dubbo 运行机制

源：http://san-yun.iteye.com/blog/1897250
评：写的 调用链很清晰


dubbo是阿里巴巴开源的单一长连接服务框架，底层通信采用nio框架，支持netty,mina,grizzly，默认是netty。对dubbo比较感兴趣的是：
1. client端的线程模型是什么样的？
传统的io client是请求应答模式，发送请求-->等待远程应答。dubbo底层是异步IO的，所有请求复用单一长连接，所以调用都不会阻在IO上，而是阻在Future超时wait上。
2. server端的线程模型是什么样的？
这个比较成熟了，现在一般的server都是基于nio,一批io thread负责处理io,一批worker thread负责处理业务。



一. 快速启动

学习dubbo最好的方式是快速运行起来，由于dubbo还是比较重量级的产品，之前遇到一些问题。
server端：


Java代码 收藏代码

import java.io.IOException;

import com.alibaba.dubbo.config.ApplicationConfig;
import com.alibaba.dubbo.config.ProtocolConfig;
import com.alibaba.dubbo.config.ServiceConfig;
import com.duitang.dboss.client.test.BlogQueryService;
import com.duitang.dboss.client.test.BlogQueryServiceImpl;

public class DubboServerTester {

public static void main(String[] args) throws IOException {
BlogQueryService blogQueryService = new BlogQueryServiceImpl();
ApplicationConfig application = new ApplicationConfig();
application.setName("dubbo-test");

ProtocolConfig protocol = new ProtocolConfig();
protocol.setName("dubbo");
protocol.setPort(8989);
protocol.setThreads(200);

// RegistryConfig registry = new RegistryConfig();
// registry.setAddress("10.20.130.230:9090");
// registry.setUsername("aaa");
// registry.setPassword("bbb");

ServiceConfig<BlogQueryService> service = new ServiceConfig<BlogQueryService>(); // 此实例很重，封装了与注册中心的连接，请自行缓存，否则可能造成内存和连接泄漏
service.setApplication(application);

// service.setRegistry(registry);
service.setRegister(false);
service.setProtocol(protocol); // 多个协议可以用setProtocols()
service.setInterface(BlogQueryService.class);
service.setRef(blogQueryService);
service.setVersion("1.0.0");
// 暴露及注册服务
service.export();

System.out.println("Press any key to exit.");
System.in.read();
}
}

注意：dubbo export服务默认依赖于RegistryConfig，如果没有配置RegistryConfig会报错.可以通过service.setRegister(false)禁用。



client:


Java代码 收藏代码

import java.io.IOException;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.Callable;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.ThreadFactory;
import java.util.concurrent.atomic.AtomicInteger;

import com.alibaba.dubbo.config.ApplicationConfig;
import com.alibaba.dubbo.config.ReferenceConfig;
import com.duitang.dboss.client.test.BlogQueryService;

public class DubboClientTester {

public static void main(String[] args) throws InterruptedException, IOException {
ApplicationConfig application = new ApplicationConfig();
application.setName("dubbo-test");

ReferenceConfig<BlogQueryService> reference = new ReferenceConfig<BlogQueryService>();
reference.setUrl("dubbo://127.0.0.1:8989/com.duitang.dboss.client.test.BlogQueryService");
reference.setTimeout(500);
reference.setConnections(10);
reference.setApplication(application);
reference.setInterface(BlogQueryService.class);
reference.setVersion("1.0.0");
final BlogQueryService blogQueryService = reference.get();

long begin = System.currentTimeMillis();
System.out.println(blogQueryService.test());
long end = System.currentTimeMillis();
System.out.println(" cost:" + (end - begin));

ExecutorService es = Executors.newFixedThreadPool(50, new NamedThreadFactory("my test"));
List<Callable<String>> tasks = new ArrayList<Callable<String>>();
for (int i = 0; i < 100000; ++i) {
tasks.add(new Callable<String>() {

@Override
public String call() throws Exception {
System.out.println("run");
System.out.println(blogQueryService.test());
System.out.println("run success");
return null;
}
});
}
List<Future<String>> futurelist = es.invokeAll(tasks);
for (Future<String> future : futurelist) {
try {
String result = future.get();
} catch (ExecutionException e) {
e.printStackTrace();
}
System.out.println("------------------------------------------------------------------------------------------------------------------------------------------------\r\n");
}
es.shutdown();
System.out.println("end");
System.in.read();
}

static class NamedThreadFactory implements ThreadFactory {

private static final AtomicInteger POOL_SEQ = new AtomicInteger(1);

private final AtomicInteger mThreadNum = new AtomicInteger(1);

private final String mPrefix;

private final boolean mDaemo;

private final ThreadGroup mGroup;

public NamedThreadFactory(){
this("pool-" + POOL_SEQ.getAndIncrement(), false);
}

public NamedThreadFactory(String prefix){
this(prefix, false);
}

public NamedThreadFactory(String prefix, boolean daemo){
mPrefix = prefix + "-thread-";
mDaemo = daemo;
SecurityManager s = System.getSecurityManager();
mGroup = (s == null) ? Thread.currentThread().getThreadGroup() : s.getThreadGroup();
}

public Thread newThread(Runnable runnable) {
String name = mPrefix + mThreadNum.getAndIncrement();
Thread ret = new Thread(mGroup, runnable, name, 0);
ret.setDaemon(mDaemo);
return ret;
}

public ThreadGroup getThreadGroup() {
return mGroup;
}

}
}





1. 通过setUrl("")来实现远程服务直连。
2. 需要注意的是默认connection只有一个，可以通过setConnections()来指定connection pool。在高负载环境下，nio的单连接也会遇到瓶颈，此时你可以通过设置连接池来让更多的连接分担dubbo的请求负载，从而提高系统的吞吐量。”

二. 代码流程
这里重点分析一下client的调用过程，client调用分为三个部分：
1). 初始化，建立连接。
2). 发送请求。
3). 等待远程应答。

(一).初始化
1. DubboProtocol.initClient()
2. Exchangers.connect(URL url, ExchangeHandler handler)
3. Exchangers.getExchanger(url).connect(url, handler)
4. HeaderExchanger.connect(URL url, ExchangeHandler handler)
5. return new HeaderExchangeClient(Transporters.connect(url, new DecodeHandler(new HeaderExchangeHandler(handler))));
6. Transporters.getTransporter().connect(URL url, ChannelHandler handler)
7. NettyTransporter.connect(URL url, ChannelHandler listener)
8. new NettyClient(url, listener) //timeout默认值：timeout=1000;connectTimeout=3000;
9. NettyClient.doOpen() //创建netty的ClientBootstrap
bootstrap = new ClientBootstrap(channelFactory);
bootstrap.setOption("keepAlive", true);
bootstrap.setOption("tcpNoDelay", true);
bootstrap.setOption("connectTimeoutMillis", getTimeout()); //注意：此timeout是timeout,而非connectTimeout
10. AbstractClient.connect()
11. NettyClient.doConnect() //如果远程地址无法连接，抛出timeout异常流程结束。
ChannelFuture future = bootstrap.connect(getConnectAddress());
boolean ret = future.awaitUninterruptibly(getConnectTimeout(), TimeUnit.MILLISECONDS);

(二).发送请求
1.DubboInvoker.doInvoke(Invocation invocation) //currentClient.request(invocation, timeout).get()
2.HeaderExchangeClient.request(invocation, timeout)
3.HeaderExchangeChannel.request(Invocation invocation,timeout)
4.AbstractPeer.send(Request request)
5.NettyChannel.send(Object message, boolean sent)
6.NioClientSocketChannel.write(message)
7.NettyHandler.writeRequested(ChannelHandlerContext ctx, MessageEvent e)
8.AbstractPeer.sent(Channel ch, Request request)

(三).等待远程应答
在调用DubboInvoker.doInvoke(Invocation invocation)中实际是调用currentClient.request(invocation, timeout).get()，此方法会返回DefaultFuture,调用get方法会阻塞直到超时，在阻塞的同时netty的io线程会接收到远程应答，如果收到响应会产生io事件调用NettyHandler.messageReceived。

1.NettyHandler.messageReceived(ChannelHandlerContext ctx, MessageEvent e)
2.AbstractPeer.received(Channel ch, Object msg)
3.MultiMessageHandler.received(Channel channel, Object message)
4.AllChannelHandler.received(Channel channel, Object message)
5.DecodeHandler.received(Channel channel, Object message)
6.HeaderExchangeHandler.received(Channel channel, Object message)
7.DefaultFuture.received(Channel channel, Response response) //注意是static方法
DefaultFuture future = FUTURES.remove(response.getId());
if (future != null) {
future.doReceived(response);
}

三. dubbo client的核心


我认为dubbo client的核心在DefaultFuture。所以远程调用都不会阻在IO上，而是阻在Future超时wait上，下面忽略掉远程调用把future抽取出来。




下面是代码实现
Java代码 收藏代码

package executor;

import java.util.concurrent.Callable;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.atomic.AtomicLong;

public class Commands {

private ExecutorService senders = Executors.newCachedThreadPool();
private ExecutorService receviers = Executors.newCachedThreadPool();
private AtomicLong counter = new AtomicLong();

public CommandResponse execute(Callable<Object> task, int timeout) {
Future<Object> result = senders.submit(task);
long id = counter.getAndIncrement();
CommandFuture commandFuture = new CommandFuture(id);
receviers.submit(new ReceiveWorker(id, result));
return commandFuture.get(timeout);
}

static class ReceiveWorker implements Runnable {

private Future<Object> result;
private Long id;

public ReceiveWorker(Long id, Future<Object> result){
super();
this.result = result;
this.id = id;
}

@Override
public void run() {
try {
Object obj = result.get();
CommandFuture.received(new CommandResponse(id, obj));
} catch (InterruptedException e) {
e.printStackTrace();
} catch (ExecutionException e) {
e.printStackTrace();
}
}
}

public void shutdown() {
senders.shutdown();
receviers.shutdown();
}
}


Java代码 收藏代码

package executor;

import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.Condition;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantLock;

public class CommandFuture {

private final Lock lock = new ReentrantLock();

private final Condition done = lock.newCondition();

private CommandResponse response;

private static final Map<Long, CommandFuture> FUTURES = new ConcurrentHashMap<Long, CommandFuture>();


public CommandFuture(Long id){
FUTURES.put(id, this);
}

public boolean isDone() {
return response != null;
}

public CommandResponse get(int timeout) {

if (!isDone()) {
long start = System.currentTimeMillis();
lock.lock();
try {
while (!isDone()) {
done.await(timeout, TimeUnit.MILLISECONDS);
if (isDone() || System.currentTimeMillis() - start >= timeout) {
break;
}
}
} catch (InterruptedException e) {
throw new RuntimeException(e);
} finally {
lock.unlock();
}
if (!isDone()) {
throw new TimeoutException("timeout");
}
}
return response;
}

public void doReceived(CommandResponse response) {
lock.lock();
try {
this.response = response;
if (done != null) {
done.signal();
}
} finally {
lock.unlock();
}

}

public static void received(CommandResponse response) {
try {
CommandFuture future = FUTURES.remove(response.getId());
if (future != null) {
future.doReceived(response);
} else {
System.out.println("some error!");
}
} finally {
// CHANNELS.remove(response.getId());
}
}
}


Java代码 收藏代码

package executor;

import java.util.concurrent.Callable;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.atomic.AtomicLong;

public class Commands {

private ExecutorService senders = Executors.newCachedThreadPool();
private ExecutorService receviers = Executors.newCachedThreadPool();
private AtomicLong counter = new AtomicLong();

public CommandResponse execute(Callable<Object> task, int timeout) {
Future<Object> result = senders.submit(task);
long id = counter.getAndIncrement();
CommandFuture commandFuture = new CommandFuture(id);
receviers.submit(new ReceiveWorker(id, result));
return commandFuture.get(timeout);
}

static class ReceiveWorker implements Runnable {

private Future<Object> result;
private Long id;

public ReceiveWorker(Long id, Future<Object> result){
super();
this.result = result;
this.id = id;
}

@Override
public void run() {
try {
Object obj = result.get();
CommandFuture.received(new CommandResponse(id, obj));
} catch (InterruptedException e) {
e.printStackTrace();
} catch (ExecutionException e) {
e.printStackTrace();
}
}
}

public void shutdown() {
senders.shutdown();
receviers.shutdown();
}
}



下面是jstack