最近嘗試了一下etcd來做服務的註冊發現
【etcd服務】
從etcd官網下載二進位檔案即可,分配了三台機器做叢集
10.0.1.98 etcd-001
10.0.1.205 etcd-002
10.0.1.182 etcd-003
然後用指令碼啟動服務
etcd --name etcd-002 --initial-advertise-peer-urls http://10.0.1.205:2380 --listen-peer-urls http://10.0.1.205:2380 --listen-client-urls http://10.0.1.205:2379,http://127.0.0.1:2379 --advertise-client-urls http://10.0.1.205:2379 --initial-cluster-token etcd-cluster --initial-cluster etcd-001=http://10.0.1.98:2380,etcd-002=http://10.0.1.205:2380,etcd-003=http://10.0.1.182:2380 --initial-cluster-state new
etcd --name etcd-001 --initial-advertise-peer-urls http://10.0.1.98:2380 --listen-peer-urls http://10.0.1.98:2380 --listen-client-urls http://10.0.1.98:2379,http://127.0.0.1:2379 --advertise-client-urls http://10.0.1.98:2379 --initial-cluster-token etcd-cluster --initial-cluster etcd-001=http://10.0.1.98:2380,etcd-002=http://10.0.1.205:2380,etcd-003=http://10.0.1.182:2380 --initial-cluster-state new
etcd --name etcd-003 --initial-advertise-peer-urls http://10.0.1.182:2380 --listen-peer-urls http://10.0.1.182:2380 --listen-client-urls http://10.0.1.182:2379,http://127.0.0.1:2379 --advertise-client-urls http://10.0.1.182:2379 --initial-cluster-token etcd-cluster --initial-cluster etcd-001=http://10.0.1.98:2380,etcd-002=http://10.0.1.205:2380,etcd-003=http://10.0.1.182:2380 --initial-cluster-state new
即可。
【服務發布】
etcd和zk不一樣,他自身沒有臨時節點,需要用戶端自己來實現。實現的大概邏輯是這樣的:
設定一個一段時間逾時的節點,比如60秒逾時,如果逾時了etcd上就找不到這個節點,
然後用戶端用一個更小的時間間隔重新整理這個節點的逾時時間,比如每隔40秒重新整理一次,重新把ttl設定成60秒。這樣就可以保證在etcd上只要服務存活節點就一定存在,當服務關掉的時候,節點過一陣就消失了。
當然,如果能探測到服務關閉發一個del給etcd主動刪除節點就更完美了。
在maven中添加依賴
<dependency><groupId>org.mousio</groupId><artifactId>etcd4j</artifactId><version>2.13.0</version></dependency>
以spring boot上為例
package com.seeplant.etcd;import java.io.IOException;import java.net.URI;import java.util.concurrent.TimeoutException;import org.springframework.context.annotation.Scope;import org.springframework.stereotype.Component;import com.seeplant.util.Property;import com.seeplant.util.ServerLogger;import mousio.client.retry.RetryOnce;import mousio.etcd4j.EtcdClient;import mousio.etcd4j.promises.EtcdResponsePromise;import mousio.etcd4j.responses.EtcdAuthenticationException;import mousio.etcd4j.responses.EtcdException;import mousio.etcd4j.responses.EtcdKeysResponse;@Component@Scope("singleton")public class EtcdUtil { private EtcdClient client; private final String serverName = Property.getProperty("serverName"); // 自訂的服務名字,我定義成roomServer private final String dirString = "/roomServerList"; private final String zoneId = Property.getProperty("zeonId"); // 自訂的一個標識,我定義成1 private final String etcdKey = dirString + "/" + zoneId + "/" + serverName; // 這裡就是發布的節點 public EtcdUtil() { int nodeCount = Integer.parseInt(Property.getProperty("etcdGroupNodeCount")); URI[] uris = new URI[nodeCount]; // 對於叢集,把所有叢集節點地址加進來,etcd的代碼裡會輪詢這些地址來發布節點,直到成功 for (int iter = 0; iter < nodeCount; iter++) { String urlString = Property.getProperty("etcdHost" + new Integer(iter).toString()); System.out.println(urlString); uris[iter] = URI.create(urlString); } client = new EtcdClient(uris); client.setRetryHandler(new RetryOnce(20)); //retry策略 } public void regist() { // 註冊節點,放在程式啟動的入口 try { // 用put方法發布一個節點 EtcdResponsePromise<EtcdKeysResponse> p = client .putDir(etcdKey + "_" + Property.getProperty("serverIp") + "_" + Property.getProperty("serverPort")) .ttl(60).send(); p.get(); // 加上這個get()用來保證設定完成,走下一步,get會阻塞,由上面client的retry策略決定阻塞的方式 new Thread(new GuardEtcd()).start(); // 啟動一個守護線程來定時重新整理節點 } catch (Exception e) { // TODO: handle exception ServerLogger.log("etcd Server not available."); } } public void destory() { try { EtcdResponsePromise<EtcdKeysResponse> p = client .deleteDir( etcdKey + "_" + Property.getProperty("serverIp") + "_" + Property.getProperty("serverPort")) .recursive().send(); p.get(); client.close(); } catch (IOException | EtcdException | EtcdAuthenticationException | TimeoutException e) { // TODO Auto-generated catch block e.printStackTrace(); } } private class GuardEtcd implements Runnable { @Override public void run() { // TODO Auto-generated method stub while (true) { try { Thread.sleep(40*1000l); client.refresh( etcdKey + "_" + Property.getProperty("serverIp") + "_" + Property.getProperty("serverPort"), 60).send(); } catch (IOException | InterruptedException e) { // TODO Auto-generated catch block e.printStackTrace(); } } } }}
對於spring boot架構來說,可以實現CommandLineRunner來完成載入和銷毀的主動調用
package com.seeplant;import javax.annotation.PreDestroy;import org.springframework.beans.factory.annotation.Autowired;import org.springframework.boot.CommandLineRunner;import org.springframework.boot.SpringApplication;import org.springframework.boot.autoconfigure.SpringBootApplication;import org.springframework.context.annotation.Bean;import com.seeplant.etcd.EtcdUtil;import com.seeplant.netty.RoomServer;import com.seeplant.util.Property;import io.netty.channel.ChannelFuture;@SpringBootApplicationpublic class App implements CommandLineRunner{@Autowiredprivate RoomServer roomServer;@Autowiredprivate EtcdUtil etcdUtil; public static void main(String[] args) { SpringApplication.run(App.class, args); } @Bean public RoomServer roomServer() { return new RoomServer(Integer.parseInt(Property.getProperty("serverPort"))); }@Overridepublic void run(String... args) throws Exception { etcdUtil.regist();ChannelFuture future = roomServer.start();Runtime.getRuntime().addShutdownHook(new Thread(){@Overridepublic void run() { roomServer.destroy();}});future.channel().closeFuture().syncUninterruptibly();}@PreDestroypublic void destory() { etcdUtil.destory();}}
以上的retry策略是嘗試一次就放棄,另外還寫了一個retry策略是等待時間軸性增長策略,僅供參考
package com.seeplant.etcd;import com.seeplant.util.ServerLogger;import mousio.client.ConnectionState;import mousio.client.retry.RetryPolicy;/** * 為Etcd提供倍增的重試策略 * @author yuantao * */public class RetryDoublePolicy extends RetryPolicy { private final int timeSlot; // 步長 private int stepTimes = 1; // 倍數 public RetryDoublePolicy(int startRetryTime) { super(startRetryTime); // TODO Auto-generated constructor stub this.timeSlot = startRetryTime; } @Override public boolean shouldRetry(ConnectionState connectionState) { // TODO Auto-generated method stub stepTimes *= 2; connectionState.msBeforeRetry = stepTimes*timeSlot; // 重設時間 System.out.println("try " + stepTimes); if (stepTimes > 128) { ServerLogger.log("etcd connection failed"); stepTimes = 1; } return true; } }
以上是註冊服務,對於服務管理、負載平衡方,只需要按照http://127.0.0.1:2379/v2/keys/roomServerList/1/ 這種方式get就可以得到節點內容。比如我的服務發布的節點通過get返回的內容如下:
{
"action": "get",
"node": {
"key": "/roomServerList/1",
"dir": true,
"nodes": [
{
"key": "/roomServerList/1/roomServer0_127.0.0.1_9090",
"dir": true,
"expiration": "2017-09-04T05:10:21.214199005Z",
"ttl": 42,
"modifiedIndex": 202,
"createdIndex": 202
}
],
"modifiedIndex": 36,
"createdIndex": 36
}
}