springcloud源码之eureka-client服务发现

springcloud源码之eureka-client服务发现

前言

请先看服务注册，这篇博文把代码入口说清楚了，服务发现和心跳的代码在一个地方

服务发现

private int registryFetchIntervalSeconds = 30;
下面这个线程就是服务发现的，也是由一个定时器每隔30s去eureka-server拿一次数据

class CacheRefreshThread implements Runnable {
        public void run() {
            refreshRegistry();
        }
    }

refreshRegistry关键代码如下

		//fetchRegistry其实在eureka-client初始化的时候拿过一次，上篇说了
		boolean success = fetchRegistry(remoteRegionsModified);
            if (success) {
                registrySize = localRegionApps.get().size();
                lastSuccessfulRegistryFetchTimestamp = System.currentTimeMillis();
            }

fetchRegistry关键代码如下

//forceFullRegistryFetch=true 代表全量拉取
private boolean fetchRegistry(boolean forceFullRegistryFetch) {
		//从缓存中拿到已存的微服务信息
       Applications applications = getApplications();
       	//如果配置了不增量拉取||配置了vip地址||applications ==null||没有一个微服务信息||之前没有拉取过
		if (clientConfig.shouldDisableDelta()
                    || (!Strings.isNullOrEmpty(clientConfig.getRegistryRefreshSingleVipAddress()))
                    || forceFullRegistryFetch
                    || (applications == null)
                    || (applications.getRegisteredApplications().size() == 0)
                    || (applications.getVersion() == -1))
            {
            	//全量拉取
                getAndStoreFullRegistry();
            } else {
            	//增量拉取
                getAndUpdateDelta(applications);
            }
        return true;
    }

全量拉取

 private void getAndStoreFullRegistry() throws Throwable {
        Applications apps = null;
        //访问eureka-server的接口ApplicationsResource#getContainers拉取全部的微服务实例
        //eureka-sever缓存设计这个博客说了getContainers
        EurekaHttpResponse<Applications> httpResponse = clientConfig.getRegistryRefreshSingleVipAddress() == null
                ? eurekaTransport.queryClient.getApplications(remoteRegionsRef.get())
                : eurekaTransport.queryClient.getVip(clientConfig.getRegistryRefreshSingleVipAddress(), remoteRegionsRef.get());
                
        if (httpResponse.getStatusCode() == Status.OK.getStatusCode()) {
            apps = httpResponse.getEntity();
        }
       //更新缓存
		localRegionApps.set(this.filterAndShuffle(apps));
       
    }

增量拉取

private void getAndUpdateDelta(Applications applications) throws Throwable {
       //存储增量微服务
        Applications delta = null;
        //去访问ApplicationsResource#getContainerDifferential接口增量拉取微服务
        //这里具体服务端是如何控制哪些微服务是要被增量拉取的待会再说
        EurekaHttpResponse<Applications> httpResponse = eurekaTransport.queryClient.getDelta(remoteRegionsRef.get());
        
        if (httpResponse.getStatusCode() == Status.OK.getStatusCode()) {
            delta = httpResponse.getEntity();
        }
		//没有拿到增量就去再全量拉取一遍
        if (delta == null) {
            getAndStoreFullRegistry();
        } else if (fetchRegistryGeneration.compareAndSet(currentUpdateGeneration, currentUpdateGeneration + 1)) {
         	//加锁，因为上面有网络请求耗时操作，可能这块代码会有并发
            if (fetchRegistryUpdateLock.tryLock()) {
                try {
               		//去更新本地缓存，根据微服务的上一次操作类型就更新
               		//MODIFIED和ADDED类型就执行add操作
               		//DELETED类型就执行remove操作
                    updateDelta(delta);
                } finally {
                    fetchRegistryUpdateLock.unlock();
                }
            } else {
                logger.warn("Cannot acquire update lock, aborting getAndUpdateDelta");
            }
    }

服务端是如何判断那些微服务属于增量的

//getContainerDifferential会构造一个ALL_APPS_DELTA类型的key
Key cacheKey = new Key(Key.EntityType.Application,
                ResponseCacheImpl.ALL_APPS_DELTA,
                keyType, CurrentRequestVersion.get(), EurekaAccept.fromString(eurekaAccept), regions
        );
responseCache.getGZIP(cacheKey) 
----------------->
下面这些代码在缓存设计博客说到了
Value getValue(final Key key, boolean useReadOnlyCache) {
        Value payload = null;
        if (useReadOnlyCache) {
        		//先从只读缓存拿
                final Value currentPayload = readOnlyCacheMap.get(key);
                if (currentPayload != null) {
                	//拿到了直接返回
                    payload = currentPayload;
                } else {
                	//去读写缓存拿，拿到了回写只读缓存
                	//readWriteCacheMap是基于guava的cache来的，如果readWriteCacheMap拿不到
                	//会回调ResponseCacheImpl的构造方法里的generatePayload(key);这行代码
                    payload = readWriteCacheMap.get(key);
                    readOnlyCacheMap.put(key, payload);
                }
            } else {
                payload = readWriteCacheMap.get(key);
            }    
        return payload;
    }

generatePayload(key)

//获取增量的逻辑
payload = getPayLoad(key, registry.getApplicationDeltas());

registry.getApplicationDeltas()

public Applications getApplicationDeltas() {
        
        Applications apps = new Applications();
    
        Map<String, Application> applicationInstancesMap = new HashMap<String, Application>();
      	//recentlyChangedQueue这个队列里面就是存储增量微服务的，这个队列有个神奇的地方，只会存储最近操作三分钟以内的微服务
      	Iterator<RecentlyChangedItem> iter = this.recentlyChangedQueue.iterator();
      	//遍历增量队列返回增量
         while (iter.hasNext()) {
          }
         return apps;     
    }

增量队列三分钟的代码

private TimerTask getDeltaRetentionTask() {
        return new TimerTask() {
            @Override
            public void run() {
                Iterator<RecentlyChangedItem> it = recentlyChangedQueue.iterator();
                while (it.hasNext()) {
                	//getRetentionTimeInMSInDeltaQueue=3min
                    if (it.next().getLastUpdateTime() <
                            System.currentTimeMillis() - serverConfig.getRetentionTimeInMSInDeltaQueue()) {
                        it.remove();
                    } else {
                        break;
                    }
                }
            }

        };
    }

总结

1：eureka-client每隔30s去eureka-client拉取一次信息
2：在默认配置情况下，可能初始化的时候全量拉取，稳定后基本上都是增量拉取
3：增量拉取失败会进行全量拉取
4：eureka-server保有一个recentlyChangeQueue存储最近三分钟发生改动的微服务实例数据(定时器实现)，客户端进行增量拉取就是拉取这个队列