Eureka 源码解析 —— 应用实例注册发现（六）之全量获取

来源：互联网发布：即时通讯软件编辑：程序博客网时间：2024/05/16 04:19

摘要: 原创出处 http://www.iocoder.cn/Eureka/instance-registry-fetch-all/ 「芋道源码」欢迎转载，保留摘要，谢谢！

本文主要基于 Eureka 1.8.X 版本

1. 概述
2. Eureka-Client 发起全量获取
- 2.1 初始化全量获取
- 2.2 定时获取
- 2.3 刷新注册信息缓存
- 2.4 发起获取注册信息
3. Eureka-Server 接收全量获取
- 3.1 接收全量获取请求
- 3.2 响应缓存 ResponseCache
- 3.3 缓存读取
- 3.4 主动过期读写缓存
- 3.5 被动过期读写缓存
- 3.6 定时刷新只读缓存

1. 概述

本文主要分享 Eureka-Client 向 Eureka-Server 获取全量注册信息的过程。

FROM 《深度剖析服务发现组件Netflix Eureka》

Eureka-Client 获取注册信息，分成全量获取和增量获取。默认配置下，Eureka-Client 启动时，首先执行一次全量获取进行本地缓存注册信息，而后每 30 秒增量获取刷新本地缓存( 非“正常”情况下会是全量获取 )。

本文重点在于全量获取。

推荐 Spring Cloud 书籍：

请支持正版。下载盗版，等于主动编写低级 BUG 。
程序猿DD —— 《Spring Cloud微服务实战》
周立 —— 《Spring Cloud与Docker微服务架构实战》
两书齐买，京东包邮。

2. Eureka-Client 发起全量获取

本小节调用关系如下：

2.1 初始化全量获取

Eureka-Client 启动时，首先执行一次全量获取进行本地缓存注册信息，首先代码如下：

// DiscoveryClient.java
/*
 Applications 在本地的缓存
*/
private final AtomicReference<Applications> localRegionApps = new AtomicReference<Applications>();
DiscoveryClient(ApplicationInfoManager applicationInfoManager, EurekaClientConfig config, AbstractDiscoveryClientOptionalArgs args,
                    Provider<BackupRegistry> backupRegistryProvider) {
                    
     // ... 省略无关代码
     
    // 【3.2.5】初始化应用集合在本地的缓存
    localRegionApps.set(new Applications());
     
    // ... 省略无关代码     
     
    // 【3.2.12】从 Eureka-Server 拉取注册信息
    if (clientConfig.shouldFetchRegistry() && !fetchRegistry(false)) {
        fetchRegistryFromBackup();
    }
     
     // ... 省略无关代码       
}

com.netflix.discovery.shared.Applications，注册的应用集合。较为容易理解，点击链接链接查看带中文注释的类，这里就不啰嗦了。Applications 与 InstanceInfo 类关系如下：
配置 eureka.shouldFetchRegistry = true，开启从 Eureka-Server 获取注册信息。默认值：true 。
调用 #fetchRegistry(false) 方法，从 Eureka-Server 全量获取注册信息，在「2.4 发起获取注册信息」详细解析。

2.2 定时获取

Eureka-Client 在初始化过程中，创建获取注册信息线程，固定间隔向 Eureka-Server 发起获取注册信息( fetch )，刷新本地注册信息缓存。实现代码如下：

// DiscoveryClient.java
DiscoveryClient(ApplicationInfoManager applicationInfoManager, EurekaClientConfig config, AbstractDiscoveryClientOptionalArgs args,
               Provider<BackupRegistry> backupRegistryProvider) {
    // ... 省略无关代码
               
    // 【3.2.9】初始化线程池
    // default size of 2 - 1 each for heartbeat and cacheRefresh
    scheduler = Executors.newScheduledThreadPool(2,
         new ThreadFactoryBuilder()
                 .setNameFormat("DiscoveryClient-%d")
                 .setDaemon(true)
                 .build());
    
    cacheRefreshExecutor = new ThreadPoolExecutor(
         1, clientConfig.getCacheRefreshExecutorThreadPoolSize(), 0, TimeUnit.SECONDS,
         new SynchronousQueue<Runnable>(),
         new ThreadFactoryBuilder()
                 .setNameFormat("DiscoveryClient-CacheRefreshExecutor-%d")
                 .setDaemon(true)
                 .build()
     );  // use direct handoff
    
    // ... 省略无关代码
    
    // 【3.2.14】初始化定时任务
    initScheduledTasks();
    
    // ... 省略无关代码
}
private void initScheduledTasks() {
    // 向 Eureka-Server 心跳（续租）执行器
    if (clientConfig.shouldFetchRegistry()) {
       // registry cache refresh timer
       int registryFetchIntervalSeconds = clientConfig.getRegistryFetchIntervalSeconds();
       int expBackOffBound = clientConfig.getCacheRefreshExecutorExponentialBackOffBound();
       scheduler.schedule(
               new TimedSupervisorTask(
                       "cacheRefresh",
                       scheduler,
                       cacheRefreshExecutor,
                       registryFetchIntervalSeconds,
                       TimeUnit.SECONDS,
                       expBackOffBound,
                       new CacheRefreshThread()
               ),
               registryFetchIntervalSeconds, TimeUnit.SECONDS);
     }
     // ... 省略无关代码
}

初始化定时任务代码，和续租的定时任务代码类似，在《Eureka 源码解析 —— 应用实例注册发现（二）之续租
》有详细解析，这里不重复分享。
com.netflix.discovery.DiscoveryClient.CacheRefreshThread，注册信息缓存刷新任务，实现代码如下：
```
class CacheRefreshThread implements Runnable {
    public void run() {
        refreshRegistry();
    }
}
```
- 调用 #refreshRegistry(false) 方法，刷新注册信息缓存，在「2.3 刷新注册信息缓存」详细解析。

2.3 刷新注册信息缓存

调用 #refreshRegistry(false) 方法，刷新注册信息缓存，实现代码如下：

// DiscoveryClient.java
  1: void refreshRegistry() {
  2:     try {
  3:         // TODO 芋艿：TODO[0009]：RemoteRegionRegistry
  4:         boolean isFetchingRemoteRegionRegistries = isFetchingRemoteRegionRegistries();
  5: 
  6:         boolean remoteRegionsModified = false;
  7:         // This makes sure that a dynamic change to remote regions to fetch is honored.
  8:         String latestRemoteRegions = clientConfig.fetchRegistryForRemoteRegions();
  9:         if (null != latestRemoteRegions) {
 10:             String currentRemoteRegions = remoteRegionsToFetch.get();
 11:             if (!latestRemoteRegions.equals(currentRemoteRegions)) {
 12:                 // Both remoteRegionsToFetch and AzToRegionMapper.regionsToFetch need to be in sync
 13:                 synchronized (instanceRegionChecker.getAzToRegionMapper()) {
 14:                     if (remoteRegionsToFetch.compareAndSet(currentRemoteRegions, latestRemoteRegions)) {
 15:                         String[] remoteRegions = latestRemoteRegions.split(",");
 16:                         remoteRegionsRef.set(remoteRegions);
 17:                         instanceRegionChecker.getAzToRegionMapper().setRegionsToFetch(remoteRegions);
 18:                         remoteRegionsModified = true;
 19:                     } else {
 20:                         logger.info("Remote regions to fetch modified concurrently," +
 21:                                 " ignoring change from {} to {}", currentRemoteRegions, latestRemoteRegions);
 22:                     }
 23:                 }
 24:             } else {
 25:                 // Just refresh mapping to reflect any DNS/Property change
 26:                 instanceRegionChecker.getAzToRegionMapper().refreshMapping();
 27:             }
 28:         }
 29: 
 30:         boolean success = fetchRegistry(remoteRegionsModified);
 31:         if (success) {
 32:             // 设置 注册信息的应用实例数
 33:             registrySize = localRegionApps.get().size();
 34:             // 设置 最后获取注册信息时间
 35:             lastSuccessfulRegistryFetchTimestamp = System.currentTimeMillis();
 36:         }
 37: 
 38:         // 打印日志
 39:         if (logger.isDebugEnabled()) {
 40:             StringBuilder allAppsHashCodes = new StringBuilder();
 41:             allAppsHashCodes.append("Local region apps hashcode: ");
 42:             allAppsHashCodes.append(localRegionApps.get().getAppsHashCode());
 43:             allAppsHashCodes.append(", is fetching remote regions? ");
 44:             allAppsHashCodes.append(isFetchingRemoteRegionRegistries);
 45:             for (Map.Entry<String, Applications> entry : remoteRegionVsApps.entrySet()) {
 46:                 allAppsHashCodes.append(", Remote region: ");
 47:                 allAppsHashCodes.append(entry.getKey());
 48:                 allAppsHashCodes.append(" , apps hashcode: ");
 49:                 allAppsHashCodes.append(entry.getValue().getAppsHashCode());
 50:             }
 51:             logger.debug("Completed cache refresh task for discovery. All Apps hash code is {} ",
 52:                     allAppsHashCodes.toString());
 53:         }
 54:     } catch (Throwable e) {
 55:         logger.error("Cannot fetch registry from server", e);
 56:     }        
 57: }

第 3 至 28 行：TODO[0009]：RemoteRegionRegistry
第 30 行：调用 #fetchRegistry(false) 方法，从 Eureka-Server 获取注册信息，在「2.4 发起获取注册信息」详细解析。

第 31 至 36 行：获取注册信息成功，设置注册信息的应用实例数，最后获取注册信息时间。变量代码如下：

/**
* 注册信息的应用实例数
*/
private volatile int registrySize = 0;
/**
* 最后成功从 Eureka-Server 拉取注册信息时间戳
*/
private volatile long lastSuccessfulRegistryFetchTimestamp = -1;

第 38 至 53 行：打印调试日志。
第 54 至 56 行：打印异常日志。

2.4 发起获取注册信息

调用 #fetchRegistry(false) 方法，从 Eureka-Server 获取注册信息( 根据条件判断，可能是全量，也可能是增量 )，实现代码如下：

 1: private boolean fetchRegistry(boolean forceFullRegistryFetch) {
 2:     Stopwatch tracer = FETCH_REGISTRY_TIMER.start();
 3: 
 4:     try {
 5:         // 获取 本地缓存的注册的应用实例集合
 6:         // If the delta is disabled or if it is the first time, get all
 7:         // applications
 8:         Applications applications = getApplications();
 9: 
10:         // 全量获取
11:         if (clientConfig.shouldDisableDelta() // 禁用增量获取
12:                 || (!Strings.isNullOrEmpty(clientConfig.getRegistryRefreshSingleVipAddress()))
13:                 || forceFullRegistryFetch
14:                 || (applications == null) // 空
15:                 || (applications.getRegisteredApplications().size() == 0) // 空
16:                 || (applications.getVersion() == -1)) //Client application does not have latest library supporting delta
17:         {
18:             logger.info("Disable delta property : {}", clientConfig.shouldDisableDelta());
19:             logger.info("Single vip registry refresh property : {}", clientConfig.getRegistryRefreshSingleVipAddress());
20:             logger.info("Force full registry fetch : {}", forceFullRegistryFetch);
21:             logger.info("Application is null : {}", (applications == null));
22:             logger.info("Registered Applications size is zero : {}",
23:                     (applications.getRegisteredApplications().size() == 0));
24:             logger.info("Application version is -1: {}", (applications.getVersion() == -1));
25:             // 执行 全量获取
26:             getAndStoreFullRegistry();
27:         } else {
28:             // 执行 增量获取
29:             getAndUpdateDelta(applications);
30:         }
31:         // 设置 应用集合 hashcode
32:         applications.setAppsHashCode(applications.getReconcileHashCode());
33:         // 打印 本地缓存的注册的应用实例数量
34:         logTotalInstances();
35:     } catch (Throwable e) {
36:         logger.error(PREFIX + appPathIdentifier + " - was unable to refresh its cache! status = " + e.getMessage(), e);
37:         return false;
38:     } finally {
39:         if (tracer != null) {
40:             tracer.stop();
41:         }
42:     }
43: 
44:     // Notify about cache refresh before updating the instance remote status
45:     onCacheRefreshed();
46: 
47:     // Update remote status based on refreshed data held in the cache
48:     updateInstanceRemoteStatus();
49: 
50:     // registry was fetched successfully, so return true
51:     return true;
52: }

第 5 至 8 行：获取本地缓存的注册的应用实例集合，实现代码如下：
```
public Applications getApplications() {
   return localRegionApps.get();
}
```
第 10 至 26 行：全量获取注册信息。
- 第 11 行：配置 eureka.disableDelta = true ，禁用增量获取注册信息。默认值：false 。
- 第 12 行：只获得一个 vipAddress 对应的应用实例们的注册信息。
- 第 13 行：方法参数 forceFullRegistryFetch 强制全量获取注册信息。
- 第 14 至 15 行：本地缓存为空。
- 第 25 至 26 行：调用 #getAndStoreFullRegistry() 方法，全量获取注册信息，并设置到本地缓存。下文详细解析。
第 27 至 30 行：增量获取注册信息，并刷新本地缓存，在《Eureka 源码解析 —— 应用实例注册发现（七）之增量获取》详细解析。
第 31 至 32 行：计算应用集合 hashcode 。该变量用于校验增量获取的注册信息和 Eureka-Server 全量的注册信息是否一致( 完整 )，在《Eureka 源码解析 —— 应用实例注册发现（七）之增量获取》详细解析。

第 33 至 34 行：打印调试日志，输出本地缓存的注册的应用实例数量。实现代码如下：

private void logTotalInstances() {
   if (logger.isDebugEnabled()) {
       int totInstances = 0;
       for (Application application : getApplications().getRegisteredApplications()) {
           totInstances += application.getInstancesAsIsFromEureka().size();
       }
       logger.debug("The total number of all instances in the client now is {}", totInstances);
   }
}

第 44 至 45 行：触发 CacheRefreshedEvent 事件，事件监听器执行。目前 Eureka 未提供默认的该事件监听器。

#onCacheRefreshed() 方法，实现代码如下：

/**
 * Eureka 事件监听器
 */
private final CopyOnWriteArraySet<EurekaEventListener> eventListeners = new CopyOnWriteArraySet<>();
protected void onCacheRefreshed() {
    fireEvent(new CacheRefreshedEvent());
}
protected void fireEvent(final EurekaEvent event) {
    for (EurekaEventListener listener : eventListeners) {
        listener.onEvent(event);
    }
}

笔者的YY ：你可以实现自定义的事件监听器监听 CacheRefreshedEvent 事件，以达到持久化最新的注册信息到存储器( 例如，本地文件 )，通过这样的方式，配合实现 BackupRegistry 接口读取存储器。BackupRegistry 接口调用如下：
```
// 【3.2.12】从 Eureka-Server 拉取注册信息
if (clientConfig.shouldFetchRegistry() && !fetchRegistry(false)) {
    fetchRegistryFromBackup();
}
```

第47 至 48 行：更新本地缓存的当前应用实例在 Eureka-Server 的状态。

 1: private volatile InstanceInfo.InstanceStatus lastRemoteInstanceStatus = InstanceInfo.InstanceStatus.UNKNOWN; 
 2: 
 3: private synchronized void updateInstanceRemoteStatus() {
 4:     // Determine this instance's status for this app and set to UNKNOWN if not found
 5:     InstanceInfo.InstanceStatus currentRemoteInstanceStatus = null;
 6:     if (instanceInfo.getAppName() != null) {
 7:         Application app = getApplication(instanceInfo.getAppName());
 8:         if (app != null) {
 9:             InstanceInfo remoteInstanceInfo = app.getByInstanceId(instanceInfo.getId());
10:             if (remoteInstanceInfo != null) {
11:                 currentRemoteInstanceStatus = remoteInstanceInfo.getStatus();
12:             }
13:         }
14:     }
15:     if (currentRemoteInstanceStatus == null) {
16:         currentRemoteInstanceStatus = InstanceInfo.InstanceStatus.UNKNOWN;
17:     }
18: 
19:     // Notify if status changed
20:     if (lastRemoteInstanceStatus != currentRemoteInstanceStatus) {
21:         onRemoteStatusChanged(lastRemoteInstanceStatus, currentRemoteInstanceStatus);
22:         lastRemoteInstanceStatus = currentRemoteInstanceStatus;
23:     }
24: }

第 4 至 14 行：从注册信息中获取当前应用在 Eureka-Server 的状态。
第 19 至 23 行：对比本地缓存和最新的的当前应用实例在 Eureka-Server 的状态，若不同，更新本地缓存( 注意，只更新该缓存变量，不更新本地当前应用实例的状态( instanceInfo.status ) )，触发 StatusChangeEvent 事件，事件监听器执行。目前 Eureka 未提供默认的该事件监听器。#onRemoteStatusChanged(...) 实现代码如下：
```
protected void onRemoteStatusChanged(InstanceInfo.InstanceStatus oldStatus, InstanceInfo.InstanceStatus newStatus) {
   fireEvent(new StatusChangeEvent(oldStatus, newStatus));
}
```
- Eureka-Client 本地应用实例与 Eureka-Server 的该应用实例状态不同的原因，因为应用实例的覆盖状态，在《Eureka 源码解析 —— 应用实例注册发现（八）之覆盖状态》有详细解析。

2.4.1 全量获取注册信息，并设置到本地缓存

调用 #getAndStoreFullRegistry() 方法，全量获取注册信息，并设置到本地缓存。下实现代码如下：

 1: private void getAndStoreFullRegistry() throws Throwable {
 2:     long currentUpdateGeneration = fetchRegistryGeneration.get();
 3: 
 4:     logger.info("Getting all instance registry info from the eureka server");
 5: 
 6:     // 全量获取注册信息
 7:     Applications apps = null;
 8:     EurekaHttpResponse<Applications> httpResponse = clientConfig.getRegistryRefreshSingleVipAddress() == null
 9:             ? eurekaTransport.queryClient.getApplications(remoteRegionsRef.get())
10:             : eurekaTransport.queryClient.getVip(clientConfig.getRegistryRefreshSingleVipAddress(), remoteRegionsRef.get());
11:     if (httpResponse.getStatusCode() == Status.OK.getStatusCode()) {
12:         apps = httpResponse.getEntity();
13:     }
14:     logger.info("The response status is {}", httpResponse.getStatusCode());
15: 
16:     // 设置到本地缓存
17:     if (apps == null) {
18:         logger.error("The application is null for some reason. Not storing this information");
19:     } else if (fetchRegistryGeneration.compareAndSet(currentUpdateGeneration, currentUpdateGeneration + 1)) {
20:         localRegionApps.set(this.filterAndShuffle(apps));
21:         logger.debug("Got full registry with apps hashcode {}", apps.getAppsHashCode());
22:     } else {
23:         logger.warn("Not updating applications as another thread is updating it already");
24:     }
25: }

第 6 至 14 行：全量获取注册信息，实现代码如下：

// AbstractJerseyEurekaHttpClient.java
@Override
public EurekaHttpResponse<Applications> getApplications(String... regions) {
   return getApplicationsInternal("apps/", regions);
}
private EurekaHttpResponse<Applications> getApplicationsInternal(String urlPath, String[] regions) {
   ClientResponse response = null;
   String regionsParamValue = null;
   try {
       WebResource webResource = jerseyClient.resource(serviceUrl).path(urlPath);
       if (regions != null && regions.length > 0) {
           regionsParamValue = StringUtil.join(regions);
           webResource = webResource.queryParam("regions", regionsParamValue);
       }
       Builder requestBuilder = webResource.getRequestBuilder();
       addExtraHeaders(requestBuilder);
       response = requestBuilder.accept(MediaType.APPLICATION_JSON_TYPE).get(ClientResponse.class); // JSON
       Applications applications = null;
       if (response.getStatus() == Status.OK.getStatusCode() && response.hasEntity()) {
           applications = response.getEntity(Applications.class);
       }
       return anEurekaHttpResponse(response.getStatus(), Applications.class)
               .headers(headersOf(response))
               .entity(applications)
               .build();
   } finally {
       if (logger.isDebugEnabled()) {
           logger.debug("Jersey HTTP GET {}/{}?{}; statusCode={}",
                   serviceUrl, urlPath,
                   regionsParamValue == null ? "" : "regions=" + regionsParamValue,
                   response == null ? "N/A" : response.getStatus()
           );
       }
       if (response != null) {
           response.close();
       }
   }
}

调用 AbstractJerseyEurekaHttpClient#getApplications(...) 方法，GET 请求 Eureka-Server 的 apps/ 接口，参数为 regions ，返回格式为 JSON ，实现全量获取注册信息。

第 16 至 24 行：设置到本地注册信息缓存。
- 第 19 行：TODO[0025] ：并发更新的情况？？？
- 第 20 行：调用 #filterAndShuffle(...) 方法，根据配置 eureka.shouldFilterOnlyUpInstances = true ( 默认值：true ) 过滤只保留状态为开启( UP )的应用实例，并随机打乱应用实例顺序。打乱后，实现调用应用服务的随机性。代码比较易懂，点击链接查看方法实现。

3. Eureka-Server 接收全量获取

3.1 接收全量获取请求

com.netflix.eureka.resources.ApplicationsResource，处理所有应用的请求操作的 Resource ( Controller )。

接收全量获取请求，映射 ApplicationsResource#getContainers() 方法，实现代码如下：

 1: @GET
 2: public Response getContainers(@PathParam("version") String version,
 3:                               @HeaderParam(HEADER_ACCEPT) String acceptHeader,
 4:                               @HeaderParam(HEADER_ACCEPT_ENCODING) String acceptEncoding,
 5:                               @HeaderParam(EurekaAccept.HTTP_X_EUREKA_ACCEPT) String eurekaAccept,
 6:                               @Context UriInfo uriInfo,
 7:                               @Nullable @QueryParam("regions") String regionsStr) {
 8:     // TODO[0009]：RemoteRegionRegistry
 9:     boolean isRemoteRegionRequested = null != regionsStr && !regionsStr.isEmpty();
10:     String[] regions = null;
11:     if (!isRemoteRegionRequested) {
12:         EurekaMonitors.GET_ALL.increment();
13:     } else {
14:         regions = regionsStr.toLowerCase().split(",");
15:         Arrays.sort(regions); // So we don't have different caches for same regions queried in different order.
16:         EurekaMonitors.GET_ALL_WITH_REMOTE_REGIONS.increment();
17:     }
18: 
19:     // 判断是否可以访问
20:     // Check if the server allows the access to the registry. The server can
21:     // restrict access if it is not
22:     // ready to serve traffic depending on various reasons.
23:     if (!registry.shouldAllowAccess(isRemoteRegionRequested)) {
24:         return Response.status(Status.FORBIDDEN).build();
25:     }
26: 
27:     // API 版本
28:     CurrentRequestVersion.set(Version.toEnum(version));
29: 
30:     // 返回数据格式
31:     KeyType keyType = Key.KeyType.JSON;
32:     String returnMediaType = MediaType.APPLICATION_JSON;
33:     if (acceptHeader == null || !acceptHeader.contains(HEADER_JSON_VALUE)) {
34:         keyType = Key.KeyType.XML;
35:         returnMediaType = MediaType.APPLICATION_XML;
36:     }
37: 
38:     // 响应缓存键( KEY )
39:     Key cacheKey = new Key(Key.EntityType.Application,
40:             ResponseCacheImpl.ALL_APPS,
41:             keyType, CurrentRequestVersion.get(), EurekaAccept.fromString(eurekaAccept), regions
42:     );
43: 
44:     //
45:     Response response;
46:     if (acceptEncoding != null && acceptEncoding.contains(HEADER_GZIP_VALUE)) {
47:         response = Response.ok(responseCache.getGZIP(cacheKey))
48:                 .header(HEADER_CONTENT_ENCODING, HEADER_GZIP_VALUE)
49:                 .header(HEADER_CONTENT_TYPE, returnMediaType)
50:                 .build();
51:     } else {
52:         response = Response.ok(responseCache.get(cacheKey))
53:                 .build();
54:     }
55:     return response;
56: }

第 8 至 17 行：TODO[0009]：RemoteRegionRegistry
第 19 至 25 行：Eureka-Server 启动完成，但是未处于就绪( Ready )状态，不接受请求全量应用注册信息的请求，例如，Eureka-Server 启动时，未能从其他 Eureka-Server 集群的节点获取到应用注册信息。

第 27 至 28 行：设置 API 版本号。默认最新 API 版本为 V2。实现代码如下：

public enum Version {
    V1, V2;
    public static Version toEnum(String v) {
        for (Version version : Version.values()) {
            if (version.name().equalsIgnoreCase(v)) {
                return version;
            }
        }
        //Defaults to v2
        return V2;
    }
}

第 30 至 36 行：设置返回数据格式，默认 JSON 。
第 38 至 42 行：创建响应缓存( ResponseCache ) 的键( KEY )，在「3.2.1 缓存键」详细解析。
第 44 至 55 行：从响应缓存读取全量注册信息，在「3.3 缓存读取」详细解析。

3.2 响应缓存 ResponseCache

com.netflix.eureka.registry.ResponseCache，响应缓存接口，接口代码如下：

public interface ResponseCache {
    String get(Key key);
    
    byte[] getGZIP(Key key);
    
    void invalidate(String appName, @Nullable String vipAddress, @Nullable String secureVipAddress);
    AtomicLong getVersionDelta();
    
    AtomicLong getVersionDeltaWithRegions();
}

其中，#getVersionDelta() 和 #getVersionDeltaWithRegions() 已经废弃。这里保留的原因主要是考虑兼容性。判断依据来自如下代码：

// Applications.java
@Deprecated
public void setVersion(Long version) {
   this.versionDelta = version;
}
// AbstractInstanceRegistry.java
public Applications getApplicationDeltas() {
    // ... 省略其它无关代码
    apps.setVersion(responseCache.getVersionDelta().get()); // 唯一调用到 ResponseCache#getVersionDelta() 方法的地方
    // ... 省略其它无关代码
}

#get() ：获得缓存。
#getGZIP() ：获得缓存，并 GZIP 。
#invalidate() ：过期缓存。

3.2.1 缓存键

com.netflix.eureka.registry.Key，缓存键。实现代码如下：

public class Key {
    public enum KeyType {
        JSON, XML
    }
    /
     * An enum to define the entity that is stored in this cache for this key.
     /
    public enum EntityType {
        Application, VIP, SVIP
    }
    /
      实体名
     /
    private final String entityName;
    /
      TODO[0009]：RemoteRegionRegistry
     /
    private final String[] regions;
    /
      请求参数类型
     /
    private final KeyType requestType;
    /
      请求 API 版本号
     /
    private final Version requestVersion;
    /
      hashKey
     /
    private final String hashKey;
    /
      实体类型
     
      {@link EntityType}
     /
    private final EntityType entityType;
    /
      {@link EurekaAccept}
     */
    private final EurekaAccept eurekaAccept;
    
    public Key(EntityType entityType, String entityName, KeyType type, Version v, EurekaAccept eurekaAccept, @Nullable String[] regions) {
        this.regions = regions;
        this.entityType = entityType;
        this.entityName = entityName;
        this.requestType = type;
        this.requestVersion = v;
        this.eurekaAccept = eurekaAccept;
        hashKey = this.entityType + this.entityName + (null != this.regions ? Arrays.toString(this.regions) : "")
                + requestType.name() + requestVersion.name() + this.eurekaAccept.name();
    }
    
    public Key(EntityType entityType, String entityName, KeyType type, Version v, EurekaAccept eurekaAccept, @Nullable String[] regions) {
        this.regions = regions;
        this.entityType = entityType;
        this.entityName = entityName;
        this.requestType = type;
        this.requestVersion = v;
        this.eurekaAccept = eurekaAccept;
        hashKey = this.entityType + this.entityName + (null != this.regions ? Arrays.toString(this.regions) : "")
                + requestType.name() + requestVersion.name() + this.eurekaAccept.name();
    }
    
    @Override
    public int hashCode() {
        String hashKey = getHashKey();
        return hashKey.hashCode();
    }
    @Override
    public boolean equals(Object other) {
        if (other instanceof Key) {
            return getHashKey().equals(((Key) other).getHashKey());
        } else {
            return false;
        }
    }
    
}

3.2.2 响应缓存实现类

com.netflix.eureka.registry.ResponseCacheImpl，响应缓存实现类。

在 ResponseCacheImpl 里，将缓存拆分成两层：

只读缓存( readOnlyCacheMap )
固定过期 + 固定大小的读写缓存( readWriteCacheMap )

默认配置下，缓存读取策略如下：

缓存过期策略如下：

应用实例注册、下线、过期时，只只只过期 readWriteCacheMap 。
readWriteCacheMap 写入一段时间( 可配置 )后自动过期。
定时任务对比 readWriteCacheMap 和 readOnlyCacheMap 的缓存值，若不一致，以前者为主。通过这样的方式，实现了 readOnlyCacheMap 的定时过期。

注意：应用实例注册、下线、过期时，不会很快刷新到 readWriteCacheMap 缓存里。默认配置下，最大延迟在 30 秒。

为什么可以使用缓存？

在 CAP 的选择上，Eureka 选择了 AP ，不同于 Zookeeper 选择了 CP 。

3.3 缓存读取

调用 ResponseCacheImpl#get(...) 方法( #getGzip(...) 类似 )，读取缓存，实现代码如下：

 1: private final ConcurrentMap<Key, Value> readOnlyCacheMap = new ConcurrentHashMap<Key, Value>();
 2: 
 3: private final LoadingCache<Key, Value> readWriteCacheMap;
 4: 
 5: public String get(final Key key) {
 6:     return get(key, shouldUseReadOnlyResponseCache);
 7: }
 8: 
 9: String get(final Key key, boolean useReadOnlyCache) {
10:     Value payload = getValue(key, useReadOnlyCache);
11:     if (payload == null || payload.getPayload().equals(EMPTY_PAYLOAD)) {
12:         return null;
13:     } else {
14:         return payload.getPayload();
15:     }
16: }
17: 
18: Value getValue(final Key key, boolean useReadOnlyCache) {
19:     Value payload = null;
20:     try {
21:         if (useReadOnlyCache) {
22:             final Value currentPayload = readOnlyCacheMap.get(key);
23:             if (currentPayload != null) {
24:                 payload = currentPayload;
25:             } else {
26:                 payload = readWriteCacheMap.get(key);
27:                 readOnlyCacheMap.put(key, payload);
28:             }
29:         } else {
30:             payload = readWriteCacheMap.get(key);
31:         }
32:     } catch (Throwable t) {
33:         logger.error("Cannot get value for key :" + key, t);
34:     }
35:     return payload;
36: }

第 5 至 7 行：调用 #get(key, useReadOnlyCache) 方法，读取缓存。其中 shouldUseReadOnlyResponseCache 通过配置 eureka.shouldUseReadOnlyResponseCache = true (默认值：true ) 开启只读缓存。如果你对数据的一致性有相对高的要求，可以关闭这个开关，当然因为少了 readOnlyCacheMap ，性能会有一定的下降。

第 9 至 16 行：调用 getValue(key, useReadOnlyCache) 方法，读取缓存。从 readOnlyCacheMap 和 readWriteCacheMap 变量可以看到缓存值的类为 com.netflix.eureka.registry.ResponseCacheImpl.Value ，实现代码如下：

public class Value {
   /**
    * 原始值
    */
   private final String payload;
   /**
    * GZIP 压缩后的值
    */
   private byte[] gzipped;
   public Value(String payload) {
       this.payload = payload;
       if (!EMPTY_PAYLOAD.equals(payload)) {
           // ... 省略 GZIP 压缩代码
           gzipped = bos.toByteArray();
       } else {
           gzipped = null;
       }
   }
   public String getPayload() {
       return payload;
   }
   public byte[] getGzipped() {
       return gzipped;
   }
}

第 21 至 31 行：读取缓存。

第 21 至 28 行：先读取 readOnlyCacheMap 。读取不到，读取 readWriteCacheMap ，并设置到 readOnlyCacheMap 。
第 29 至 31 行：读取 readWriteCacheMap 。

readWriteCacheMap 实现代码如下：

this.readWriteCacheMap =
      CacheBuilder.newBuilder().initialCapacity(1000)
              .expireAfterWrite(serverConfig.getResponseCacheAutoExpirationInSeconds(), TimeUnit.SECONDS)
              .removalListener(new RemovalListener<Key, Value>() {
                  @Override
                  public void onRemoval(RemovalNotification<Key, Value> notification) {
                      // TODO[0009]：RemoteRegionRegistry
                      Key removedKey = notification.getKey();
                      if (removedKey.hasRegions()) {
                          Key cloneWithNoRegions = removedKey.cloneWithoutRegions();
                          regionSpecificKeys.remove(cloneWithNoRegions, removedKey);
                      }
                  }
              })
              .build(new CacheLoader<Key, Value>() {
                  @Override
                  public Value load(Key key) throws Exception {
                      // // TODO[0009]：RemoteRegionRegistry
                      if (key.hasRegions()) {
                          Key cloneWithNoRegions = key.cloneWithoutRegions();
                          regionSpecificKeys.put(cloneWithNoRegions, key);
                      }
                      Value value = generatePayload(key);
                      return value;
                  }
              });

readWriteCacheMap 最大缓存数量为 1000 。
调用 #generatePayload(key) 方法，生成缓存值。

#generatePayload(key) 方法，实现代码如下：

 1: private Value generatePayload(Key key) {
 2:     Stopwatch tracer = null;
 3:     try {
 4:         String payload;
 5:         switch (key.getEntityType()) {
 6:             case Application:
 7:                 boolean isRemoteRegionRequested = key.hasRegions();
 8: 
 9:                 if (ALL_APPS.equals(key.getName())) {
10:                     if (isRemoteRegionRequested) { // TODO[0009]：RemoteRegionRegistry
11:                         tracer = serializeAllAppsWithRemoteRegionTimer.start();
12:                         payload = getPayLoad(key, registry.getApplicationsFromMultipleRegions(key.getRegions()));
13:                     } else {
14:                         tracer = serializeAllAppsTimer.start();
15:                         payload = getPayLoad(key, registry.getApplications());
16:                     }
17:                 } else if (ALL_APPS_DELTA.equals(key.getName())) {
18:                     // ... 省略增量获取相关的代码
19:                  } else {
20:                     tracer = serializeOneApptimer.start();
21:                     payload = getPayLoad(key, registry.getApplication(key.getName()));
22:                 }
23:                 break;
24:             // ... 省略部分代码 
25:         }
26:         return new Value(payload);
27:     } finally {
28:         if (tracer != null) {
29:             tracer.stop();
30:         }
31:     }
32: }

第 10 至 12 行：TODO[0009]：RemoteRegionRegistry
第 13 至 16 行：调用 AbstractInstanceRegistry#getApplications() 方法，获得注册的应用集合。后调用 #getPayLoad() 方法，将注册的应用集合转换成缓存值。�� 这两个方法代码较多，下面详细解析。
第 17 至 18 行：获取增量注册信息的缓存值，在《Eureka 源码解析 —— 应用实例注册发现（七）之增量获取》详细解析。

3.3.1 获得注册的应用集合

调用 AbstractInstanceRegistry#getApplications() 方法，获得注册的应用集合，实现代码如下：

 1: // AbstractInstanceRegistry.java
 2: 
 3: private static final String[] EMPTY_STR_ARRAY = new String[0];
 4: 
 5: public Applications getApplications() {
 6:    boolean disableTransparentFallback = serverConfig.disableTransparentFallbackToOtherRegion();
 7:    if (disableTransparentFallback) { // TODO[0009]：RemoteRegionRegistry
 8:        return getApplicationsFromLocalRegionOnly();
 9:    } else {
10:        return getApplicationsFromAllRemoteRegions();  // Behavior of falling back to remote region can be disabled.
11:    }
12: }
13: 
14: public Applications getApplicationsFromLocalRegionOnly() {
15:    return getApplicationsFromMultipleRegions(EMPTY_STR_ARRAY);
16: }

第 6 至 8 行：TODO[0009]：RemoteRegionRegistry

第 9 至 16 行：调用 #getApplicationsFromMultipleRegions(...) 方法，获得注册的应用集合，实现代码如下：

 1: public Applications getApplicationsFromMultipleRegions(String[] remoteRegions) {
 2:     // TODO[0009]：RemoteRegionRegistry
 3:     boolean includeRemoteRegion = null != remoteRegions && remoteRegions.length != 0;
 4:     logger.debug("Fetching applications registry with remote regions: {}, Regions argument {}",
 5:             includeRemoteRegion, Arrays.toString(remoteRegions));
 6:     if (includeRemoteRegion) {
 7:         GET_ALL_WITH_REMOTE_REGIONS_CACHE_MISS.increment();
 8:     } else {
 9:         GET_ALL_CACHE_MISS.increment();
10:     }
11:     // 获得获得注册的应用集合
12:     Applications apps = new Applications();
13:     apps.setVersion(1L);
14:     for (Entry<String, Map<String, Lease<InstanceInfo>>> entry : registry.entrySet()) {
15:         Application app = null;
16: 
17:         if (entry.getValue() != null) {
18:             for (Entry<String, Lease<InstanceInfo>> stringLeaseEntry : entry.getValue().entrySet()) {
19:                 Lease<InstanceInfo> lease = stringLeaseEntry.getValue();
20:                 if (app == null) {
21:                     app = new Application(lease.getHolder().getAppName());
22:                 }
23:                 app.addInstance(decorateInstanceInfo(lease));
24:             }
25:         }
26:         if (app != null) {
27:             apps.addApplication(app);
28:         }
29:     }
30:     // TODO[0009]：RemoteRegionRegistry
31:     if (includeRemoteRegion) {
32:         for (String remoteRegion : remoteRegions) {
33:             RemoteRegionRegistry remoteRegistry = regionNameVSRemoteRegistry.get(remoteRegion);
34:             if (null != remoteRegistry) {
35:                 Applications remoteApps = remoteRegistry.getApplications();
36:                 for (Application application : remoteApps.getRegisteredApplications()) {
37:                     if (shouldFetchFromRemoteRegistry(application.getName(), remoteRegion)) {
38:                         logger.info("Application {}  fetched from the remote region {}",
39:                                 application.getName(), remoteRegion);
40: 
41:                         Application appInstanceTillNow = apps.getRegisteredApplications(application.getName());
42:                         if (appInstanceTillNow == null) {
43:                             appInstanceTillNow = new Application(application.getName());
44:                             apps.addApplication(appInstanceTillNow);
45:                         }
46:                         for (InstanceInfo instanceInfo : application.getInstances()) {
47:                             appInstanceTillNow.addInstance(instanceInfo);
48:                         }
49:                     } else {
50:                         logger.debug("Application {} not fetched from the remote region {} as there exists a "
51:                                         + "whitelist and this app is not in the whitelist.",
52:                                 application.getName(), remoteRegion);
53:                     }
54:                 }
55:             } else {
56:                 logger.warn("No remote registry available for the remote region {}", remoteRegion);
57:             }
58:         }
59:     }
60:     // 设置 应用集合 hashcode
61:     apps.setAppsHashCode(apps.getReconcileHashCode());
62:     return apps;
63: }

第 2 至第 10 行：TODO[0009]：RemoteRegionRegistry
第 11 至 29 行：获得获得注册的应用集合。
第 30 至 59 行：TODO[0009]：RemoteRegionRegistry
第 61 行：计算应用集合 hashcode 。该变量用于校验增量获取的注册信息和 Eureka-Server 全量的注册信息是否一致( 完整 )，在《Eureka 源码解析 —— 应用实例注册发现（七）之增量获取》详细解析。

3.3.2 转换成缓存值

调用 #getPayLoad() 方法，将注册的应用集合转换成缓存值，实现代码如下：

/*
 Generate pay load with both JSON and XML formats for all applications.
*/
private String getPayLoad(Key key, Applications apps) {
   // 获得编码器
   EncoderWrapper encoderWrapper = serverCodecs.getEncoder(key.getType(), key.getEurekaAccept());
   String result;
   try {
       // 编码
       result = encoderWrapper.encode(apps);
   } catch (Exception e) {
       logger.error("Failed to encode the payload for all apps", e);
       return "";
   }
   if(logger.isDebugEnabled()) {
       logger.debug("New application cache entry {} with apps hashcode {}", key.toStringCompact(), apps.getAppsHashCode());
   }
   return result;
}

3.4 主动过期读写缓存

应用实例注册、下线、过期时，调用 ResponseCacheImpl#invalidate() 方法，主动过期读写缓存( readWriteCacheMap )，实现代码如下：

public void invalidate(String appName, @Nullable String vipAddress, @Nullable String secureVipAddress) {
   for (Key.KeyType type : Key.KeyType.values()) {
       for (Version v : Version.values()) {
           invalidate(
                   new Key(Key.EntityType.Application, appName, type, v, EurekaAccept.full),
                   new Key(Key.EntityType.Application, appName, type, v, EurekaAccept.compact),
                   new Key(Key.EntityType.Application, ALL_APPS, type, v, EurekaAccept.full),
                   new Key(Key.EntityType.Application, ALL_APPS, type, v, EurekaAccept.compact),
                   new Key(Key.EntityType.Application, ALL_APPS_DELTA, type, v, EurekaAccept.full),
                   new Key(Key.EntityType.Application, ALL_APPS_DELTA, type, v, EurekaAccept.compact)
           );
           if (null != vipAddress) {
               invalidate(new Key(Key.EntityType.VIP, vipAddress, type, v, EurekaAccept.full));
           }
           if (null != secureVipAddress) {
               invalidate(new Key(Key.EntityType.SVIP, secureVipAddress, type, v, EurekaAccept.full));
           }
       }
   }
}

调用 #invalidate(keys) 方法，逐个过期每个缓存键值，实现代码如下：

public void invalidate(Key... keys) {
   for (Key key : keys) {
       logger.debug("Invalidating the response cache key : {} {} {} {}, {}", key.getEntityType(), key.getName(), key.getVersion(), key.getType(), key.getEurekaAccept());
       // 过期读写缓存
       readWriteCacheMap.invalidate(key);
       // TODO[0009]：RemoteRegionRegistry
       Collection<Key> keysWithRegions = regionSpecificKeys.get(key);
       if (null != keysWithRegions && !keysWithRegions.isEmpty()) {
           for (Key keysWithRegion : keysWithRegions) {
               logger.debug("Invalidating the response cache key : {} {} {} {} {}",
                       key.getEntityType(), key.getName(), key.getVersion(), key.getType(), key.getEurekaAccept());
               readWriteCacheMap.invalidate(keysWithRegion);
           }
       }
   }
}

3.5 被动过期读写缓存

读写缓存( readWriteCacheMap ) 写入后，一段时间自动过期，实现代码如下：

expireAfterWrite(serverConfig.getResponseCacheAutoExpirationInSeconds())

配置 eureka.responseCacheAutoExpirationInSeconds ，设置写入过期时长。默认值：180 秒。

3.6 定时刷新只读缓存

定时任务对比 readWriteCacheMap 和 readOnlyCacheMap 的缓存值，若不一致，以前者为主。通过这样的方式，实现了 readOnlyCacheMap 的定时过期。实现代码如下：

 1: ResponseCacheImpl(EurekaServerConfig serverConfig, ServerCodecs serverCodecs, AbstractInstanceRegistry registry) {
 2:     // ... 省略无关代码 
 3: 
 4:     long responseCacheUpdateIntervalMs = serverConfig.getResponseCacheUpdateIntervalMs();
 5:     // ... 省略无关代码
 6: 
 7:     if (shouldUseReadOnlyResponseCache) {
 8:         timer.schedule(getCacheUpdateTask(),
 9:                 new Date(((System.currentTimeMillis() / responseCacheUpdateIntervalMs) * responseCacheUpdateIntervalMs)
10:                         + responseCacheUpdateIntervalMs),
11:                 responseCacheUpdateIntervalMs);
12:     }
13: 
14:     // ... 省略无关代码
15: }
16: 
17: private TimerTask getCacheUpdateTask() {
18:     return new TimerTask() {
19:         @Override
20:         public void run() {
21:             logger.debug("Updating the client cache from response cache");
22:             for (Key key : readOnlyCacheMap.keySet()) { // 循环 readOnlyCacheMap 的缓存键
23:                 if (logger.isDebugEnabled()) {
24:                     Object[] args = {key.getEntityType(), key.getName(), key.getVersion(), key.getType()};
25:                     logger.debug("Updating the client cache from response cache for key : {} {} {} {}", args);
26:                 }
27:                 try {
28:                     CurrentRequestVersion.set(key.getVersion());
29:                     Value cacheValue = readWriteCacheMap.get(key);
30:                     Value currentCacheValue = readOnlyCacheMap.get(key);
31:                     if (cacheValue != currentCacheValue) { // 不一致时，进行替换
32:                         readOnlyCacheMap.put(key, cacheValue);
33:                     }
34:                 } catch (Throwable th) {
35:                     logger.error("Error while updating the client cache from response cache for key {}", key.toStringCompact(), th);
36:                 }
37:             }
38:         }
39:     };
40: }

第 7 至 12 行：初始化定时任务。配置 eureka.responseCacheUpdateIntervalMs，设置任务执行频率，默认值：30 * 1000 毫秒。
第 17 至 39 行：创建定时任务。
- 第 22 行：循环 readOnlyCacheMap 的缓存键。为什么不循环 readWriteCacheMap 呢？ readOnlyCacheMap 的缓存过期依赖 readWriteCacheMap，因此缓存键会更多。
- 第 28 行至 33 行：对比 readWriteCacheMap 和 readOnlyCacheMap 的缓存值，若不一致，以前者为主。通过这样的方式，实现了 readOnlyCacheMap 的定时过期。

阅读全文

0 0