分布式架构基础:Java RMI详解

RMI简介

Java RMI，即 远程方法调用(Remote Method Invocation)，一种用于实现远程过程调用(RPC)(Remote procedure call)的Java API，能直接传输序列化后的Java对象和分布式垃圾收集。它的实现依赖于Java虚拟机(JVM)，因此它仅支持从一个JVM到另一个JVM的调用。

rmi架构图

rmi的实现

(1) 直接使用Registry实现rmi

服务端:

public class RegistryService {
    public static void main(String[] args) {
        try {
            // 本地主机上的远程对象注册表Registry的实例,默认端口1099
            Registry registry = LocateRegistry.createRegistry(1099);
            // 创建一个远程对象
            HelloRegistryFacade hello = new HelloRegistryFacadeImpl();
            // 把远程对象注册到RMI注册服务器上，并命名为HelloRegistry
            registry.rebind("HelloRegistry", hello);
            System.out.println("======= 启动RMI服务成功! =======");
        } catch (RemoteException e) {
            e.printStackTrace();
        }
    }
}

接口:

继承Remote接口

public interface HelloRegistryFacade extends Remote {

    String helloWorld(String name) throws RemoteException;

}

接口实现:

继承UnicastRemoteObject

public class HelloRegistryFacadeImpl extends UnicastRemoteObject implements HelloRegistryFacade{

    public HelloRegistryFacadeImpl() throws RemoteException {
        super();
    }

    @Override
    public String helloWorld(String name) {
        return "[Registry] 你好! " + name;
    }

}

客户端:

public class RegistryClient {
    public static void main(String[] args) {
        try {
            Registry registry = LocateRegistry.getRegistry(1099);
            HelloRegistryFacade hello = (HelloRegistryFacade) registry.lookup("HelloRegistry");
            String response = hello.helloWorld("ZhenJin");
            System.out.println("=======> " + response + " <=======");
        } catch (NotBoundException | RemoteException e) {
            e.printStackTrace();
        }
    }
}

图解:

出处:https://www.tutorialspoint.com/java_rmi/java_rmi_introduction.htm

rmi调用过程

Registry(注册表)是放置所有服务器对象的命名空间。
每次服务端创建一个对象时，它都会使用bind()或rebind()方法注册该对象。
这些是使用称为绑定名称的唯一名称注册的。

要调用远程对象，客户端需要该对象的引用,如(HelloRegistryFacade)。
即通过服务端绑定的名称(HelloRegistry)从注册表中获取对象(lookup()方法)。

(2) 使用Naming方法实现rmi

服务端:

public class NamingService {
    public static void main(String[] args) {
        try {
            // 本地主机上的远程对象注册表Registry的实例
            LocateRegistry.createRegistry(1100);
            // 创建一个远程对象
            HelloNamingFacade hello = new HelloNamingFacadeImpl();
            // 把远程对象注册到RMI注册服务器上，并命名为Hello 
            //绑定的URL标准格式为：rmi://host:port/name
            Naming.bind("rmi://localhost:1100/HelloNaming", hello);
            System.out.println("======= 启动RMI服务成功! =======");
        } catch (RemoteException | MalformedURLException | AlreadyBoundException e) {
            e.printStackTrace();
        }
    }
}

接口和接口实现和Registry的方式一样

客户端:

public class NamingClient {
    public static void main(String[] args) {
        try {
            String remoteAddr="rmi://localhost:1100/HelloNaming";
            HelloNamingFacade hello = (HelloNamingFacade) Naming.lookup(remoteAddr);
            String response = hello.helloWorld("ZhenJin");
            System.out.println("=======> " + response + " <=======");
        } catch (NotBoundException | RemoteException | MalformedURLException e) {
            e.printStackTrace();
        }
    }
}

Naming部分源码:

public static Remote lookup(String name)
    throws NotBoundException,java.net.MalformedURLException,RemoteException{
    ParsedNamingURL parsed = parseURL(name);
    Registry registry = getRegistry(parsed);

    if (parsed.name == null)
        return registry;
    return registry.lookup(parsed.name);
}

Naming其实是对Registry的一个封装

Scala实现rmi

上面说了rmi是通过JVM虚拟机进行一个远程调用的,我们通过Scala,kotlin等jvm语言印证下

服务端:

object ScalaRmiService extends App {
  try {
    val user:UserScalaFacade = new UserScalaFacadeImpl
    LocateRegistry.createRegistry(1103)
    Naming.rebind("rmi://localhost:1103/UserScala", user)
    println("======= 启动RMI服务成功! =======")
  } catch {
    case e: IOException => println(e)
  }
}

接口

trait UserScalaFacade extends Remote {

  /**
    * 通过用户名获取用户信息
    */
  @throws(classOf[RemoteException])
  def getByName(userName: String): User

  /**
    * 通过用户性别获取用户信息
    */
  @throws(classOf[RemoteException])
  def getBySex(userSex: String): List[User]

}

接口实现:

class UserScalaFacadeImpl extends UnicastRemoteObject with UserScalaFacade {

  /**
    * 模拟一个数据库表
    */
  private lazy val userList = List(
    new User("Jane", "女", 16),
    new User("jack", "男", 17),
    new User("ZhenJin", "男", 18)
  )

  override def getByName(userName: String): User = userList.filter(u => userName.equals(u.userName)).head

  override def getBySex(userSex: String): List[User] = userList.filter(u => userSex.equals(u.userSex))

}

实体类:

实体类必须实现序列化(Serializable)才能进行一个远程传输

class User(name: String, sex: String, age: Int) extends Serializable {

  var userName: String = name
  var userSex: String = sex
  var userAge: Int = age
  override def toString = s"User(userName=$userName, userSex=$userSex, userAge=$userAge)"

}

Scala客户端:

object ScalaRmiClient extends App {

  try {

    val remoteAddr="rmi://localhost:1103/UserScala"
    val userFacade = Naming.lookup(remoteAddr).asInstanceOf[UserScalaFacade]

    println(userFacade.getByName("ZhenJin"))
    System.out.println("--------------------------------------")
    for (user <- userFacade.getBySex("男")) println(user)

  } catch {
    case e: NotBoundException => println(e)
    case e: RemoteException => println(e)
    case e: MalformedURLException => println(e)
  }

}

Java客户端:

public class JavaRmiClient {

    public static void main(String[] args) {

        try {
            String remoteAddr="rmi://localhost:1103/UserScala";
            UserScalaFacade userFacade = (UserScalaFacade) Naming.lookup();

            User zhenJin = userFacade.getByName("ZhenJin");
            System.out.println(zhenJin);
            System.out.println("--------------------------------------");
            List<User> userList = userFacade.getBySex("男");
            System.out.println(userList);

        } catch (NotBoundException | RemoteException | MalformedURLException e) {
            e.printStackTrace();
        }
    }
}

上面试验可以证明Scala和Java是可以互通的,Scala本身也是可以直接引用Java类的

序列化简介

序列化(Serialization)是将数据结构或对象状态转换为可以存储（例如，在文件或存储器缓冲区中）或传输（例如，通过网络连接）的格式的过程, 反序列化(Deserialization)则是从一系列字节中提取数据结构的相反操作.

序列化与反序列化

Kotlin实现rmi

服务端:

fun main(args: Array<String>) {
    try {
        val hello: HelloKotlinFacade = HelloKotlinFacadeImpl()
        LocateRegistry.createRegistry(1102)
        Naming.rebind("rmi://localhost:1101/HelloKotlin", hello)
        println("======= 启动RMI服务成功! =======")
    } catch (e: IOException) {
        e.printStackTrace()
    }
}

客户端:

fun main(args: Array<String>) {
    try {
        val hello = Naming.lookup("rmi://localhost:1102/HelloKotlin") as HelloKotlinFacade
        val response = hello.helloWorld("ZhenJin")
        println("=======> $response <=======")
    } catch (e: NotBoundException) {
        e.printStackTrace()
    } catch (e: RemoteException) {
        e.printStackTrace()
    } catch (e: MalformedURLException) {
        e.printStackTrace()
    }
}

实现和接口省略...

SpringBoot实现rmi

StringBoot通过配置就可以简单实现rmi了

服务端:

@Configuration
public class RmiServiceConfig {
    @Bean
    public RmiServiceExporter registerService(UserFacade userFacade) {
        RmiServiceExporter rmiServiceExporter = new RmiServiceExporter();
        rmiServiceExporter.setServiceName("UserInfo");
        rmiServiceExporter.setService(userFacade);
        rmiServiceExporter.setServiceInterface(UserFacade.class);
        rmiServiceExporter.setRegistryPort(1101);
        return rmiServiceExporter;
    }
}

客户端:

@Configuration
public class RmiClientConfig {

    @Bean
    public UserFacade userInfo() {
        RmiProxyFactoryBean rmiProxyFactoryBean = new RmiProxyFactoryBean();
        rmiProxyFactoryBean.setServiceUrl("rmi://localhost:1101/UserInfo");
        rmiProxyFactoryBean.setServiceInterface(UserFacade.class);
        rmiProxyFactoryBean.afterPropertiesSet();
        return (UserFacade) rmiProxyFactoryBean.getObject();
    }

}

客户端测试类:

@Autowired
private UserFacade userFacade;
    
@Test
public void userBySexTest() {
    try {
        List<User> userList = userFacade.getBySex("男");
        userList.forEach(System.out::println);
    } catch (RemoteException e) {
        e.printStackTrace();
    }
}

通过测试类可以看出,这和我们平时的程序调用内部方法没什么区别!

rmi调用过程

大家可以通过下面文章加深了解:

https://stuff.mit.edu/afs/athena/software/java/java_v1.2.2/distrib/sun4x_56/docs/guide/rmi/Factory.html

rmi工厂调用的过程

有两个远程服务接口可供client调用，Factory和Product接口

FactoryImpl类实现了Factory接口，ProductImpl类实现了Product接口

1. FactoryImpl被注册到了rmi-registry中
2. client端请求一个Factory的引用 
3. rmi-registry返回client端一个FactoryImpl的引用 
4. client端调用FactoryImpl的远程方法请求一个ProductImpl的远程引用
5. FactoryImpl返回给client端一个ProductImpl引用 
6. client通过ProductImpl引用调用远程方法

socket工厂文档:
https://docs.oracle.com/javase/8/docs/technotes/guides/rmi/socketfactory/index.html

Zookeeper实现rmi

出处:http://www.importnew.com/20344.html

安装Zookeeper

解压 ZooKeeper

tar -zxvf zookeeper-3.4.12.tar.gz

在 conf 目录新建 zoo.cfg

cd zookeeper-3.4.12/conf
vim zoo.cfg

zoo.cfg 代码如下（自己指定 log 文件目录）：

tickTime=2000
dataDir=/usr/local/zookeeper-3.4.12/data 
dataLogDir=/usr/local/zookeeper-3.4.12/log
clientPort=2181

在 bin 目录下，启动 Zookeeper：

cd zookeeper-3.4.12/bin
./zkServer.sh start

消费者:

public class RmiConsumer {

    // 用于等待 SyncConnected 事件触发后继续执行当前线程
    private CountDownLatch latch = new CountDownLatch(1);

    // 定义一个 volatile 成员变量，用于保存最新的 RMI 地址（考虑到该变量或许会被其它线程所修改，一旦修改后，该变量的值会影响到所有线程）
    private volatile List<String> urlList = new ArrayList<>();

    // 构造器
    public RmiConsumer() {
        ZooKeeper zk = connectServer(); // 连接 ZooKeeper 服务器并获取 ZooKeeper 对象
        if (zk != null) {
            watchNode(zk); // 观察 /registry 节点的所有子节点并更新 urlList 成员变量
        }
    }

    // 查找 RMI 服务
    public <T extends Remote> T lookup() {
        T service = null;
        int size = urlList.size();
        if (size > 0) {
            String url;
            if (size == 1) {
                url = urlList.get(0); // 若 urlList 中只有一个元素，则直接获取该元素
                log.debug("using only url: {}", url);
            } else {
                url = urlList.get(ThreadLocalRandom.current().nextInt(size)); // 若 urlList 中存在多个元素，则随机获取一个元素
                log.debug("using random url: {}", url);
            }
            service = lookupService(url); // 从 JNDI 中查找 RMI 服务
        }
        return service;
    }

    // 连接 ZooKeeper 服务器
    private ZooKeeper connectServer() {
        ZooKeeper zk = null;
        try {
            zk = new ZooKeeper(Constant.ZK_CONNECTION_STRING, Constant.ZK_SESSION_TIMEOUT, new Watcher() {
                @Override
                public void process(WatchedEvent event) {
                    if (event.getState() == Event.KeeperState.SyncConnected) {
                        latch.countDown(); // 唤醒当前正在执行的线程
                    }
                }
            });
            latch.await(); // 使当前线程处于等待状态
        } catch (IOException | InterruptedException e) {
            log.error("", e);
        }
        return zk;
    }

    // 观察 /registry 节点下所有子节点是否有变化
    private void watchNode(final ZooKeeper zk) {
        try {
            List<String> nodeList = zk.getChildren(Constant.ZK_REGISTRY_PATH, event -> {
                if (event.getType() == Watcher.Event.EventType.NodeChildrenChanged) {
                    watchNode(zk); // 若子节点有变化，则重新调用该方法（为了获取最新子节点中的数据）
                }
            });
            List<String> dataList = new ArrayList<>(); // 用于存放 /registry 所有子节点中的数据
            for (String node : nodeList) {
                byte[] data = zk.getData(Constant.ZK_REGISTRY_PATH + "/" + node, false, null); // 获取 /registry 的子节点中的数据
                dataList.add(new String(data));
            }
            log.debug("node data: {}", dataList);
            urlList = dataList; // 更新最新的 RMI 地址
        } catch (KeeperException | InterruptedException e) {
            log.error("", e);
        }
    }

    // 在 JNDI 中查找 RMI 远程服务对象
    @SuppressWarnings("unchecked")
    private <T> T lookupService(String url) {
        T remote = null;
        try {
            remote = (T) Naming.lookup(url);
        } catch (NotBoundException | MalformedURLException | RemoteException e) {
            log.error("远程查找出错!", e);
        }
        return remote;
    }
}

生产者:

public class RmiProvider {


    /**
     * 用于等待 SyncConnected 事件触发后继续执行当前线程
     */
    private CountDownLatch latch = new CountDownLatch(1);

    // 发布 RMI 服务并注册 RMI 地址到 ZooKeeper 中
    public void publish(Remote remote, String host, int port) {
        String url = publishService(remote, host, port); // 发布 RMI 服务并返回 RMI 地址
        if (url != null) {
            ZooKeeper zk = connectServer(); // 连接 ZooKeeper 服务器并获取 ZooKeeper 对象
            if (zk != null) {
                createNode(zk, url); // 创建 ZNode 并将 RMI 地址放入 ZNode 上
            }
        }
    }

     /**
      *发布 RMI 服务
      */
    private String publishService(Remote remote, String host, int port) {
        String url = null;
        try {
            url = String.format("rmi://%s:%d/%s", host, port, remote.getClass().getName());
            LocateRegistry.createRegistry(port);
            Naming.rebind(url, remote);
            log.debug("publish rmi service (url: {})", url);
        } catch (RemoteException | MalformedURLException e) {
            log.error("", e);
        }
        return url;
    }

    // 连接 ZooKeeper 服务器
    private ZooKeeper connectServer() {
        ZooKeeper zk = null;
        try {
            zk = new ZooKeeper(Constant.ZK_CONNECTION_STRING, Constant.ZK_SESSION_TIMEOUT, new Watcher() {
                @Override
                public void process(WatchedEvent event) {
                    if (event.getState() == Event.KeeperState.SyncConnected) {
                        latch.countDown(); // 唤醒当前正在执行的线程
                    }
                }
            });
            latch.await(); // 使当前线程处于等待状态
        } catch (IOException | InterruptedException e) {
            log.error("", e);
        }
        return zk;
    }

    /**
     * 创建节点
     */
    private void createNode(ZooKeeper zk, String url) {
        try {
            byte[] data = url.getBytes();
            String path = zk.create(Constant.ZK_PROVIDER_PATH, data, ZooDefs.Ids.OPEN_ACL_UNSAFE, CreateMode.EPHEMERAL_SEQUENTIAL);    // 创建一个临时性且有序的 ZNode
            log.debug("create zookeeper node ({} => {})", path, url);
        } catch (KeeperException | InterruptedException e) {
            log.error("", e);
        }
    }
}

图解:

zookeeper

代码已上传到GitHub上:https://github.com/jayknoxqu/rmi-example