目录
一、介绍
二、依赖引入
三、公共部分实现
四、server端实现
五、client端实现
六、测试
一、介绍
本片文章将实现请求响应同步,什么是请求响应同步呢?就是当我们发起一个请求时,希望能够在一定时间内同步(线程阻塞)等待响应结果。
我们通过netty实现rpc调用时,由于客户端和服务端保持连接,在此期间客户端会有无数的接口调用(并发),而此时,每次发送的请求需要能够及时响应获取调用结果,服务端一次次返回调用结果,客户端在处理响应结果时,需要与请求建立联系,确保每一次的请求能够正确获取到对应的调用结果。
由于在一个应用中,客户端与服务端的channel只有一条,所有线程都通过该channel进行rpc调用,所以,在接下来客户端设计中,每个线程发送的请求将会分配一个id,当请求发送完毕之后,该线程会进行阻塞状态,等待channel收到请求id对应返回的响应消息时唤醒或超时唤醒。在接下来服务端设计中,服务端收到客户端的rpc调用请求,对该请求进行处理,将该请求的id和处理结果写入响应类中进行返回。
二、依赖引入
<dependencies><dependency><groupId>io.netty</groupId><artifactId>netty-all</artifactId><version>4.1.101.Final</version></dependency><dependency><groupId>io.protostuff</groupId><artifactId>protostuff-core</artifactId><version>1.8.0</version></dependency><dependency><groupId>io.protostuff</groupId><artifactId>protostuff-runtime</artifactId><version>1.8.0</version></dependency<dependency><groupId>org.projectlombok</groupId><artifactId>lombok</artifactId><version>1.18.30</version></dependency></dependencies>
三、公共部分实现
1、结构
2、Message类,所有Request和Response类的父类,最关键的字段就是messageType,子类继承之后进行赋值,该值与类的类型进行绑定,用于byte字节数组反序列化时能够获取到需要反序列化的类型。
@Data
public abstract class Message {protected Byte messageType;}
RpcRequest,用于客户端向服务端发起调用的消息通信类
@Data
@ToString
public class RpcRequest extends Message{private String id;private String param;public RpcRequest() {this.id = UUID.randomUUID().toString();super.messageType = MessageConstant.rpcRequest;}}
RpcResponse,用于服务端向客户端返回结构的消息通信类
@Data
@ToString
public class RpcResponse extends Message{private String id;private String result;public RpcResponse() {super.messageType = MessageConstant.rpcResponse;}}
3、MessageConstant,通过数值常量messageType绑定消息类型,在序列化对象时,会在数据中记录对象的messageType,在反序列化对象时,会从数据包中拿到messageType,将其转化为对应的消息类型进行处理
public class MessageConstant {public final static Byte rpcRequest = 1;public final static Byte rpcResponse = 2;public static Map<Byte, Class<? extends Message>> messageTypeMap = new ConcurrentHashMap<>();static {messageTypeMap.put(rpcRequest, RpcRequest.class);messageTypeMap.put(rpcResponse, RpcResponse.class);}public static Class<? extends Message> getMessageClass(Byte messageType){return messageTypeMap.get(messageType);}}
4、序列化工具,用于将类对象序列化为字节数组,以及将字节数组反序列化为对象
public class SerializationUtil {private final static Map<Class<?>, Schema<?>> schemaCache = new ConcurrentHashMap<>();/*** 序列化*/public static <T> byte[] serialize(T object){LinkedBuffer buffer = LinkedBuffer.allocate(LinkedBuffer.DEFAULT_BUFFER_SIZE);try {Class<T> cls = (Class<T>) object.getClass();Schema<T> schema = getSchema(cls);return ProtostuffIOUtil.toByteArray(object, schema, buffer);} catch (Exception e) {throw e;} finally {buffer.clear();}}/*** 反序列化*/public static <T> T deserialize(Class<T> cls, byte[] data) {Schema<T> schema = getSchema(cls);T message = schema.newMessage();ProtostuffIOUtil.mergeFrom(data, message, schema);return message;}public static <T> Schema<T> getSchema(Class<T> cls) {Schema<T> schema = (Schema<T>) schemaCache.get(cls);if(schema == null) {schema = RuntimeSchema.getSchema(cls);schemaCache.put(cls, schema);}return schema;}}
5、MesasgeEncode和MessageDecode实现
MessageEncode,用于将消息对象序列化为字节数组
字节数组主要包括三部分:
·有效数组长度,占4个字节,长度不包括自己,用于半包黏包判断
·消息的类型,占1个字节,用于反序列选择类型使用
·消息对象,占n个字节
public class MessageEncode extends MessageToByteEncoder<Message> {@Overrideprotected void encode(ChannelHandlerContext channelHandlerContext, Message message, ByteBuf byteBuf) throws Exception {// 将对象进行序列化byte[] data = SerializationUtil.serialize(message);// 写数据长度,前4个字节用于记录数据总长度(对象 + 类型(1个字节))byteBuf.writeInt(data.length + 1);// 写记录消息类型,用于反序列选择类的类型byteBuf.writeByte(message.getMessageType());// 写对象byteBuf.writeBytes(data);}}
MesageDecode,用于将字节数组反序列化为消息对象
反序列时会进行判断数据是否足够读取,足够的话就会读取到符合长度的字节数组进行序列化,否则的话等到下一个数据包到来再进行重新判断处理(解决半包黏包方案)
public class MessageDecode extends ByteToMessageDecoder {@Overrideprotected void decode(ChannelHandlerContext channelHandlerContext, ByteBuf byteBuf, List<Object> list) throws Exception {// 由于数据包的前4个字节用于记录总数据大小,如果数据不够4个字节,不进行读if(byteBuf.readableBytes() < 4) {return;}// 标记开始读的位置byteBuf.markReaderIndex();// 前四个字节记录了数据大小int dataSize = byteBuf.readInt();// 查看剩余可读字节是否足够,如果不是,重置读取位置,等待下一次解析if(byteBuf.readableBytes() < dataSize) {byteBuf.resetReaderIndex();return;}// 读取消息类型byte messageType = byteBuf.readByte();// 读取数据, 数组大小需要剔除1个字节的消息类型byte[] data = new byte[dataSize -1];byteBuf.readBytes(data);Message message = SerializationUtil.deserialize(MessageConstant.getMessageClass(messageType), data);list.add(message);}}
四、server端实现
1、结构
2、RpcRequestHandler,用于处理客户端rpc请求
public class RpcRequestHandler extends SimpleChannelInboundHandler<RpcRequest> {private final static EventLoopGroup worker = new DefaultEventLoopGroup(Runtime.getRuntime().availableProcessors() + 1);@Overrideprotected void channelRead0(ChannelHandlerContext ctx, RpcRequest msg) throws Exception {// 为避免占用网络io,此处异步进行处理worker.submit(() -> {System.out.println("[RpcRequestHandler] "+ Thread.currentThread().getName() +" 处理请求,msg: " + msg);// 模拟处理耗时try {Thread.sleep(3000);} catch (InterruptedException e) {throw new RuntimeException(e);}RpcResponse rpcResponse = new RpcResponse();rpcResponse.setId(msg.getId());rpcResponse.setResult("处理" + msg.getParam());ctx.writeAndFlush(rpcResponse);});}}
3、ServerChannelInitializer,该类用于初始化Server与Client通信的Channel,需要将我们前面写的编解码器以及RequestHandler添加进pipeline
public class ServerChannelInitializer extends ChannelInitializer<SocketChannel> {@Overrideprotected void initChannel(SocketChannel socketChannel) throws Exception {ChannelPipeline pipeline = socketChannel.pipeline();pipeline.addLast(new MessageEncode());pipeline.addLast(new MessageDecode());pipeline.addLast(new RpcRequestHandler());}}
4、RpcServer,用于启动一个Netty Server服务
public class RpcServer {public void bind(Integer port) {EventLoopGroup parent = new NioEventLoopGroup();EventLoopGroup child = new NioEventLoopGroup();Channel channel = null;try{ServerBootstrap serverBootstrap = new ServerBootstrap();serverBootstrap.group(parent, child).channel(NioServerSocketChannel.class).option(ChannelOption.SO_BACKLOG, 1024).childHandler(new ServerChannelInitializer());ChannelFuture channelFuture = serverBootstrap.bind(port).sync();System.out.println("server启动");// 非阻塞等待关闭channelFuture.channel().closeFuture().addListener(new ChannelFutureListener() {@Overridepublic void operationComplete(ChannelFuture channelFuture) throws Exception {System.out.println("server关闭");parent.shutdownGracefully();child.shutdownGracefully();}});channel = channelFuture.channel();} catch (Exception e) {e.printStackTrace();if(channel == null || !channel.isActive()) {System.out.println("server关闭");parent.shutdownGracefully();child.shutdownGracefully();} else {channel.close();}}}}
五、client端实现
1、结构
2、SyncPromise,用于Netty客户端的工作线程与外部发起RpcRequest的线程通信的类,通过该类可以阻塞与唤醒外部发起RpcRequest的线程,以及设置线程之间通信的内容(功能有点像Netty提供的Promise,不过此处我加了超时机制)
此处使用CountDownLatch来阻塞与唤醒线程有以下好处:
1、能够通过await(long timeout, TimeUnit unit)返回值true/false进行判断线程等待返回结果是否超时。因为线程进入阻塞时,CountDownLatch的值为1,当netty客户端的工作线程调用countDown()唤醒线程时,CountDownLatch值减为0,await(long timeout, TimeUnit unit)返回true,意味着线程等待响应结果时,没有超时。当netty客户端的工作线程没有来得及调用countDown()唤醒线程时。也就是说服务端返回结果超时,CountDownLatch值为1,线程超时唤醒,await(long timeout, TimeUnit unit)返回false。
综上所述,以await(long timeout, TimeUnit unit)返回值进行判断线程是否超时唤醒。此处给一个对比,就是有人认为为什么不使用LockSupport进行线程的阻塞与唤醒,原因如下:虽然LockSupport提供了超时唤醒的方法,但是该方法既没有返回值,也没有抛出异常,线程唤醒时,我们没有办法判断该线程是否超时了。
2、在我们实现的流程中,我们先发送了请求,才进行线程阻塞。那么存在一种情况,如果结果在我们线程阻塞之前就返回了,那么当线程进入阻塞时,就再也没有唤醒线程的时机了,导致线程每次调用接口都是超时的。
CountDownLatch的await(long timeout, TimeUnit unit)方法很好的规避了上诉问题,如果netty客户端的工作线程调用countDown()唤醒线程,那么此时CountDownLatch值减为0,线程需要调用await()进入阻塞,此时由于CountDownLatch为0,线程将不会进入阻塞,方法返回true,我们线程也能够正常的拿到请求的响应结果。
具体妙处需要大家仔细感受,一开始可能不太能理解,但把流程仔细梳理一下,就能够有更好的体验。
public class SyncPromise {// 用于接收结果private RpcResponse rpcResponse;private final CountDownLatch countDownLatch = new CountDownLatch(1);// 用于判断是否超时private boolean isTimeout = false;/*** 同步等待返回结果*/public RpcResponse get(long timeout, TimeUnit unit) throws InterruptedException {// 等待阻塞,超时时间内countDownLatch减到0,将提前唤醒,以此作为是否超时判断boolean earlyWakeUp = countDownLatch.await(timeout, unit);if(earlyWakeUp) {// 超时时间内countDownLatch减到0,提前唤醒,说明已有结果return rpcResponse;} else {// 超时时间内countDownLatch没有减到0,自动唤醒,说明超时时间内没有等到结果isTimeout = true;return null;}}public void wake() {countDownLatch.countDown();}public RpcResponse getRpcResponse() {return rpcResponse;}public void setRpcResponse(RpcResponse rpcResponse) {this.rpcResponse = rpcResponse;}public boolean isTimeout() {return isTimeout;}}
3、RpcUtil,封装的请求发送工具类,需要调用rpc发送的请求的线程,将通过该工具的send方法进行远程调用,不能简单的通过channel.writeAndFlush()进行客户端与服务端的通信
syncPromiseMap的作用:记录请求对应的SyncPromise对象(一次请求对应一个SyncPromise对象),由于外部线程与netty客户端的工作线程是通过SyncPromise进行通信的,我们需要通过请求的id与SyncPromise建立关系,确保netty客户端在处理RpcResopnse时,能够根据其中的请求id属性值,找到对应SyncPromise对象,为其设置响应值,以及唤醒等待结果的线程。
public class RpcUtil {private final static Map<String, SyncPromise> syncPromiseMap = new ConcurrentHashMap<>();private final static Channel channel;static{channel = new RpcClient().connect("127.0.0.1", 8888);}public static RpcResponse send(RpcRequest rpcRequest, long timeout, TimeUnit unit) throws Exception{if(channel == null) {throw new NullPointerException("channel");}if(rpcRequest == null) {throw new NullPointerException("rpcRequest");}if(timeout <= 0) {throw new IllegalArgumentException("timeout must greater than 0");}// 创造一个容器,用于存放当前线程与rpcClient中的线程交互SyncPromise syncPromise = new SyncPromise();syncPromiseMap.put(rpcRequest.getId(), syncPromise);// 发送消息,此处如果发送玩消息并且在get之前返回了结果,下一行的get将不会进入阻塞,也可以顺利拿到结果channel.writeAndFlush(rpcRequest);// 等待获取结果RpcResponse rpcResponse = syncPromise.get(timeout, unit);if(rpcResponse == null) {if(syncPromise.isTimeout()) {throw new TimeoutException("等待响应结果超时");} else{throw new Exception("其他异常");}}// 移除容器syncPromiseMap.remove(rpcRequest.getId());return rpcResponse;}public static Map<String, SyncPromise> getSyncPromiseMap(){return syncPromiseMap;}}
4、RpcResponseHandler,处理返回的调用结果,在该处理器中,将唤醒等待返回结果的线程
public class RpcResponseHandler extends SimpleChannelInboundHandler<RpcResponse> {@Overrideprotected void channelRead0(ChannelHandlerContext ctx, RpcResponse msg) throws Exception {// 根据请求id,在集合中找到与外部线程通信的SyncPromise对象SyncPromise syncPromise = RpcUtil.getSyncPromiseMap().get(msg.getId());if(syncPromise != null) {// 设置响应结果syncPromise.setRpcResponse(msg);// 唤醒外部线程syncPromise.wake();}}}
5、ClientChannelInitializer,该类用于初始化Server与Client通信的Channel,需要将我们前面写的编解码器以及ResponseHandler添加进pipeline
public class ClientChannelInitializer extends ChannelInitializer<SocketChannel> {@Overrideprotected void initChannel(SocketChannel socketChannel) throws Exception {ChannelPipeline pipeline = socketChannel.pipeline();pipeline.addLast(new MessageEncode());pipeline.addLast(new MessageDecode());pipeline.addLast(new RpcResponseHandler());}}
6、RpcClient实现,用于启动客户端
public class RpcClient {public Channel connect(String host, Integer port) {EventLoopGroup worker = new NioEventLoopGroup();Channel channel = null;try {Bootstrap bootstrap = new Bootstrap();bootstrap.group(worker).channel(NioSocketChannel.class).option(ChannelOption.AUTO_READ, true).handler(new ClientChannelInitializer());ChannelFuture channelFuture = bootstrap.connect(host, port).sync();System.out.println("客户端启动");channel = channelFuture.channel();// 添加关闭监听器channel.closeFuture().addListener(new ChannelFutureListener() {@Overridepublic void operationComplete(ChannelFuture channelFuture) throws Exception {System.out.println("关闭客户端");worker.shutdownGracefully();}});} catch (Exception e) {e.printStackTrace();if(channel == null || !channel.isActive()) {worker.shutdownGracefully();} else {channel.close();}}return channel;}}
六、测试
1、启动服务端
public static void main(String[] args) {new RpcServer().bind(8888);
}
启动结果如下:
server启动
2、启动客户端,并且通过两个异步线程发送请求
public static void main(String[] args) throws Exception{//Channel channel = new RpcClient().connect("127.0.0.1", 8888);Thread thread1 = new Thread(new Runnable() {@Overridepublic void run() {RpcRequest rpcRequest = new RpcRequest();rpcRequest.setParam("参数1");try {System.out.println("thread1发送请求");RpcResponse rpcResponse = RpcUtil.send(rpcRequest, 5, TimeUnit.SECONDS);System.out.println("thread1处理结果:" + rpcResponse);} catch (Exception e) {throw new RuntimeException(e);}}});Thread thread2 = new Thread(new Runnable() {@Overridepublic void run() {RpcRequest rpcRequest2 = new RpcRequest();rpcRequest2.setParam("参数2");try {System.out.println("thread2发送请求");RpcResponse rpcResponse = RpcUtil.send(rpcRequest2, 5, TimeUnit.SECONDS);System.out.println("thread2处理结果:" + rpcResponse);} catch (Exception e) {throw new RuntimeException(e);}}});// 休眠一下,等待客户端与服务端进行连接Thread.sleep(1000);thread1.start();thread2.start();}
服务端结果:
[RpcRequestHandler] defaultEventLoopGroup-4-3 处理请求,msg: RpcRequest(id=ade6af01-2bcf-4a4c-a42a-381731010027, param=参数1)
[RpcRequestHandler] defaultEventLoopGroup-4-4 处理请求,msg: RpcRequest(id=db57bf9a-3220-44ca-8e4f-d74237a3d5b2, param=参数2)
客户端结果
thread1发送请求
thread2发送请求
thread1处理结果:RpcResponse(id=ade6af01-2bcf-4a4c-a42a-381731010027, result=处理参数1)
thread2处理结果:RpcResponse(id=db57bf9a-3220-44ca-8e4f-d74237a3d5b2, result=处理参数2)
以上由于我们在RpcRequestHandler中模拟处理请求为3秒,而线程等待结果超时为5秒,所以接下来将线程调用rpc请求的的超时时间设置为2秒,重启客户端,客户端结果如下:
thread1发送请求
thread2发送请求
Exception in thread "Thread-1" Exception in thread "Thread-0" java.lang.RuntimeException: java.util.concurrent.TimeoutException: 等待响应结果超时at org.ricardo.sync.client.RpcClientTest$1.run(RpcClientTest.java:32)at java.lang.Thread.run(Thread.java:748)
Caused by: java.util.concurrent.TimeoutException: 等待响应结果超时at org.ricardo.sync.client.rpc.RpcUtil.send(RpcUtil.java:56)at org.ricardo.sync.client.RpcClientTest$1.run(RpcClientTest.java:29)... 1 more
java.lang.RuntimeException: java.util.concurrent.TimeoutException: 等待响应结果超时at org.ricardo.sync.client.RpcClientTest$2.run(RpcClientTest.java:48)at java.lang.Thread.run(Thread.java:748)
Caused by: java.util.concurrent.TimeoutException: 等待响应结果超时at org.ricardo.sync.client.rpc.RpcUtil.send(RpcUtil.java:56)at org.ricardo.sync.client.RpcClientTest$2.run(RpcClientTest.java:45)... 1 more