java 从零开始手写 RPC (01) 基于 socket 实现
java 从零开始手写 RPC (02)-netty4 实现客户端和服务端
java 从零开始手写 RPC (03) 如何实现客户端调用服务端?
java 从零开始手写 RPC (04) -序列化
上一篇我们介绍了,如何实现基于反射的通用服务端。
这一节我们来一起学习下如何实现通用客户端。
因为内容较多,所以拆分为 2 个部分。
基本思路所有的方法调用,基于反射进行相关处理实现。
核心类为了便于拓展,我们把核心类调整如下:
package com.github.houbb.rpc.client.core;
import com.github.houbb.heaven.annotation.ThreadSafe;
import com.github.houbb.log.integration.core.Log;
import com.github.houbb.log.integration.core.LogFactory;
import com.github.houbb.rpc.client.core.context.RpcClientContext;
import com.github.houbb.rpc.client.handler.RpcClientHandler;
import com.github.houbb.rpc.common.constant.RpcConstant;
import io.netty.bootstrap.Bootstrap;
import io.netty.channel.*;
import io.netty.channel.nio.NioEventLoopGroup;
import io.netty.channel.socket.nio.NioSocketChannel;
import io.netty.handler.codec.serialization.ClassResolvers;
import io.netty.handler.codec.serialization.ObjectDecoder;
import io.netty.handler.codec.serialization.ObjectEncoder;
import io.netty.handler.logging.LogLevel;
import io.netty.handler.logging.LoggingHandler;
@ThreadSafe
public class RpcClient {
private static final Log log = LogFactory.getLog(RpcClient.class);
private final String address;
private final int port;
private final ChannelHandler channelHandler;
public RpcClient(final RpcClientContext clientContext) {
this.address = clientContext.address();
this.port = clientContext.port();
this.channelHandler = clientContext.channelHandler();
}
public ChannelFuture connect() {
// 启动服务端
log.info("RPC 服务开始启动客户端");
EventLoopGroup workerGroup = new NioEventLoopGroup();
ChannelFuture channelFuture;
try {
Bootstrap bootstrap = new Bootstrap();
channelFuture = bootstrap.group(workerGroup)
.channel(NioSocketChannel.class)
.option(ChannelOption.SO_KEEPALIVE, true)
.handler(new ChannelInitializer(){
@Override
protected void initChannel(Channel ch) throws Exception {
ch.pipeline()
// 解码 bytes=>resp
.addLast(new ObjectDecoder(Integer.MAX_VALUE, ClassResolvers.cacheDisabled(null)))
// request=>bytes
.addLast(new ObjectEncoder())
// 日志输出
.addLast(new LoggingHandler(LogLevel.INFO))
.addLast(channelHandler);
}
})
.connect(address, port)
.syncUninterruptibly();
log.info("RPC 服务启动客户端完成,监听地址 {}:{}", address, port);
} catch (Exception e) {
log.error("RPC 客户端遇到异常", e);
throw new RuntimeException(e);
}
// 不要关闭线程池!!!
return channelFuture;
}
}
可以灵活指定对应的服务端地址、端口信息。
ChannelHandler 作为处理参数传入。
ObjectDecoder、ObjectEncoder、LoggingHandler 都和服务端类似,是 netty 的内置实现。
RpcClientHandler客户端的 handler 实现如下:
package com.github.houbb.rpc.client.handler;
import com.github.houbb.log.integration.core.Log;
import com.github.houbb.log.integration.core.LogFactory;
import com.github.houbb.rpc.client.core.RpcClient;
import com.github.houbb.rpc.client.invoke.InvokeService;
import com.github.houbb.rpc.common.rpc.domain.RpcResponse;
import io.netty.channel.ChannelHandlerContext;
import io.netty.channel.SimpleChannelInboundHandler;
public class RpcClientHandler extends SimpleChannelInboundHandler {
private static final Log log = LogFactory.getLog(RpcClient.class);
private final InvokeService invokeService;
public RpcClientHandler(InvokeService invokeService) {
this.invokeService = invokeService;
}
@Override
protected void channelRead0(ChannelHandlerContext ctx, Object msg) throws Exception {
RpcResponse rpcResponse = (RpcResponse)msg;
invokeService.addResponse(rpcResponse.seqId(), rpcResponse);
log.info("[Client] response is :{}", rpcResponse);
}
@Override
public void channelReadComplete(ChannelHandlerContext ctx) throws Exception {
// 每次用完要关闭,不然拿不到response,我也不知道为啥(目测得了解netty才行)
// 个人理解:如果不关闭,则永远会被阻塞。
ctx.flush();
ctx.close();
}
@Override
public void exceptionCaught(ChannelHandlerContext ctx, Throwable cause) throws Exception {
cause.printStackTrace();
ctx.close();
}
}
只有 channelRead0 做了调整,基于 InvokeService 对结果进行处理。
InvokeService 接口package com.github.houbb.rpc.client.invoke;
import com.github.houbb.rpc.common.rpc.domain.RpcResponse;
public interface InvokeService {
InvokeService addRequest(final String seqId);
InvokeService addResponse(final String seqId, final RpcResponse rpcResponse);
RpcResponse getResponse(final String seqId);
}
主要是对入参、出参的设置,以及出参的获取。
实现package com.github.houbb.rpc.client.invoke.impl;
import com.github.houbb.heaven.util.guava.Guavas;
import com.github.houbb.heaven.util.lang.ObjectUtil;
import com.github.houbb.log.integration.core.Log;
import com.github.houbb.log.integration.core.LogFactory;
import com.github.houbb.rpc.client.core.RpcClient;
import com.github.houbb.rpc.client.invoke.InvokeService;
import com.github.houbb.rpc.common.exception.RpcRuntimeException;
import com.github.houbb.rpc.common.rpc.domain.RpcResponse;
import java.util.Set;
import java.util.concurrent.ConcurrentHashMap;
public class DefaultInvokeService implements InvokeService {
private static final Log LOG = LogFactory.getLog(DefaultInvokeService.class);
private final Set requestSet;
private final ConcurrentHashMap responseMap;
public DefaultInvokeService() {
requestSet = Guavas.newHashSet();
responseMap = new ConcurrentHashMap<>();
}
@Override
public InvokeService addRequest(String seqId) {
LOG.info("[Client] start add request for seqId: {}", seqId);
requestSet.add(seqId);
return this;
}
@Override
public InvokeService addResponse(String seqId, RpcResponse rpcResponse) {
// 这里放入之前,可以添加判断。
// 如果 seqId 必须处理请求集合中,才允许放入。或者直接忽略丢弃。
LOG.info("[Client] 获取结果信息,seq: {}, rpcResponse: {}", seqId, rpcResponse);
responseMap.putIfAbsent(seqId, rpcResponse);
// 通知所有等待方
LOG.info("[Client] seq 信息已经放入,通知所有等待方", seqId);
synchronized (this) {
this.notifyAll();
}
return this;
}
@Override
public RpcResponse getResponse(String seqId) {
try {
RpcResponse rpcResponse = this.responseMap.get(seqId);
if(ObjectUtil.isNotNull(rpcResponse)) {
LOG.info("[Client] seq {} 对应结果已经获取: {}", seqId, rpcResponse);
return rpcResponse;
}
// 进入等待
while (rpcResponse == null) {
LOG.info("[Client] seq {} 对应结果为空,进入等待", seqId);
// 同步等待锁
synchronized (this) {
this.wait();
}
rpcResponse = this.responseMap.get(seqId);
LOG.info("[Client] seq {} 对应结果已经获取: {}", seqId, rpcResponse);
}
return rpcResponse;
} catch (InterruptedException e) {
throw new RpcRuntimeException(e);
}
}
}
使用 requestSet 存储对应的请求入参。
使用 responseMap 存储对应的请求出参,在获取的时候通过同步 while 循环等待,获取结果。
此处,通过 notifyAll() 和 wait() 进行等待和唤醒。
ReferenceConfig-服务端配置 说明我们想调用服务端,首先肯定要定义好要调用的对象。
ReferenceConfig 就是要告诉 rpc 框架,调用的服务端信息。
接口package com.github.houbb.rpc.client.config.reference; import com.github.houbb.rpc.common.config.component.RpcAddress; import java.util.List; public interface ReferenceConfig实现{ ReferenceConfig serviceId(final String serviceId); String serviceId(); Class serviceInterface(); ReferenceConfig serviceInterface(final Class serviceInterface); ReferenceConfig addresses(final String addresses); T reference(); }
package com.github.houbb.rpc.client.config.reference.impl; import com.github.houbb.heaven.constant.PunctuationConst; import com.github.houbb.heaven.util.common.ArgUtil; import com.github.houbb.heaven.util.guava.Guavas; import com.github.houbb.heaven.util.lang.NumUtil; import com.github.houbb.rpc.client.config.reference.ReferenceConfig; import com.github.houbb.rpc.client.core.RpcClient; import com.github.houbb.rpc.client.core.context.impl.DefaultRpcClientContext; import com.github.houbb.rpc.client.handler.RpcClientHandler; import com.github.houbb.rpc.client.invoke.InvokeService; import com.github.houbb.rpc.client.invoke.impl.DefaultInvokeService; import com.github.houbb.rpc.client.proxy.ReferenceProxy; import com.github.houbb.rpc.client.proxy.context.ProxyContext; import com.github.houbb.rpc.client.proxy.context.impl.DefaultProxyContext; import com.github.houbb.rpc.common.config.component.RpcAddress; import io.netty.channel.ChannelFuture; import io.netty.channel.ChannelHandler; import java.util.List; public class DefaultReferenceConfigimplements ReferenceConfig { private String serviceId; private Class serviceInterface; private List rpcAddresses; private List channelFutures; @Deprecated private RpcClientHandler channelHandler; private InvokeService invokeService; public DefaultReferenceConfig() { // 初始化信息 this.rpcAddresses = Guavas.newArrayList(); this.channelFutures = Guavas.newArrayList(); this.invokeService = new DefaultInvokeService(); } @Override public String serviceId() { return serviceId; } @Override public DefaultReferenceConfig serviceId(String serviceId) { this.serviceId = serviceId; return this; } @Override public Class serviceInterface() { return serviceInterface; } @Override public DefaultReferenceConfig serviceInterface(Class serviceInterface) { this.serviceInterface = serviceInterface; return this; } @Override public ReferenceConfig addresses(String addresses) { ArgUtil.notEmpty(addresses, "addresses"); String[] addressArray = addresses.split(PunctuationConst.COMMA); ArgUtil.notEmpty(addressArray, "addresses"); for(String address : addressArray) { String[] addressSplits = address.split(PunctuationConst.colon); if(addressSplits.length < 2) { throw new IllegalArgumentException("Address must be has ip and port, like 127.0.0.1:9527"); } String ip = addressSplits[0]; int port = NumUtil.toIntegerThrows(addressSplits[1]); // 包含权重信息 int weight = 1; if(addressSplits.length >= 3) { weight = NumUtil.toInteger(addressSplits[2], 1); } RpcAddress rpcAddress = new RpcAddress(ip, port, weight); this.rpcAddresses.add(rpcAddress); } return this; } @Override public T reference() { // 1. 启动 client 端到 server 端的连接信息 // 1.1 为了提升性能,可以将所有的 client=>server 的连接都调整为一个 thread。 // 1.2 初期为了简单,直接使用同步循环的方式。 // 创建 handler // 循环连接 for(RpcAddress rpcAddress : rpcAddresses) { final ChannelHandler channelHandler = new RpcClientHandler(invokeService); final DefaultRpcClientContext context = new DefaultRpcClientContext(); context.address(rpcAddress.address()).port(rpcAddress.port()).channelHandler(channelHandler); ChannelFuture channelFuture = new RpcClient(context).connect(); // 循环同步等待 // 如果出现异常,直接中断?捕获异常继续进行?? channelFutures.add(channelFuture); } // 2. 接口动态代理 ProxyContext proxyContext = buildReferenceProxyContext(); return ReferenceProxy.newProxyInstance(proxyContext); } private ProxyContext buildReferenceProxyContext() { DefaultProxyContext proxyContext = new DefaultProxyContext<>(); proxyContext.serviceId(this.serviceId); proxyContext.serviceInterface(this.serviceInterface); proxyContext.channelFutures(this.channelFutures); proxyContext.invokeService(this.invokeService); return proxyContext; } }
这里主要根据指定的服务端信息,初始化对应的代理实现。
这里还可以拓展指定权重,便于后期负载均衡拓展,本期暂时不做实现。
ReferenceProxy 说明所有的 rpc 调用,客户端只有服务端的接口。
那么,怎么才能和调用本地方法一样调用远程方法呢?
答案就是动态代理。
实现实现如下:
package com.github.houbb.rpc.client.proxy; import com.github.houbb.heaven.util.lang.ObjectUtil; import com.github.houbb.heaven.util.lang.reflect.ReflectMethodUtil; import com.github.houbb.log.integration.core.Log; import com.github.houbb.log.integration.core.LogFactory; import com.github.houbb.rpc.client.proxy.context.ProxyContext; import com.github.houbb.rpc.common.rpc.domain.RpcResponse; import com.github.houbb.rpc.common.rpc.domain.impl.DefaultRpcRequest; import com.github.houbb.rpc.common.support.id.impl.Uuid; import com.github.houbb.rpc.common.support.time.impl.DefaultSystemTime; import io.netty.channel.Channel; import java.lang.reflect.InvocationHandler; import java.lang.reflect.Method; import java.lang.reflect.Proxy; public class ReferenceProxyimplements InvocationHandler { private static final Log LOG = LogFactory.getLog(ReferenceProxy.class); private final ProxyContext proxyContext; private ReferenceProxy(ProxyContext proxyContext) { this.proxyContext = proxyContext; } @Override public Object invoke(Object proxy, Method method, Object[] args) throws Throwable { // 反射信息处理成为 rpcRequest final String seqId = Uuid.getInstance().id(); final long createTime = DefaultSystemTime.getInstance().time(); DefaultRpcRequest rpcRequest = new DefaultRpcRequest(); rpcRequest.serviceId(proxyContext.serviceId()); rpcRequest.seqId(seqId); rpcRequest.createTime(createTime); rpcRequest.paramValues(args); rpcRequest.paramTypeNames(ReflectMethodUtil.getParamTypeNames(method)); rpcRequest.methodName(method.getName()); // 调用远程 LOG.info("[Client] start call remote with request: {}", rpcRequest); proxyContext.invokeService().addRequest(seqId); // 这里使用 load-balance 进行选择 channel 写入。 final Channel channel = getChannel(); LOG.info("[Client] start call channel id: {}", channel.id().asLongText()); // 对于信息的写入,实际上有着严格的要求。 // writeAndFlush 实际是一个异步的操作,直接使用 sync() 可以看到异常信息。 // 支持的必须是 ByteBuf channel.writeAndFlush(rpcRequest).sync(); // 循环获取结果 // 通过 Loop+match wait/notifyAll 来获取 // 分布式根据 redis+queue+loop LOG.info("[Client] start get resp for seqId: {}", seqId); RpcResponse rpcResponse = proxyContext.invokeService().getResponse(seqId); LOG.info("[Client] start get resp for seqId: {}", seqId); Throwable error = rpcResponse.error(); if(ObjectUtil.isNotNull(error)) { throw error; } return rpcResponse.result(); } private Channel getChannel() { return proxyContext.channelFutures().get(0).channel(); } @SuppressWarnings("unchecked") public static T newProxyInstance(ProxyContext proxyContext) { final Class interfaceClass = proxyContext.serviceInterface(); ClassLoader classLoader = interfaceClass.getClassLoader(); Class>[] interfaces = new Class[]{interfaceClass}; ReferenceProxy proxy = new ReferenceProxy(proxyContext); return (T) Proxy.newProxyInstance(classLoader, interfaces, proxy); } }
客户端初始化 newProxyInstance 的就是创建的代理的过程。
客户端调用远程方法,实际上是调用 invoke 的过程。
(1)构建反射 invoke 请求信息,添加 reqId
(2)netty 远程调用服务端
(3)同步获取响应信息
测试 引入 maven测试代码com.github.houbb rpc-client 0.0.6
public static void main(String[] args) {
// 服务配置信息
ReferenceConfig config = new DefaultReferenceConfig();
config.serviceId(ServiceIdConst.CALC);
config.serviceInterface(CalculatorService.class);
config.addresses("localhost:9527");
CalculatorService calculatorService = config.reference();
CalculateRequest request = new CalculateRequest();
request.setOne(10);
request.setTwo(20);
CalculateResponse response = calculatorService.sum(request);
System.out.println(response);
}
测试日志:
[DEBUG] [2021-10-05 14:16:17.534] [main] [c.g.h.l.i.c.LogFactory.setImplementation] - Logging initialized using 'class com.github.houbb.log.integration.adaptors.stdout.StdOutExImpl' adapter.
[INFO] [2021-10-05 14:16:17.625] [main] [c.g.h.r.c.c.RpcClient.connect] - RPC 服务开始启动客户端
...
[INFO] [2021-10-05 14:16:19.328] [main] [c.g.h.r.c.c.RpcClient.connect] - RPC 服务启动客户端完成,监听地址 localhost:9527
[INFO] [2021-10-05 14:16:19.346] [main] [c.g.h.r.c.p.ReferenceProxy.invoke] - [Client] start call remote with request: DefaultRpcRequest{seqId='a525c5a6196545f5a5241b2cdc2ec2c2', createTime=1633414579339, serviceId='calc', methodName='sum', paramTypeNames=[com.github.houbb.rpc.server.facade.model.CalculateRequest], paramValues=[CalculateRequest{one=10, two=20}]}
[INFO] [2021-10-05 14:16:19.347] [main] [c.g.h.r.c.i.i.DefaultInvokeService.addRequest] - [Client] start add request for seqId: a525c5a6196545f5a5241b2cdc2ec2c2
[INFO] [2021-10-05 14:16:19.348] [main] [c.g.h.r.c.p.ReferenceProxy.invoke] - [Client] start call channel id: 00e04cfffe360988-000017bc-00000000-399b9d7e1b88839d-5ccc4a29
十月 05, 2021 2:16:19 下午 io.netty.handler.logging.LoggingHandler write
信息: [id: 0x5ccc4a29, L:/127.0.0.1:50596 - R:localhost/127.0.0.1:9527] WRITE: DefaultRpcRequest{seqId='a525c5a6196545f5a5241b2cdc2ec2c2', createTime=1633414579339, serviceId='calc', methodName='sum', paramTypeNames=[com.github.houbb.rpc.server.facade.model.CalculateRequest], paramValues=[CalculateRequest{one=10, two=20}]}
十月 05, 2021 2:16:19 下午 io.netty.handler.logging.LoggingHandler flush
信息: [id: 0x5ccc4a29, L:/127.0.0.1:50596 - R:localhost/127.0.0.1:9527] FLUSH
[INFO] [2021-10-05 14:16:19.412] [main] [c.g.h.r.c.p.ReferenceProxy.invoke] - [Client] start get resp for seqId: a525c5a6196545f5a5241b2cdc2ec2c2
[INFO] [2021-10-05 14:16:19.413] [main] [c.g.h.r.c.i.i.DefaultInvokeService.getResponse] - [Client] seq a525c5a6196545f5a5241b2cdc2ec2c2 对应结果为空,进入等待
十月 05, 2021 2:16:19 下午 io.netty.handler.logging.LoggingHandler channelRead
信息: [id: 0x5ccc4a29, L:/127.0.0.1:50596 - R:localhost/127.0.0.1:9527] READ: DefaultRpcResponse{seqId='a525c5a6196545f5a5241b2cdc2ec2c2', error=null, result=CalculateResponse{success=true, sum=30}}
...
[INFO] [2021-10-05 14:16:19.505] [nioEventLoopGroup-2-1] [c.g.h.r.c.i.i.DefaultInvokeService.addResponse] - [Client] 获取结果信息,seq: a525c5a6196545f5a5241b2cdc2ec2c2, rpcResponse: DefaultRpcResponse{seqId='a525c5a6196545f5a5241b2cdc2ec2c2', error=null, result=CalculateResponse{success=true, sum=30}}
[INFO] [2021-10-05 14:16:19.505] [nioEventLoopGroup-2-1] [c.g.h.r.c.i.i.DefaultInvokeService.addResponse] - [Client] seq 信息已经放入,通知所有等待方
[INFO] [2021-10-05 14:16:19.506] [nioEventLoopGroup-2-1] [c.g.h.r.c.c.RpcClient.channelRead0] - [Client] response is :DefaultRpcResponse{seqId='a525c5a6196545f5a5241b2cdc2ec2c2', error=null, result=CalculateResponse{success=true, sum=30}}
[INFO] [2021-10-05 14:16:19.506] [main] [c.g.h.r.c.i.i.DefaultInvokeService.getResponse] - [Client] seq a525c5a6196545f5a5241b2cdc2ec2c2 对应结果已经获取: DefaultRpcResponse{seqId='a525c5a6196545f5a5241b2cdc2ec2c2', error=null, result=CalculateResponse{success=true, sum=30}}
[INFO] [2021-10-05 14:16:19.507] [main] [c.g.h.r.c.p.ReferenceProxy.invoke] - [Client] start get resp for seqId: a525c5a6196545f5a5241b2cdc2ec2c2
CalculateResponse{success=true, sum=30}
小结
现在看来有一个小问题,要求服务端必须指定 port,这有点不太合理,比如代理域名,后续需要优化。
这里的启动声明方式也比较基础,后续可以考虑和 spring 进行整合。
为了便于大家学习,以上源码已经开源:
https://github.com/houbb/rpc
希望本文对你有所帮助,如果喜欢,欢迎点赞收藏转发一波。
我是老马,期待与你的下次重逢。



