Hadoop2源码分析－RPC探索实战

1.概述

　　在《Hadoop2源码分析－RPC机制初识》博客中，我们对RPC机制有了初步的认识和了解，下面我们对Hadoop V2的RPC机制做进一步探索，在研究Hadoop V2的RPC机制，我们需要掌握相关的Java基础知识，如：Java NIO、动态代理与反射等。本篇博客介绍的内容目录如下所示：

• Java NIO简述
• Java NIO实例演示
• 动态代理与反射简述
• 动态代理与反射实例演示
• Hadoop V2 RPC框架使用实例

　　下面开始今天的博客介绍。

2.Java NIO简述

　　Java NIO又称Java New IO，它替代了Java IO API，提供了与标准IO不同的IO工作方式。Java NIO由一下核心组件组成：

• Channels：连接通道，即能从通道读取数据，又能写数据到通道。可以异步读写，读写从Buffer开始。
• Buffers：消息缓冲区，用于和NIO通道进行交互。所谓缓冲区，它是一块可以读写的内存，该内存被封装成NIO的Buffer对象，并提供相应的方法，以便于访问。
• Selectors：通道管理器，它能检测到Java NIO中多个通道，单独的线程可以管理多个通道，间接的管理多个网络连接。

　　下图为Java NIO的工作原理图，如下图所示：

3.Java NIO实例演示

• NIOServer

　　首先，我们来看NIOServer的代码块。代码内容如下所示：

package cn.hadoop.nio;

import java.io.IOException;
import java.net.InetSocketAddress;
import java.nio.ByteBuffer;
import java.nio.channels.SelectionKey;
import java.nio.channels.Selector;
import java.nio.channels.ServerSocketChannel;
import java.nio.channels.SocketChannel;
import java.util.Iterator;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import cn.hadoop.conf.ConfigureAPI;

/**
 * @Date May 8, 2015
 *
 * @Author dengjie
 *
 * @Note Defined nio server
 */
public class NIOServer {

    private static final Logger LOGGER = LoggerFactory.getLogger(NIOServer.class);

    // The channel manager
    private Selector selector;

    /**
     * Get ServerSocket channel and initialize
     * 
     * 1.Get a ServerSocket channel
     * 
     * 2.Set channel for non blocking
     * 
     * 3.The channel corresponding to the ServerSocket binding to port port
     * 
     * 4.Get a channel manager
     * 
     * 5.The channel manager and the channel binding, and the channel registered
     * SelectionKey.OP_ACCEPT event
     * 
     * @param port
     * @throws IOException
     */
    public void init(int port) throws IOException {
        ServerSocketChannel serverChannel = ServerSocketChannel.open();
        serverChannel.configureBlocking(false);
        serverChannel.socket().bind(new InetSocketAddress(port));
        this.selector = Selector.open();
        serverChannel.register(selector, SelectionKey.OP_ACCEPT);
    }

    /**
     * listen selector
     * 
     * @throws IOException
     */
    public void listen() throws IOException {
        LOGGER.info("Server has start success");
        while (true) {
            selector.select();
            Iterator<SelectionKey> ite = this.selector.selectedKeys().iterator();
            while (ite.hasNext()) {
                SelectionKey key = (SelectionKey) ite.next();
                ite.remove();
                if (key.isAcceptable()) {
                    ServerSocketChannel server = (ServerSocketChannel) key.channel();
                    SocketChannel channel = server.accept();
                    channel.configureBlocking(false);// 非阻塞
                    channel.write(ByteBuffer.wrap(new String("Send test info to client").getBytes()));
                    channel.register(this.selector, SelectionKey.OP_READ);// 设置读的权限
                } else if (key.isReadable()) {
                    read(key);
                }
            }
        }
    }

    /**
     * Deal client send event
     */
    public void read(SelectionKey key) throws IOException {
        SocketChannel channel = (SocketChannel) key.channel();
        ByteBuffer buffer = ByteBuffer.allocate(1024);
        channel.read(buffer);
        byte[] data = buffer.array();
        String info = new String(data).trim();
        LOGGER.info("Server receive info : " + info);
        ByteBuffer outBuffer = ByteBuffer.wrap(info.getBytes());
        channel.write(outBuffer);// 将消息回送给客户端
    }

    public static void main(String[] args) {
        try {
            NIOServer server = new NIOServer();
            server.init(ConfigureAPI.ServerAddress.NIO_PORT);
            server.listen();
        } catch (Exception ex) {
            ex.printStackTrace();
            LOGGER.error("NIOServer main run error,info is " + ex.getMessage());
        }
    }
}

• NIOClient

　　然后，我们在来看NIOClient的代码块，代码具体内容如下所示：

package cn.hadoop.nio;

import java.io.IOException;
import java.net.InetSocketAddress;
import java.nio.ByteBuffer;
import java.nio.channels.SelectionKey;
import java.nio.channels.Selector;
import java.nio.channels.SocketChannel;
import java.util.Iterator;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import cn.hadoop.conf.ConfigureAPI;

/**
 * @Date May 8, 2015
 *
 * @Author dengjie
 *
 * @Note Defined NIO client
 */
public class NIOClient {

    private static final Logger LOGGER = LoggerFactory.getLogger(NIOClient.class);

    private Selector selector;

    /**
     * Get ServerSocket channel and initialize
     */
    public void init(String ip, int port) throws Exception {
        SocketChannel channel = SocketChannel.open();
        channel.configureBlocking(false);
        this.selector = Selector.open();
        channel.connect(new InetSocketAddress(ip, port));
        channel.register(selector, SelectionKey.OP_CONNECT);
    }

    /**
     * listen selector
     */
    public void listen() throws Exception {
        while (true) {
            selector.select();
            Iterator<SelectionKey> ite = this.selector.selectedKeys().iterator();
            while (ite.hasNext()) {
                SelectionKey key = (SelectionKey) ite.next();
                ite.remove();
                if (key.isConnectable()) {
                    SocketChannel channel = (SocketChannel) key.channel();
                    if (channel.isConnectionPending()) {
                        channel.finishConnect();
                    }
                    channel.configureBlocking(false);// 非阻塞

                    channel.write(ByteBuffer.wrap(new String("Send test info to server").getBytes()));
                    channel.register(this.selector, SelectionKey.OP_READ);
                } else if (key.isReadable()) {
                    read(key);
                }

            }

        }
    }

    /**
     * Deal client send event
     */
    public void read(SelectionKey key) throws IOException {
        SocketChannel channel = (SocketChannel) key.channel();
        ByteBuffer buffer = ByteBuffer.allocate(1024);
        channel.read(buffer);
        byte[] data = buffer.array();
        String info = new String(data).trim();
        LOGGER.info("Client receive info : " + info);
        ByteBuffer outBuffer = ByteBuffer.wrap(info.getBytes());
        channel.write(outBuffer);
    }

    public static void main(String[] args) {
        try {
            NIOClient client = new NIOClient();
            client.init(ConfigureAPI.ServerAddress.NIO_IP, ConfigureAPI.ServerAddress.NIO_PORT);
            client.listen();
        } catch (Exception ex) {
            ex.printStackTrace();
            LOGGER.error("NIOClient main run has error,info is " + ex.getMessage());
        }
    }
}

• ConfigureAPI

　　下面给出ConfigureAPI类的代码，内容如下所示：

package cn.hadoop.conf;

/**
 * @Date May 7, 2015
 *
 * @Author dengjie
 *
 * @Note Defined rpc info
 */
public class ConfigureAPI {

    public interface VersionID {
        public static final long RPC_VERSION = 7788L;
    }

    public interface ServerAddress {
        public static final int NIO_PORT = 8888;
        public static final String NIO_IP = "127.0.0.1";
    }

}

4.动态代理和反射简述

　　在Java中，动态代理主要用来做方法的增强，可以在不修改源码的情况下，增强一些方法。另外，还有一个作用就是做远程调用，比如现在有Java接口，该接口的实现部署在非本地服务器上，在编写客户端代码时，由于没法直接生成该对象，这个时候就需要考虑使用动态代理了。

　　而反射，利用了Class类作为反射实例化对象的基本应用，对于一个实例化对象而言，它需要调用类中的构造方法，属性和一般方法，这些操作都可以通过反射机制来完成。下面我们用一个实例来理解这些理论。

5.动态代理和反射实例演示

5.1动态代理

• JProxy

package cn.java.base;

import java.lang.reflect.InvocationHandler;
import java.lang.reflect.Method;
import java.lang.reflect.Proxy;

/**
 * @Date May 7, 2015
 * 
 * @Author dengjie
 */
public class JProxy {
    
    public static void main(String[] args) {
        JInvocationHandler ji = new JInvocationHandler();
        Subject sub = (Subject) ji.bind(new RealSubject());
        System.out.println(sub.say("dengjie", 25));
    }

}

interface Subject {
    public String say(String name, int age);
}

class RealSubject implements Subject {

    @Override
    public String say(String name, int age) {
        return name + "," + age;
    }

}

class JInvocationHandler implements InvocationHandler {

    private Object object = null;

    public Object bind(Object object) {
        this.object = object;
        return Proxy.newProxyInstance(object.getClass().getClassLoader(), object.getClass().getInterfaces(), this);
    }

    @Override
    public Object invoke(Object proxy, Method method, Object[] args) throws Throwable {
        Object tmp = method.invoke(this.object, args);
        return tmp;
    }

}

5.2反射

• JReflect

package cn.java.base;

/**
 * @Date May 7, 2015
 * 
 * @Author dengjie
 */
public class JReflect {
    public static void main(String[] args) {
        Fruit f = Factory.getInstance(Orange.class.getName());
        if (f != null) {
            f.eat();
        }
    }
}

interface Fruit {
    public abstract void eat();
}

class Apple implements Fruit {

    @Override
    public void eat() {
        System.out.println("apple");
    }

}

class Orange implements Fruit {

    @Override
    public void eat() {
        System.out.println("orange");
    }

}

class Factory {
    public static Fruit getInstance(String className) {
        Fruit f = null;
        try {
            f = (Fruit) Class.forName(className).newInstance();
        } catch (Exception e) {
            e.printStackTrace();
        }
        return f;
    }
}

6.Hadoop V2 RPC框架使用实例

　　本实例主要演示通过Hadoop V2的RPC框架实现一个计算两个整数的Add和Sub，服务接口为 CaculateService ，继承于 VersionedProtocol ，具体代码如下所示：

• CaculateService

package cn.hadoop.service;

import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.ipc.ProtocolInfo;
import org.apache.hadoop.ipc.VersionedProtocol;

import cn.hadoop.conf.ConfigureAPI;

/**
 * @Date May 7, 2015
 *
 * @Author dengjie
 *
 * @Note Data calculate service interface
 */
@ProtocolInfo(protocolName = "", protocolVersion = ConfigureAPI.VersionID.RPC_VERSION)
public interface CaculateService extends VersionedProtocol {

    // defined add function
    public IntWritable add(IntWritable arg1, IntWritable arg2);

    // defined sub function
    public IntWritable sub(IntWritable arg1, IntWritable arg2);

}

　　注意，本工程使用的是Hadoop-2.6.0版本，这里CaculateService接口需要加入注解，来声明版本号。

　　CaculateServiceImpl类实现CaculateService接口。代码如下所示：

• CaculateServiceImpl

package cn.hadoop.service.impl;

import java.io.IOException;

import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.ipc.ProtocolSignature;

import cn.hadoop.conf.ConfigureAPI;
import cn.hadoop.service.CaculateService;

/**
 * @Date May 7, 2015
 *
 * @Author dengjie
 *
 * @Note Implements CaculateService class
 */
public class CaculateServiceImpl implements CaculateService {

    public ProtocolSignature getProtocolSignature(String arg0, long arg1, int arg2) throws IOException {
        return this.getProtocolSignature(arg0, arg1, arg2);
    }

    /**
     * Check the corresponding version
     */
    public long getProtocolVersion(String arg0, long arg1) throws IOException {
        return ConfigureAPI.VersionID.RPC_VERSION;
    }

    /**
     * Add nums
     */
    public IntWritable add(IntWritable arg1, IntWritable arg2) {
        return new IntWritable(arg1.get() + arg2.get());
    }

    /**
     * Sub nums
     */
    public IntWritable sub(IntWritable arg1, IntWritable arg2) {
        return new IntWritable(arg1.get() - arg2.get());
    }

}

　　CaculateServer服务类，对外提供服务，具体代码如下所示：

• CaculateServer

package cn.hadoop.rpc;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.ipc.RPC;
import org.apache.hadoop.ipc.RPC.Server;
import org.slf4j.LoggerFactory;
import org.slf4j.Logger;

import cn.hadoop.service.CaculateService;
import cn.hadoop.service.impl.CaculateServiceImpl;

/**
 * @Date May 7, 2015
 *
 * @Author dengjie
 *
 * @Note Server Main
 */
public class CaculateServer {

    private static final Logger LOGGER = LoggerFactory.getLogger(CaculateServer.class);

    public static final int IPC_PORT = 9090;

    public static void main(String[] args) {
        try {
            Server server = new RPC.Builder(new Configuration()).setProtocol(CaculateService.class)
                    .setBindAddress("127.0.0.1").setPort(IPC_PORT).setInstance(new CaculateServiceImpl()).build();
            server.start();
            LOGGER.info("CaculateServer has started");
            System.in.read();
        } catch (Exception ex) {
            ex.printStackTrace();
            LOGGER.error("CaculateServer server error,message is " + ex.getMessage());
        }
    }

}

　　注意，在Hadoop V2版本中，获取RPC下的Server对象不能在使用RPC.getServer()方法了，该方法已被移除，取而代之的是使用Builder方法来构建新的Server对象。

　　RPCClient客户端类，用于访问Server端，具体代码实现如下所示：

• RPCClient

package cn.hadoop.rpc;

import java.net.InetSocketAddress;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.ipc.RPC;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import cn.hadoop.service.CaculateService;

/**
 * @Date May 7, 2015
 *
 * @Author dengjie
 *
 * @Note RPC Client Main
 */
public class RPCClient {

    private static final Logger LOGGER = LoggerFactory.getLogger(RPCClient.class);

    public static void main(String[] args) {
        InetSocketAddress addr = new InetSocketAddress("127.0.0.1", CaculateServer.IPC_PORT);
        try {
            RPC.getProtocolVersion(CaculateService.class);
            CaculateService service = (CaculateService) RPC.getProxy(CaculateService.class,
                    RPC.getProtocolVersion(CaculateService.class), addr, new Configuration());
            int add = service.add(new IntWritable(2), new IntWritable(3)).get();
            int sub = service.sub(new IntWritable(5), new IntWritable(2)).get();
            LOGGER.info("2+3=" + add);
            LOGGER.info("5-2=" + sub);
        } catch (Exception ex) {
            ex.printStackTrace();
            LOGGER.error("Client has error,info is " + ex.getMessage());
        }
    }

}

　　Hadoop V2 RPC服务端截图预览，如下所示：

　　Hadoop V2 RPC客户端截图预览，如下所示：

7.总结

　　Hadoop V2 RPC框架对Socket通信进行了封装，定义了自己的基类接口VersionProtocol。该框架需要通过网络以序列化的方式传输对象，关于Hadoop V2的序列化可以参考《Hadoop2源码分析－序列化篇》，传统序列化对象较大。框架内部实现了基于Hadoop自己的服务端对象和客户端对象。服务端对象通过new RPC.Builder().builder()的方式来获取，客户端对象通过RPC.getProxy()的方式来获取。并且都需要接受Configuration对象，该对象实现了Hadoop相关文件的配置。

8.结束语

　　这篇博客就和大家分享到这里，如果大家在研究学习的过程当中有什么问题，可以加群进行讨论或发送邮件给我，我会尽我所能为您解答，与君共勉！