阿里二面：使用 Nacos 做注冊中心怎么做優雅發布？

作者：朱晉君 2023-06-13 08:25:14

開發前端

無論是哪一款注冊中心，優雅發布要解決的問題都是優雅上線和優雅下線。本文結合 Nacos 的原理講解了 Nacos 的優雅發布，希望對你有所幫助。

大家好，我是君哥。

今天來聊一聊使用 Nacos 做注冊中心怎么做優雅發布。

跟其他的注冊中心一樣，Nacos 作為注冊中心的使用如下圖：

Service Provider 啟動后注冊到 Nacos Server，Service Consumer 則從 Nacos Server 拉取服務列表，根據一定算法選擇一個 Service Provider 來發送請求。

1.優雅要求

對于優雅發布，要求是 Service Provider 上線（注冊到 Nacos）后，服務能夠正常地接收和處理請求，而 Service Provider 停服后，則不會再收到請求。這就有兩個要求：

優雅上線：Service Provider 發布完成之前，Service Consumer 不應該從服務列表中拉取到這個服務地址；
優雅下線：Service Provider 下線后，Service Consumer 不會從服務列表中拉取到這個服務地址。

解決了這兩個問題，優雅發布就可以做到了。

2.搭建環境

搭建環境是為了看 Nacos 日志，通過日志找到對應的源代碼。本文搭建的環境如下圖：

2.1 啟動 provider

啟動 springboot-provider 的應用，注冊到 Nacos，啟動日志如下：

2023-06-11 18:58:10,120 [main] [INFO] com.alibaba.nacos.client.naming - [BEAT] adding beat: BeatInfo{port=8083, ip='192.168.31.94', weight=1.0, serviceName='DEFAULT_GROUP@@springboot-provider', cluster='DEFAULT', metadata={management.endpoints.web.base-path=/actuator, management.port=18082, preserved.register.source=SPRING_CLOUD, management.address=127.0.0.1}, scheduled=false, period=5000, stopped=false} to beat map.
2023-06-11 18:58:10,121 [main] [INFO] com.alibaba.nacos.client.naming - [REGISTER-SERVICE] public registering service DEFAULT_GROUP@@springboot-provider with instance: Instance{instanceId='null', ip='192.168.31.94', port=8083, weight=1.0, healthy=true, enabled=true, ephemeral=true, clusterName='DEFAULT', serviceName='null', metadata={management.endpoints.web.base-path=/actuator, management.port=18082, preserved.register.source=SPRING_CLOUD, management.address=127.0.0.1}}
2023-06-11 18:58:10,133 [main] [INFO] com.alibaba.cloud.nacos.registry.NacosServiceRegistry - nacos registry, DEFAULT_GROUP springboot-provider 192.168.31.94:8083 register finished
2023-06-11 18:58:10,221 [main] [INFO] org.springframework.boot.web.embedded.tomcat.TomcatWebServer - Tomcat initialized with port(s): 18082 (http)
2023-06-11 18:58:10,222 [main] [INFO] org.apache.coyote.http11.Http11NioProtocol - Initializing ProtocolHandler ["http-nio-127.0.0.1-18082"]
2023-06-11 18:58:10,223 [main] [INFO] org.apache.catalina.core.StandardService - Starting service [Tomcat]
2023-06-11 18:58:10,223 [main] [INFO] org.apache.catalina.core.StandardEngine - Starting Servlet engine: [Apache Tomcat/9.0.21]
2023-06-11 18:58:10,239 [main] [INFO] org.apache.catalina.core.ContainerBase.[Tomcat-1].[localhost].[/] - Initializing Spring embedded WebApplicationContext
2023-06-11 18:58:10,239 [main] [INFO] org.springframework.web.context.ContextLoader - Root WebApplicationContext: initialization completed in 99 ms
2023-06-11 18:58:10,268 [main] [INFO] org.springframework.boot.actuate.endpoint.web.EndpointLinksResolver - Exposing 22 endpoint(s) beneath base path '/actuator'
2023-06-11 18:58:10,336 [main] [INFO] org.apache.coyote.http11.Http11NioProtocol - Starting ProtocolHandler ["http-nio-127.0.0.1-18082"]
2023-06-11 18:58:10,340 [main] [INFO] org.springframework.boot.web.embedded.tomcat.TomcatWebServer - Tomcat started on port(s): 18082 (http) with context path ''
2023-06-11 18:58:10,342 [main] [INFO] boot.Application - Started Application in 7.051 seconds (JVM running for 7.874)
2023-06-11 18:58:10,358 [main] [INFO] com.alibaba.nacos.client.config.impl.ClientWorker - [fixed-39.105.183.91_8848] [subscribe] springboot-provider.properties+DEFAULT_GROUP
2023-06-11 18:58:10,359 [main] [INFO] com.alibaba.nacos.client.config.impl.CacheData - [fixed-39.105.183.91_8848] [add-listener] ok, tenant=, dataId=springboot-provider.properties, group=DEFAULT_GROUP, cnt=1
2023-06-11 18:58:10,359 [main] [INFO] com.alibaba.nacos.client.config.impl.ClientWorker - [fixed-39.105.183.91_8848] [subscribe] springboot-provider-dev.properties+DEFAULT_GROUP
2023-06-11 18:58:10,359 [main] [INFO] com.alibaba.nacos.client.config.impl.CacheData - [fixed-39.105.183.91_8848] [add-listener] ok, tenant=, dataId=springboot-provider-dev.properties, group=DEFAULT_GROUP, cnt=1
2023-06-11 18:58:10,360 [main] [INFO] com.alibaba.nacos.client.config.impl.ClientWorker - [fixed-39.105.183.91_8848] [subscribe] springboot-provider+DEFAULT_GROUP
2023-06-11 18:58:10,360 [main] [INFO] com.alibaba.nacos.client.config.impl.CacheData - [fixed-39.105.183.91_8848] [add-listener] ok, tenant=, dataId=springboot-provider, group=DEFAULT_GROUP, cnt=1
2023-06-11 18:58:10,639 [RMI TCP Connection(1)-192.168.31.94] [INFO] org.apache.catalina.core.ContainerBase.[Tomcat].[localhost].[/] - Initializing Spring DispatcherServlet 'dispatcherServlet'
2023-06-11 18:58:10,839 [com.alibaba.nacos.client.naming.updater] [INFO] com.alibaba.nacos.client.naming - [BEAT] adding beat: BeatInfo{port=8083, ip='192.168.31.94', weight=1.0, serviceName='DEFAULT_GROUP@@springboot-provider', cluster='DEFAULT', metadata={management.endpoints.web.base-path=/actuator, management.port=18082, preserved.register.source=SPRING_CLOUD, management.address=127.0.0.1}, scheduled=false, period=5000, stopped=false} to beat map.
2023-06-11 18:58:10,840 [com.alibaba.nacos.client.naming.updater] [INFO] com.alibaba.nacos.client.naming - modified ips(1) service: DEFAULT_GROUP@@springboot-provider@@DEFAULT -> [{"instanceId":"192.168.31.94#8083#DEFAULT#DEFAULT_GROUP@@springboot-provider","ip":"192.168.31.94","port":8083,"weight":1.0,"healthy":true,"enabled":true,"ephemeral":true,"clusterName":"DEFAULT","serviceName":"DEFAULT_GROUP@@springboot-provider","metadata":{"management.endpoints.web.base-path":"/actuator","management.port":"18082","preserved.register.source":"SPRING_CLOUD","management.address":"127.0.0.1"},"ipDeleteTimeout":30000,"instanceHeartBeatInterval":5000,"instanceHeartBeatTimeOut":15000}]
2023-06-11 18:58:10,841 [com.alibaba.nacos.client.naming.updater] [INFO] com.alibaba.nacos.client.naming - current ips:(1) service: DEFAULT_GROUP@@springboot-provider@@DEFAULT -> [{"instanceId":"192.168.31.94#8083#DEFAULT#DEFAULT_GROUP@@springboot-provider","ip":"192.168.31.94","port":8083,"weight":1.0,"healthy":true,"enabled":true,"ephemeral":true,"clusterName":"DEFAULT","serviceName":"DEFAULT_GROUP@@springboot-provider","metadata":{"management.endpoints.web.base-path":"/actuator","management.port":"18082","preserved.register.source":"SPRING_CLOUD","management.address":"127.0.0.1"},"ipDeleteTimeout":30000,"instanceHeartBeatInterval":5000,"instanceHeartBeatTimeOut":15000}]

我們再看下 Nacos 的日志，這里看的文件 naming-server.log，日志如下：

2023-06-11 18:58:09,723 INFO Client connection 192.168.31.94:51885#true connect
2023-06-11 18:58:10,105 INFO Client change for service Service{namespace='public', group='DEFAULT_GROUP', name='springboot-provider', ephemeral=true, revisinotallow=1}, 192.168.31.94:8083#true
2023-06-11 18:58:18,204 INFO Client connection 192.168.31.94:60850#true disconnect, remove instances and subscribers

springboot-provider 啟動成功后，從Nacos 管理后臺可以看到下圖：

2.2 provider 下線

服務下線后，Nacos 日志如下：

2023-06-11 19:01:03,375 INFO Client connection 192.168.31.94:51885#true disconnect, remove instances and subscribers
2023-06-11 19:01:05,048 INFO [AUTO-DELETE-IP] service: Service{namespace='public', group='DEFAULT_GROUP', name='springboot-provider', ephemeral=true, revisinotallow=2}, ip: {"ip":"192.168.31.94","port":8083,"healthy":false,"cluster":"DEFAULT","extendDatum":{"management.endpoints.web.base-path":"/actuator","management.port":"18082","preserved.register.source":"SPRING_CLOUD","management.address":"127.0.0.1","customInstanceId":"192.168.31.94#8083#DEFAULT#DEFAULT_GROUP@@springboot-provider"},"lastHeartBeatTime":1686481231604,"metadataId":"192.168.31.94:8083:DEFAULT"}
2023-06-11 19:01:05,048 INFO Client remove for service Service{namespace='public', group='DEFAULT_GROUP', name='springboot-provider', ephemeral=true, revisinotallow=2}, 192.168.31.94:8083#true
2023-06-11 19:01:08,379 INFO Client connection 192.168.31.94:8083#true disconnect, remove instances and subscribers

2.3 服務調用

在 springboot-consumer 上跑一個單元測試的用例，用 FeignClient 調用下面的方法：

@FeignClient(value = "springboot-provider", configuration = FeignMultipartSupportConfig.class)
public interface FeignAsEurekaClient {

    @PostMapping("/employee/save")
    String saveEmployeebyName(@RequestBody Employee employee);

}

日志如下：

2023-06-11 19:15:47,694 [main] [INFO] org.springframework.test.context.transaction.TransactionContext - Began transaction (1) for test context [DefaultTestContext@5bf0d49 testClass = TestFeignAsEurekaClient, testInstance = boot.service.TestFeignAsEurekaClient@10683d9d, testMethod = testPostEmployByFeign@TestFeignAsEurekaClient, testException = [null], mergedContextConfiguration = [WebMergedContextConfiguration@5b7a5baa testClass = TestFeignAsEurekaClient, locations = '{}', classes = '{class boot.Application, class boot.Application}', contextInitializerClasses = '[]', activeProfiles = '{}', propertySourceLocations = '{}', propertySourceProperties = '{org.springframework.boot.test.context.SpringBootTestCnotallow=true, server.port=0}', contextCustomizers = set[org.springframework.boot.test.context.filter.ExcludeFilterContextCustomizer@166fa74d, org.springframework.boot.test.json.DuplicateJsonObjectContextCustomizerFactory$DuplicateJsonObjectContextCustomizer@588df31b, org.springframework.boot.test.mock.mockito.MockitoContextCustomizer@0, org.springframework.boot.test.web.client.TestRestTemplateContextCustomizer@7fad8c79, org.springframework.boot.test.autoconfigure.properties.PropertyMappingContextCustomizer@0, org.springframework.boot.test.autoconfigure.web.servlet.WebDriverContextCustomizerFactory$Customizer@10b48321], resourceBasePath = 'src/main/webapp', contextLoader = 'org.springframework.boot.test.context.SpringBootContextLoader', parent = [null]], attributes = map['org.springframework.test.context.web.ServletTestExecutionListener.activateListener' -> false]]; transaction manager [org.springframework.jdbc.datasource.DataSourceTransactionManager@693676d]; rollback [true]
2023-06-11 19:15:47,941 [main] [INFO] com.netflix.config.ChainedDynamicProperty - Flipping property: springboot-provider.ribbon.ActiveConnectionsLimit to use NEXT property: niws.loadbalancer.availabilityFilteringRule.activeConnectionsLimit = 2147483647
2023-06-11 19:15:47,962 [main] [INFO] com.netflix.loadbalancer.BaseLoadBalancer - Client: springboot-provider instantiated a LoadBalancer: DynamicServerListLoadBalancer:{NFLoadBalancer:name=springboot-provider,current list of Servers=[],Load balancer stats=Zone stats: {},Server stats: []}ServerList:null
2023-06-11 19:15:47,969 [main] [INFO] com.netflix.loadbalancer.DynamicServerListLoadBalancer - Using serverListUpdater PollingServerListUpdater
2023-06-11 19:15:48,064 [main] [INFO] com.netflix.config.ChainedDynamicProperty - Flipping property: springboot-provider.ribbon.ActiveConnectionsLimit to use NEXT property: niws.loadbalancer.availabilityFilteringRule.activeConnectionsLimit = 2147483647
2023-06-11 19:15:48,064 [main] [INFO] com.netflix.loadbalancer.DynamicServerListLoadBalancer - DynamicServerListLoadBalancer for client springboot-provider initialized: DynamicServerListLoadBalancer:{NFLoadBalancer:name=springboot-provider,current list of Servers=[192.168.31.94:8083],Load balancer stats=Zone stats: {unknown=[Zone:unknown; Instance count:1; Active connections count: 0; Circuit breaker tripped count: 0; Active connections per server: 0.0;]
},Server stats: [[Server:192.168.31.94:8083; Zone:UNKNOWN; Total Requests:0; Successive connection failure:0; Total blackout seconds:0; Last connection made:Thu Jan 01 08:00:00 CST 1970; First connection made: Thu Jan 01 08:00:00 CST 1970; Active Connections:0; total failure count in last (1000) msecs:0; average resp time:0.0; 90 percentile resp time:0.0; 95 percentile resp time:0.0; min resp time:0.0; max resp time:0.0; stddev resp time:0.0]
]}ServerList:com.alibaba.cloud.nacos.ribbon.NacosServerList@24d998ba

注意，這里使用了 OpenFeign，其中用到了 Ribbon 做負載均衡，那就需要考慮到 Ribbon 的刷新本地服務列表的時間，從源代碼中看，刷新周期是 30s。如下圖：

Ribbon 刷新緩存的邏輯參考下面代碼：

public synchronized void start(final UpdateAction updateAction) {
 if (isActive.compareAndSet(false, true)) {
  final Runnable wrapperRunnable = new Runnable() {
   @Override
   public void run() {
    //...
   }
  };

  scheduledFuture = getRefreshExecutor().scheduleWithFixedDelay(
    wrapperRunnable,
    initialDelayMs,
    refreshIntervalMs,//這里定義的是30s
    TimeUnit.MILLISECONDS
  );
 }//...
}

3.優雅發布

前面第一節提到過，優雅發布有兩個要求：優雅上線和優雅下線。

Nacos 客戶端和服務端的交互采用長輪詢的方式，服務端收到客戶端的請求后，首先會判斷服務端本地的服務列表是否跟客戶端的相比是否發生變化（比較 MD5），如果發生變化則立即通知客戶端，否則放入長輪詢隊列掛起，如果這段時間內服務列表發生變化，則立刻通知客戶端，否則等到超時后再通知客戶端。代碼如下：

//LongPollingService.java
public void addLongPollingClient(HttpServletRequest req, HttpServletResponse rsp, Map<String, String> clientMd5Map,
  int probeRequestSize) {
 
 String str = req.getHeader(LongPollingService.LONG_POLLING_HEADER);
 int delayTime = SwitchService.getSwitchInteger(SwitchService.FIXED_DELAY_TIME, 500);
 
 // Add delay time for LoadBalance, and one response is returned 500 ms in advance to avoid client timeout.
 long timeout = -1L;
 if (isFixedPolling()) {
  //...
 } else {
  timeout = Math.max(10000, Long.parseLong(str) - delayTime);//29.5s
  long start = System.currentTimeMillis();
  List<String> changedGroups = MD5Util.compareMd5(req, rsp, clientMd5Map);
  if (changedGroups.size() > 0) {
      //服務列表發生變化，直接返回給客戶端
   generateResponse(req, rsp, changedGroups);
   return;
  } //...
 }
 String ip = RequestUtil.getRemoteIp(req);
 //..
 
 // Must be called by http thread, or send response.
 final AsyncContext asyncContext = req.startAsync();
 // AsyncContext.setTimeout() is incorrect, Control by oneself
 asyncContext.setTimeout(0L);
 
 String appName = req.getHeader(RequestUtil.CLIENT_APPNAME_HEADER);
 String tag = req.getHeader("Vipserver-Tag");
 //服務列表沒有發生變化，放入長輪詢隊列等待調度
 ConfigExecutor.executeLongPolling(
   new ClientLongPolling(asyncContext, clientMd5Map, ip, probeRequestSize, timeout, appName, tag));
}

從上面服務端源代碼可以看到，這里超時時間是 30s，其中 29.5s 用于掛起等待，0.5s 檢查服務列表是否發生變化。這里使用了長輪詢，如果服務端列表發生變化，會立刻通知客戶端，所以對優雅發布影響非常小。

服務列表發生變化后，客戶端用單獨的線程通知監聽的 listener，代碼如下：

public void startInternal() {
 executor.schedule(() -> {
  while (!executor.isShutdown() && !executor.isTerminated()) {
   try {
    listenExecutebell.poll(5L, TimeUnit.SECONDS);
    //...
    executeConfigListen();
   } catch (Throwable e) {
    //...
   }
  }
 }, 0L, TimeUnit.MILLISECONDS);
 
}

3.1 優雅上線

優雅上線存在的問題主要在于 Service Provider 注冊到 Nacos 后，服務還沒有完成初始化，請求已經到來。這種情況主要原因是 Service Provider 啟動后立刻注冊 Naocs，而本身提供的接口可能還沒有初始化完成。

這種情況的解決方法是關閉自動注冊：

spring.cloud.nacos.discovery.registerEnabled=false

在服務初始化后使用代碼手動注冊，代碼如下：

Properties setting8 = new Properties();
String serverIp8 = "127.0.0.1:8848";
setting8.put(PropertyKeyConst.SERVER_ADDR, serverIp8);
setting8.put(PropertyKeyConst.USERNAME, "nacos");
setting8.put(PropertyKeyConst.PASSWORD, "nacos");
NamingService inaming8 = NacosFactory.createNamingService(setting7);
inaming8.registerInstance("springboot-provider", "192.168.31.94", 8083);

3.2 優雅下線

服務下線分兩種情況，一個是正常停服，一個是服務故障。

3.2.1 正常停服

對于正常停服，Nacos 采用心跳檢測來實現服務在線。心跳周期是 5s，Nacos Server 如果 15s 沒收到心跳就會將實例設置為不健康，在 30s 沒收到心跳才會講這個服務刪除。當然這個時間可以設置：

spring.cloud.nacos.discovery.metadata.preserved.heart.beat.interval=1000 #心跳間隔5s->1s
spring.cloud.nacos.discovery.metadata.preserved.heart.beat.timeout=3000 #超時時間15s->3s
spring.cloud.nacos.discovery.metadata.preserved.ip.delete.timeout=5000 #刪除時間30s->5s

但這樣并不能保證服務停止后能夠立刻從 Nacos Server 下線，很有可能服務停止后還能再收到請求，最好的方式是手動下線，比如增加一個 API 接口，服務下線之前增加 preStopHook 函數調用這個 API 接口來實現下線。API 接口示例代碼如下：

@GetMapping(value = "/nacos/deregisterInstance")
public String deregisterInstance() {
 Properties prop = new Properties();
 prop.setProperty("serverAddr", "localhost");
 prop.put(PropertyKeyConst.NAMESPACE, "test");
 NacosNamingService client = new NacosNamingService(prop);
 client.deregisterInstance("springboot-provider", "192.168.31.94", 8083);
 return "success";
}

在使用 Ribbon 的場景，也需要考慮 Ribbon 更新本地緩存服務列表的機制，手動下線后，可以再等待 30s 再關閉服務。

3.2.1 服務故障

第二種情況是服務故障，但是并沒有停服，這種情況是很難避免外部請求再發送過來的。處理方式是對這個服務本身的健康檢查結果進行處理，比如連續三次健康檢查失敗，可以調用上面的 API 接口讓服務下線。

4 總結