网络异常");cause.printStackTrace();if(channel.isActive())ctx.close();}}
这里我们定义了一个变量: refreshTime,当我们从channel中read到了服务端发来的心跳响应消息的话,就刷新refreshTime为当前时间
当连接成功时,会触发channelActive 方法,在这里我们开启了一个定时任务去判断refreshTime和当前时间的时间差,超过5秒说明断线了,要进行重连,我这里由于配置了两个服务器,所有在我的逻辑中,尝试连接2次以上连不上就去连另一个服务器去了
下面的handler用于发送心跳消息,实现userEventTriggered方法,并在state是WRITER_IDLE的时候发送一个心跳包到sever端,告诉server端我还活着
@Component
public class ClientHeartBeatHandler extends ChannelInboundHandlerAdapter {private static final Logger logger = LoggerFactory.getLogger(ClientHeartBeatHandler.class);@Overridepublic void userEventTriggered(ChannelHandlerContext ctx, Object evt) throws Exception {clientname = ReadFileUtil.readFile("C:/CrawlProgram/wrapper_nettyClient/name.txt");if (evt instanceof IdleStateEvent) {IdleState state = ((IdleStateEvent) evt).state();if (state == IdleState.WRITER_IDLE) {//用于心跳的客户端类型为0int type = 0;//客户端机器名String body = clientname;TcpMsg msg = new TcpMsg(type,body);try {ctx.writeAndFlush(msg).sync();logger.info("发送消息成功,消息类型为:"+type+",请求id为" + msg.getGuid() + ",客户端机器号为:" + msg.getBody());} catch (Exception ex) {ex.printStackTrace();logger.info("发送失败");}}} else {super.userEventTriggered(ctx, evt);}}}
然后就是和服务端一样的decoder、encoder过程,不同的是,我们在decoder的时候使用了线程池去将任务放入队列中去,防止请求慢的时候丢失任务请求
MessageDecoder.java
public class MessageDecoder extends ByteToMessageDecoder {private static final Logger logger = LoggerFactory.getLogger(MessageDecoder.class);@Autowiredprivate VisiableThreadPoolTaskExecutor visiableThreadPoolTaskExecutor;//线程池常量public static VisiableThreadPoolTaskExecutor executor;private TcpMsg tcpMsg;List<Object> out;// 用@PostConstruct方法引导绑定@PostConstructpublic void init() {executor = visiableThreadPoolTaskExecutor;encryptService = encrypt;orderService = order;}@Overridepublic void decode(ChannelHandlerContext ctx, ByteBuf in, List<Object> out) throws Exception {this.context = ctx;this.out = out;int len = in.readableBytes();if (len > 0) {logger.info("得到返回数据,长度为" + len);byte[] bytes = new byte[len];in.readBytes(bytes);TcpMsg msg = TcpMsg.ByteToObj(bytes);this.tcpMsg = msg;logger.info("start asyncServiceExecutor");executor.execute(new Runnable() {@Overridepublic void run() {executeTask();}});logger.info("end executeAsync");}}}
这里,我们使用了netty来实现了服务端、客户端通信、心跳检测的功能。体会到了netty的传输效率高、封装好的特性,用起来简单、实用。我们不仅可以做断线重连、还可以做很多业务请求,可以配置多台客户端去做不同的事情,来达到服务器调度的目的。
归根结底,netty还是一个框架的东西,我们还是没有过多的去看透nio的本质、我们要做的不仅仅是会用netty,而且还要了解nio、了解netty的实现原理,它的底层是如何封装的,希望大家多去研究,我们一起去搞懂它
Netty 的 Channel 过滤器实现原理与 Servlet Filter 机制一致,它将 Channel 的数据管道抽象为 ChannelPipeline,消息在 ChannelPipeline 中流动和传递。ChannelPipeline 持有 I/O 事件拦截器 ChannelHandler 的链表,由 ChannelHandler 来对 I/O 事件进行具体的拦截和处理,可以方便地通过新增和删除 ChannelHandler 来实现不同业务逻辑的定制,能够实现对修改封闭和对扩展到支持。