Rpc-實現Zookeeper註冊中心

2023-02-17 18:02:01

1.前言

本文章是筆主在聲哥的手寫RPC框架的學習下,對註冊中心的一個拓展。因為聲哥某些部分沒有保留拓展性,所以本文章的專案與聲哥的工程有部分割區別,核心內容在Curator的註冊發現與登出,思想看準即可。

本文章Git倉庫:zko0/zko0-rpc

聲哥的RPC專案寫的確實很詳細,跟學一遍受益匪淺:

何人聽我楚狂聲的部落格

在聲哥的專案裡使用Nacos作為了服務註冊中心。本人拓展新增了ZooKeeper實現服務註冊。

Nacos的服務註冊和發現,設計的不是非常好,每次服務的發現都需要去註冊中心拉取。本人實現ZooKeeper註冊中心時,參考了Dubbo的設計原理,結合本人自身想法,新增了本地快取:

  • Client發現服務後快取在本地,維護一個服務——範例列表
  • 當監聽到註冊中心的服務列表發生了變化,Client更新本地列表
  • 當註冊中心宕機,Client能夠依靠原生的服務列表繼續提供服務

問題:

  1. 實現服務註冊的本地快取,還需要實現註冊中心的監聽,當註冊中心的服務發生更改時能夠實現動態更新。或者用輪訓的方式,定時更新,不過這種方式的服務實時性較差
  2. 當Server宕機,非臨時節點註冊容易出現服務殘留無法清除的問題。所以我建議全部使用臨時節點去註冊。

2.內容

zookeeper需要簡單學一下,知識內容非常簡單,搭建也很簡單,在此跳過。

如果你感興趣,可以參考我的ZooKeeper的文章:Zookeeper學習筆記 - zko0

①新增依賴

Curator:(簡化ZooKeeper使用者端使用)(Netfix研發,捐給Apache,是Apache頂級專案)

這裡排除slf4j依賴,因為筆主使用的slf4j存在衝突

<!-- https://mvnrepository.com/artifact/org.apache.curator/curator-recipes -->
<dependency>
    <groupId>org.apache.curator</groupId>
    <artifactId>curator-recipes</artifactId>
    <version>5.2.0</version>
    <exclusions>
        <exclusion>
            <groupId>org.slf4j</groupId>
            <artifactId>slf4j-api</artifactId>
        </exclusion>
    </exclusions>
</dependency>

②程式碼編寫

1.首先建立一個連線類:

@Slf4j
public class ZookeeperUtil {

    //內部化構造方法
    private ZookeeperUtil(){
    }

    private static final String SERVER_HOSTNAME= RegisterCenterConfig.getHostName();

    private static final Integer SERVER_PORT=RegisterCenterConfig.getServerPort();

    private static CuratorFramework zookeeperClient;

    public static CuratorFramework getZookeeperClient(){
        if (zookeeperClient==null){
            synchronized (ZookeeperUtil.class){
                if (zookeeperClient==null){
                    RetryPolicy retryPolic=new ExponentialBackoffRetry(3000,10);
                    zookeeperClient = CuratorFrameworkFactory.builder()
                            .connectString(SERVER_HOSTNAME+":"+SERVER_PORT)
                            .retryPolicy(retryPolic)
                            //  zookeeper根目錄為/serviceRegister,不為/
                            .namespace("serviceRegister")
                            .build();
                    zookeeperClient.start();
                }
            }
        }
        return zookeeperClient;
    }

    public static String getServerHostname(){
        return SERVER_HOSTNAME;
    }

    public static Integer getServerPort(){
        return SERVER_PORT;
    }

}

其中HOST,PORT資訊我儲存在regiserCenter.properties組態檔夾中,使用類讀取:

public class RpcConfig {


    //註冊中心型別
    private static String registerCenterType;

    //序列化型別
    private static String serializerType;

    //負載均衡型別
    private static String loadBalanceType;

    //設定Nacos地址
    private static String registerCenterHost;

    private static Integer registerCenterPort;


    private static boolean zookeeperDestoryIsEphemeral;

    private static String serverHostName;

    private static Integer serverPort;

    static {
        ResourceBundle bundle = ResourceBundle.getBundle("rpc");
        registerCenterType=bundle.getString("registerCenter.type");
        loadBalanceType=bundle.getString("loadBalance.type");
        registerCenterHost=bundle.getString("registerCenter.host");
        registerCenterPort = Integer.parseInt(bundle.getString("registerCenter.port"));
        try {
            zookeeperDestoryIsEphemeral="true".equals(bundle.getString("registerCenter.destory.isEphemeral"));
        } catch (Exception e) {
            zookeeperDestoryIsEphemeral=false;
        }
        serializerType=bundle.getString("serializer.type");
        serverHostName=bundle.getString("server.hostName");
        serverPort=Integer.parseInt(bundle.getString("server.port"));
    }

    public static String getRegisterCenterType() {
        return registerCenterType;
    }

    public static String getSerializerType() {
        return serializerType;
    }

    public static String getLoadBalanceType() {
        return loadBalanceType;
    }

    public static String getRegisterCenterHost() {
        return registerCenterHost;
    }

    public static Integer getRegisterCenterPort() {
        return registerCenterPort;
    }


    public static String getServerHostName() {
        return serverHostName;
    }

    public static Integer getServerPort() {
        return serverPort;
    }

    public static boolean isZookeeperDestoryIsEphemeral() {
        return zookeeperDestoryIsEphemeral;
    }

}

下面的程式碼我和聲哥有些不同,我將服務註冊,登出方法放在ServerUtils中,服務發現方法放在ClientUtils中:

服務的高一致性存在兩種做法:

  • 因為ZooKeeper存在臨時節點,註冊中心可以實現Client(RPC的Server)斷開,註冊服務資訊的自動丟失
  • 不設定為臨時節點,手動的服務註冊清除

我這裡兩種都實現了,雖然做兩種方式不同但是功能相同的程式碼放在一起看起來很奇怪,這裡只是做演示。選擇其中一種即可。(我建議使用臨時節點,當Server宕機,殘留的服務資訊也能及時清除)

註冊實現原理圖:

介面:

public interface ServiceDiscovery {
    InetSocketAddress searchService(String serviceName);

    void cleanLoaclCache(String serviceName);
}
public interface ServiceRegistry {
    //服務註冊
    void register(String serviceName, InetSocketAddress inetAddress);

    void cleanRegistry();
}

ZooKeeper介面實現:

public class ZookeeperServiceDiscovery implements ServiceDiscovery{

    private final LoadBalancer loadBalancer;

    public ZookeeperServiceDiscovery(LoadBalancer loadBalancer) {
        this.loadBalancer = loadBalancer;
    }
    
    @Override
    public InetSocketAddress searchService(String serviceName) {
        return ZookeeperClientUtils.searchService(serviceName,loadBalancer);
    }

    @Override
    public void cleanLoaclCache(String serviceName) {
        ZookeeperClientUtils.cleanLocalCache(serviceName);
    }
}
public class ZookeeperServiceRegistry implements ServiceRegistry{
    @Override
    public void register(String serviceName, InetSocketAddress inetAddress) {
        ZookeeperServerUitls.register(serviceName,inetAddress);
    }

    @Override
    public void cleanRegistry() {
        ZookeeperServerUitls.cleanRegistry();
    }
}

Factory工廠:

public class ServiceFactory {

    private static String center = RpcConfig.getRegisterCenterType();
    private static String lb= RpcConfig.getLoadBalanceType();

    private static  ServiceRegistry registry;

    private static  ServiceDiscovery discovery;

    private static Object registerLock=new Object();

    private static Object discoveryLock=new Object();

    public static ServiceDiscovery getServiceDiscovery(){
        if (discovery==null){
            synchronized (discoveryLock){
                if (discovery==null){
                    if ("nacos".equalsIgnoreCase(center)){
                        discovery= new NacosServiceDiscovery(LoadBalancerFactory.getLoadBalancer(lb));
                    }else if ("zookeeper".equalsIgnoreCase(center)){
                        discovery= new ZookeeperServiceDiscovery(LoadBalancerFactory.getLoadBalancer(lb));
                    }
                }
            }
        }
        return discovery;
    }

    public static ServiceRegistry getServiceRegistry(){
        if (registry==null){
            synchronized (registerLock){
                if (registry==null){
                    if ("nacos".equalsIgnoreCase(center)){
                        registry=  new NacosServiceRegistry();
                    }else if ("zookeeper".equalsIgnoreCase(center)){
                        registry= new ZookeeperServiceRegistry();
                    }
                }
            }
        }
        return registry;
    }

}

使用Gson序列化InetSocketAddress存在問題,編寫Util:

public class InetSocketAddressSerializerUtil {
    public static String getJsonByInetSockerAddress(InetSocketAddress address){
        HashMap<String, String> map = new HashMap<>();
        map.put("host",address.getHostName());
        map.put("port",address.getPort()+"");
        return new Gson().toJson(map);
    }

    public static InetSocketAddress getInetSocketAddressByJson(String json){
        HashMap<String,String> hashMap = new Gson().fromJson(json, HashMap.class);
        String host = hashMap.get("host");
        Integer port=Integer.parseInt(hashMap.get("port"));
        return new InetSocketAddress(host,port);
    }

}

上面主要是註冊,發現的邏輯,我把主要方法寫在了Utils中:

@Slf4j
public class ZookeeperServerUitls {

    private static CuratorFramework client = ZookeeperUtil.getZookeeperClient();

    private static final Set<String> instances=new ConcurrentHashSet<>();

    public static void register(String serviceName, InetSocketAddress inetSocketAddress){

        serviceName=ZookeeperUtil.serviceName2Path(serviceName);;
        String uuid = UUID.randomUUID().toString();
        serviceName=serviceName+"/"+uuid;
        String json = InetSocketAddressSerializerUtil.getJsonByInetSockerAddress(inetSocketAddress);
        try {
            if (RpcConfig.isZookeeperDestoryIsEphemeral()){
                //對談結束節點,建立消失
                client.create()
                        .creatingParentsIfNeeded()
                        .withMode(CreateMode.EPHEMERAL)
                        .forPath(serviceName,json.getBytes());
            } else {
                client.create()
                        .creatingParentsIfNeeded()
                        .forPath(serviceName,json.getBytes());
            }
        }
            catch (Exception e) {
            log.error("服務註冊失敗");
            throw  new RpcException(RpcError.REGISTER_SERVICE_FAILED);
        }
        //放入map
        instances.add(serviceName);
    }

    public static void cleanRegistry(){
        log.info("登出所有註冊的服務");
        //如果自動銷燬,不需要清除
        if (RpcConfig.isZookeeperDestoryIsEphemeral()) return;
        if (ZookeeperUtil.getServerHostname()!=null&&ZookeeperUtil.getServerPort()!=null&&!instances.isEmpty()){
            for (String path:instances) {
                try {
                    client.delete().forPath(path);
                } catch (Exception e) {
                    log.error("服務登出失敗");
                    throw new RpcException(RpcError.DESTORY_REGISTER_FALL);
                }
            }
        }
    }
}
@Slf4j
public class ZookeeperClientUtils {

    private static CuratorFramework client = ZookeeperUtil.getZookeeperClient();

    private static final Map<String,  List<InetSocketAddress>> instances=new ConcurrentHashMap<>();

    public static InetSocketAddress searchService(String serviceName, LoadBalancer loadBalancer) {
        InetSocketAddress address;
        //本地快取查詢
        if (instances.containsKey(serviceName)){
            List<InetSocketAddress> addressList = instances.get(serviceName);
            if (!addressList.isEmpty()){
                //使用lb進行負載均衡
                return loadBalancer.select(addressList);
            }
        }
        try {
            String path = ZookeeperUtil.serviceName2Path(serviceName);
            //獲取路徑下所有的實現
            List<String> instancePaths = client.getChildren().forPath(path);
            List<InetSocketAddress> addressList = new ArrayList<>();
            for (String instancePath : instancePaths) {
                byte[] bytes = client.getData().forPath(path+"/"+instancePath);
                String json = new String(bytes);
                InetSocketAddress instance = InetSocketAddressSerializerUtil.getInetSocketAddressByJson(json);
                addressList.add(instance);
            }
            addLocalCache(serviceName,addressList);
            return loadBalancer.select(addressList);
        } catch (Exception e) {
            log.error("服務獲取失敗====>{}",e);
            throw new RpcException(RpcError.SERVICE_NONE_INSTANCE);
        }
    }

    public static void cleanLocalCache(String serviceName){
        log.info("服務呼叫失敗,清除本地快取,重新獲取範例===>{}",serviceName);
        instances.remove(serviceName);
    }

    public static void addLocalCache(String serviceName,List<InetSocketAddress> addressList){
        //直接替換原本的快取
        instances.put(serviceName,addressList);
    }
}

③組態檔

rpc.properties放在resources下

#nacos    zookeeper
#registerCenter.type=nacos
registerCenter.type=zookeeper

#registerCenter.host=127.0.0.1
registerCenter.host=101.43.244.40

#zookeeper port default 2181
#registerCenter.port=9000
registerCenter.port=2181

registerCenter.destory.isEphemeral=false

#??random?roundRobin
loadBalance.type=random

#kryo json jdk
serializer.type=kryo

server.hostName=127.0.0.1
server.port=9999

④更多

聲哥的程式碼我做了很多修改,如果上述程式碼和你參考的專案程式碼出入比較大,可以檢視本文章的工程閱讀。