dubbo服务调用过程

3,222 阅读8分钟

参考资料:

总结

dubbo服务调用主要分为两个过程,分别是消费者发送请求和接收响应结果,提供者接收请求。

消费者端:

  • 发送请求:服务接口的代理对象执行目标方法,被InvokerInvocationHandler#invoke方法拦截,经过路由过滤、负载均衡后选择一个DubboInvoker对象,调用doInvoke方法。创建一个Request对象,并生成全局唯一的请求ID,接着实例化一个DefaultFuture对象,将请求ID作为key,把DefaultFuture保存到一个ConcurrentHashMap。最后,通过NettyClient把封装了目标方法信息的RpcInvocation序列化后发送出去
  • 接收响应:先通过响应ID,即请求ID,在缓存中找到对应的Future,执行doReceived方法。保存结果,接着唤醒对应的请求线程来处理响应结果

提供者端:

NettyServer接收到请求后,根据协议得到信息并反序列化成对象,派发到线程池等待处理。信息会被封转成ChannelEventRunnable对象,类型为RECEIVED。工作线程最终会调用DubboProtocol#reply方法,根据port、path、version、group构建serviceKey,从缓存中找到对应Exporter,经过层层调用,最后会找到真正实现类,执行目标方法返回结果。

image-20201225152029705

调用流程图

image-20201224160303845

客户端发起本地调用,实际上调用的是代理类,代理类通过远程客户端(默认是NettyClient)发起请求。先是构建协议头,指定通信协议、序列化器类型和body长度,接着将Java对象序列化成协议体,然后发送数据。

服务端(NettyServer)接收请求,分发给业务线程池处理,由业务线程找到对应的实现类,执行相应方法并返回结果。

这里省略了应用层的通信协议和序列化器的知识点,具体可查看参考资料。

客户端源码分析

上篇文章说到,消费者引用服务时,ReferenceConfig#init会通过动态代理创建一个代理对象,所以当我们调用服务接口的方法时,都会被InvokerInvocationHandler#invoke拦截

image-20201224165013621

com.alibaba.dubbo.rpc.proxy.InvokerInvocationHandler#invoke

@Override
public Object invoke(Object proxy, Method method, Object[] args) throws Throwable {
    String methodName = method.getName();
    Class<?>[] parameterTypes = method.getParameterTypes();
    if (method.getDeclaringClass() == Object.class) {
        return method.invoke(invoker, args);
    }
    // ... 省略部分代码
    
    // 先创建RpcInvocation对象,然后调用MockClusterInvoker#invoke
    // Invocation类的作用:基本上都是 将原本由InvocationHandler调用目标方法的执行时机交给其他类决定
    return invoker.invoke(new RpcInvocation(method, args)).recreate();
}

我们顺着代码,看下执行过程。

com.alibaba.dubbo.rpc.cluster.support.wrapper.MockClusterInvoker#invoke

 @Override
public Result invoke(Invocation invocation) throws RpcException {
    Result result = null;
    // 获取mock配置
    String value = directory.getUrl()
        			.getMethodParameter(invocation.getMethodName(), Constants.MOCK_KEY, Boolean.FALSE.toString()).trim();
    
    if (value.length() == 0 || value.equalsIgnoreCase("false")) {
        //no mock
        // 调用AbstractClusterInvoker
        result = this.invoker.invoke(invocation);
    }
    // ..... 省略部分代码
    return result;
}

com.alibaba.dubbo.rpc.cluster.support.AbstractClusterInvoker#invoke

@Override
public Result invoke(final Invocation invocation) throws RpcException {
    // 检查是否被销毁
    checkWhetherDestroyed();
    LoadBalance loadbalance = null;

    // binding attachments into invocation.
    // 看下上下文有没有attachments,有的话绑定到Invocation
    Map<String, String> contextAttachments = RpcContext.getContext().getAttachments();
    if (contextAttachments != null && contextAttachments.size() != 0) {
        ((RpcInvocation) invocation).addAttachments(contextAttachments);
    }

    // 调用directory#list,里面做的是路由过滤
    List<Invoker<T>> invokers = list(invocation);
    
    if (invokers != null && !invokers.isEmpty()) {
        // 通过spi获取loadbalance实现类
        loadbalance = ExtensionLoader.getExtensionLoader(LoadBalance.class).getExtension(
            invokers.get(0).getUrl().getMethodParameter(
                RpcUtils.getMethodName(invocation), Constants.LOADBALANCE_KEY, Constants.DEFAULT_LOADBALANCE));
    }
    RpcUtils.attachInvocationIdIfAsync(getUrl(), invocation);
    
    // 最终调用DubboInvoker#doInvoke方法,通过负载均衡策略选择一个Invoker
    return doInvoke(invocation, invokers, loadbalance);
}


protected List<Invoker<T>> list(Invocation invocation) throws RpcException {
    List<Invoker<T>> invokers = directory.list(invocation);
    return invokers;
}

DubboInvoker#doInvoke

我们先忽略路由、集群、负载均衡的相关知识点,重点看com.alibaba.dubbo.rpc.protocol.dubbo.DubboInvoker#doInvoke

@Override
protected Result doInvoke(final Invocation invocation) throws Throwable {
    RpcInvocation inv = (RpcInvocation) invocation;
    final String methodName = RpcUtils.getMethodName(invocation);
    // 设置path和version到attachment
    inv.setAttachment(Constants.PATH_KEY, getUrl().getPath());
    inv.setAttachment(Constants.VERSION_KEY, version);

    // 选择一个客户端
    ExchangeClient currentClient;
    if (clients.length == 1) {
        currentClient = clients[0];
    } else {
        currentClient = clients[index.getAndIncrement() % clients.length];
    }
    try {
        // 异步调用标记
        boolean isAsync = RpcUtils.isAsync(getUrl(), invocation);
        // oneway发送方式标记
        boolean isOneway = RpcUtils.isOneway(getUrl(), invocation);
        // 超时时间
        int timeout = getUrl().getMethodParameter(methodName, Constants.TIMEOUT_KEY, Constants.DEFAULT_TIMEOUT);

        if (isOneway) {
            // oneway方式发送,不管发送结果
            boolean isSent = getUrl().getMethodParameter(methodName, Constants.SENT_KEY, false);
            currentClient.send(inv, isSent);
            RpcContext.getContext().setFuture(null);
            return new RpcResult();
        } else if (isAsync) {
            // 异步发送
            ResponseFuture future = currentClient.request(inv, timeout);
            RpcContext.getContext().setFuture(new FutureAdapter<Object>(future));
            return new RpcResult();
        } else {
            RpcContext.getContext().setFuture(null);
            // 同步发送,直接调用future.get()阻塞等待结果
            // 异步和同步的区别是,future.get()是用户调用还是组件调用
            return (Result) currentClient.request(inv, timeout).get();
        }
    } 
    // ... 省略部分代码
}

调用的三种方式

从上面的代码可以看到调用一共分为三种方式,包括oneway、异步、同步。

  • **oneway:**不需要关心发送结果时,可以选择这种方式,消耗最小,啥都不用管。
  • **异步调用:**client发送请求后返回一个ResponseFuture,然后塞到上下文。在用户需要结果时,通过上下文获取future,调用future#get得到请求结果
  • **同步调用:**这是我们最常用的,其实只不过组件帮我们手动调用future#get方法,使得给用户的感觉好像同步调用一样

发送请求

发送请求currentClient.request(inv, timeout)的执行过程是,创建一个请求对象Request,设置版本号和请求内容,构造Future对象并缓存,调用NettyClient发送请求。

com.alibaba.dubbo.remoting.exchange.support.header.HeaderExchangeChannel#request(java.lang.Object, int)

@Override
public ResponseFuture request(Object request, int timeout) throws RemotingException {
    // ... 省略部分代码
    
    // create request.
    // 创建请求对象,同时生成唯一的请求ID
    Request req = new Request();
    req.setVersion(Version.getProtocolVersion());
    req.setTwoWay(true);
    req.setData(request);
    
    // 新建并缓存Future
    DefaultFuture future = new DefaultFuture(channel, req, timeout);
    try {
        // 最终调用NettyChannel#send
        channel.send(req);
    } catch (RemotingException e) {
        future.cancel();
        throw e;
    }
    return future;
}

构建Request

com.alibaba.dubbo.remoting.exchange.Request

public class Request {

    private static final AtomicLong INVOKE_ID = new AtomicLong(0);

    private final long mId;

    public Request() {
        mId = newId();
    }

    private static long newId() {
        // getAndIncrement() When it grows to MAX_VALUE, it will grow to MIN_VALUE, and the negative can be used as ID
        return INVOKE_ID.getAndIncrement();
    }
    
    public long getId() {
        return mId;
    }

	// ...省略部分代码
}

构建Future

com.alibaba.dubbo.remoting.exchange.support.DefaultFuture#DefaultFuture

public DefaultFuture(Channel channel, Request request, int timeout) {
    this.channel = channel;
    this.request = request;
    // 生成请求id,作为key,用于接收响应结果时从Map中获取到对应的Future和Channel
    this.id = request.getId();
    this.timeout = timeout > 0 ?
        			timeout : channel.getUrl().getPositiveParameter(Constants.TIMEOUT_KEY, Constants.DEFAULT_TIMEOUT);
    // put into waiting map.
    FUTURES.put(id, this);
    CHANNELS.put(id, channel);
}

// FUTURES和CHANNELS都是一个ConcurrentHashMap
private static final Map<Long, Channel> CHANNELS = new ConcurrentHashMap<Long, Channel>();
private static final Map<Long, DefaultFuture> FUTURES = new ConcurrentHashMap<Long, DefaultFuture>();

可以看到,当我们创建请求对象时,Request内部会调用AtomicLong#getAndIncrement生成一个全局唯一的ID。当创建DefaultFuture对象时,将请求ID作为key,把Future保存到Map。当接收响应结果时,最终调用的是DefaultFuture#received来接收结果

接收响应结果

com.alibaba.dubbo.remoting.exchange.support.DefaultFuture#received

public static void received(Channel channel, Response response) {
    try {
        // 响应返回的id,就是服务端请求时发送的id
        // 根据请求id到缓存找到对应的Future
        DefaultFuture future = FUTURES.remove(response.getId());
        if (future != null) {
            future.doReceived(response);
        } else {
            logger.warn("The timeout response finally returned at "
                        + (new SimpleDateFormat("yyyy-MM-dd HH:mm:ss.SSS").format(new Date()))
                        + ", response " + response
                        + (channel == null ? "" : ", channel: " + channel.getLocalAddress()
                           + " -> " + channel.getRemoteAddress()));
        }
    } finally {
        CHANNELS.remove(response.getId());
    }
}


private void doReceived(Response res) {
    lock.lock();
    try {
        // 接收响应结果
        response = res;
        if (done != null) {
            // 唤醒请求线程
            done.signal();
        }
    } finally {
        lock.unlock();
    }
    if (callback != null) {
        invokeCallback(callback);
    }
}

客户端接收到响应结果的处理过程:**根据响应结果ID,也就是请求ID,到缓存中找到对应的Future对象,调用doReceived方法,保存响应结果,并唤醒对应的请求线程。**过程如下图所示:

image-20201225112559980

小结

主要过程:

1.服务接口的代理对象执行目标方法,被InvokerInvocationHandler#invoke方法拦截,进入代理过程。经过路由过滤、负载均衡,选择一个DubboInvoker,执行doInvoker方法。先是创建一个请求对象(Request),并生成全局唯一的ID。接着,创建DefaultFuture,并保存到ConcurrentHashMap中,key是请求ID。最后通过NettyClient把封装了目标方法信息的RpcInvocation序列化成消息发送出去

2.接收到返回的响应结果,调用DefaultFuture#received方法,根据返回的请求ID,找到对应的DefaultFuture。接着调用doReceived方法,保存响应结果,并唤醒请求线程。最后,解析并返回执行结果

服务端源码分析

服务端接收到请求后,会解析得到消息。消息有五种派发策略:

image-20201225120055547

默认是all,也就是所有消息都派发到业务线程池中,我们来看下AllChannelHandler的实现

com.alibaba.dubbo.remoting.transport.dispatcher.all.AllChannelHandler#received

@Override
public void received(Channel channel, Object message) throws RemotingException {
    // 获取线程池
    // Executors.newCachedThreadPool(new NamedThreadFactory("DubboSharedHandler", true));
    ExecutorService cexecutor = getExecutorService();
    try {
        // ChannelEventRunnable#run方法,包含多种类型处理
        cexecutor.execute(new ChannelEventRunnable(channel, handler, ChannelState.RECEIVED, message));
    }
    // ...省略部分代码
}

ChannelEventRunnable

ChannelEventRunnable实现Runnable接口,我们看下它的run方法。

com.alibaba.dubbo.remoting.transport.dispatcher.ChannelEventRunnable#run

@Override
public void run() {
    if (state == ChannelState.RECEIVED) {
        try {
            handler.received(channel, message);
        } catch (Exception e) {
            logger.warn("ChannelEventRunnable handle " + state + " operation error, channel is " + channel
                        + ", message is " + message, e);
        }
    } else {
        switch (state) {
            case CONNECTED:
                try {
                    handler.connected(channel);
                } catch (Exception e) {
                    logger.warn("ChannelEventRunnable handle " + state + " operation error, channel is " + channel, e);
                }
                break;
            case DISCONNECTED:
                try {
                    handler.disconnected(channel);
                } catch (Exception e) {
                    logger.warn("ChannelEventRunnable handle " + state + " operation error, channel is " + channel, e);
                }
                break;
            case SENT:
                try {
                    handler.sent(channel, message);
                } catch (Exception e) {
                    logger.warn("ChannelEventRunnable handle " + state + " operation error, channel is " + channel
                                + ", message is " + message, e);
                }
            case CAUGHT:
                try {
                    handler.caught(channel, exception);
                } catch (Exception e) {
                    logger.warn("ChannelEventRunnable handle " + state + " operation error, channel is " + channel
                                + ", message is: " + message + ", exception is " + exception, e);
                }
                break;
            default:
                logger.warn("unknown state: " + state + ", message is " + message);
        }
    }

}

结合上面的代码,服务端接收到请求后解析成消息,接着获取线程池,把消息封装成ChannelEventHandler,类型是 ChannelState.RECEIVED。当空闲线程处理请求时,执行ChannelEventRunnable#received方法,最终调用HeaderExchangeHandler#handleRequest

HeaderExchangeHandler

com.alibaba.dubbo.remoting.exchange.support.header.HeaderExchangeHandler#handleRequest

Response handleRequest(ExchangeChannel channel, Request req) throws RemotingException {
    // 封装一个响应对象,保存请求ID
    Response res = new Response(req.getId(), req.getVersion());
   	
    // ...省略部分代码
    
    // find handler by message class.
    // 获取请求消息,比如方法名称、参数类型、参数值。也就是封装了目标方法信息的RpcInvocation
    Object msg = req.getData();
    try {
        // handle data.
        // 最终调用DubboProtocol#reply
        Object result = handler.reply(channel, msg);
        res.setStatus(Response.OK);
        res.setResult(result);
    } catch (Throwable e) {
        res.setStatus(Response.SERVICE_ERROR);
        res.setErrorMessage(StringUtils.toString(e));
    }
    return res;
}

image-20201225160939134

DubboProtocol#reply

这是个关键方法,源码如下

com.alibaba.dubbo.remoting.exchange.support.ExchangeHandlerAdapter#reply

@Override
public Object reply(ExchangeChannel channel, Object message) throws RemotingException {
    if (message instanceof Invocation) {
        Invocation inv = (Invocation) message;
        // 找到Invoker
        Invoker<?> invoker = getInvoker(channel, inv);
        // ... 省略部分代码
        RpcContext.getContext().setRemoteAddress(channel.getRemoteAddress());
        // 最终找到真正实现类,调用目标方法
        return invoker.invoke(inv);
    }
}

Invoker<?> getInvoker(Channel channel, Invocation inv) throws RemotingException {
    // ... 省略部分代码
    
    // 构建key,生成规则:group/serviceName:version:port
    String serviceKey = serviceKey(port, path, inv.getAttachments().get(Constants.VERSION_KEY), 
                                   											inv.getAttachments().get(Constants.GROUP_KEY));
    // 根据key,从之前暴露服务的Map中找到对应的Exporter
    DubboExporter<?> exporter = (DubboExporter<?>) exporterMap.get(serviceKey);
    // 返回Exporter封装的Invoker
    return exporter.getInvoker();
}

关键是serviceKey,在服务暴露时,生成的exporter会保存到exporterMap,它的key也是serviceKey。

image-20201225142220523

小结

主要过程:NettyServer接收到请求后,解析成信息,并全部派发到线程池,封转成ChannelEventRunnable对象,等待线程执行。工作线程调用DubboProtocol#reply方法,构建serviceKey,从缓存中找到对应的Exporter,并执行invoke方法,最终找到真正实现类,执行目标方法并返回结果。