Nacos10# 健康检查类型与场景
共 21897字,需浏览 44分钟
·
2021-07-26 16:31
引言
Nacos支持众多健康检查类型,心跳、HTTP、TCP、MySQL等类型,这些都作用于什么场景?他们又是如何实现的呢?本文就撸一撸这个。
临时节点续约
临时节点续约通过gRPC连接保鲜实现 执行频率5秒一次 检查结果健康刷新保鲜时间 检查结果不可用标记节点不健康 当节点不健康时重新连接时会从server列表选择下一个节点连接
持久节点心跳检测
心跳执行器通过每隔五秒中向Nacos Server发起HTTP请求 如果返回的server not found会向Nacos Server发起注册请求重新注册
持久节点探活
Nacos探活只有在持久节点注册时才会支持 探活支持HTTP、TCP、Mysql三种探活类型 HTTP通过检测返回200状态码标记是否健康 TPC通过Channel连接方式标记是否健康 Mysql则保证当前节点为主节点,可用于主从切换场景
在《Nacos2# 服务注册与发现客户端示例与源码解析(二)》分析gRPC Client启动逻辑时有分析连接健康检查逻辑。具体代码在RpcClient#start()中,下面再次聚焦下。
client连接心跳检查
clientEventExecutor.submit(new Runnable() {
@Override
public void run() {
while (true) {
try {
// 获取重定向连接上下文,指重新连接到其他server节点
ReconnectContext reconnectContext = reconnectionSignal
.poll(keepAliveTime, TimeUnit.MILLISECONDS);
if (reconnectContext == null) {
// check alive time.
// client活动时间超过5秒钟,向Nacos Server发起健康检测
if (System.currentTimeMillis() - lastActiveTimeStamp >= keepAliveTime) {
// 发送健康检查
boolean isHealthy = healthCheck();
// 非健康节点
if (!isHealthy) {
if (currentConnection == null) {
continue;
}
LoggerUtils.printIfInfoEnabled(LOGGER,
"[{}]Server healthy check fail,currentConnection={}", name,
currentConnection.getConnectionId());
// 标记客户端状态为unhealthy
rpcClientStatus.set(RpcClientStatus.UNHEALTHY);
// 重置ReconnectContext移除serverInfo
reconnectContext = new ReconnectContext(null, false);
// 健康连接更新时间戳
lastActiveTimeStamp = System.currentTimeMillis();
continue;
}
} else {
// 心跳保鲜未过期,跳过本次检测
continue;
}
}
// ...
} catch (Throwable throwable) {
//Do nothing
}
}
}
});
服务端响应
代码翻到GrpcRequestAcceptor#request部分,执行RequestHandler逻辑。
Response response = requestHandler.handleRequest(request, requestMeta);
@TpsControl(pointName = "HealthCheck")
public HealthCheckResponse handle(HealthCheckRequest request, RequestMeta meta) {
return new HealthCheckResponse();
}
当服务端收到健康检查请求时,通过HealthCheckRequestHandler#handle返回HealthCheckResponse。
节点选择
while (startUpRetryTimes > 0 && connectToServer == null) {
try {
startUpRetryTimes--;
ServerInfo serverInfo = nextRpcServer(); // 需连接的节点
LoggerUtils.printIfInfoEnabled(LOGGER, "[{}] Try to connect to server on start up, server: {}", name,
serverInfo);
connectToServer = connectToServer(serverInfo); // 发起rpc连接,连接到集群中其他节点
} catch (Throwable e) {
LoggerUtils.printIfWarnEnabled(LOGGER,
"[{}]Fail to connect to server on start up, error message={}, start up retry times left: {}",
name, e.getMessage(), startUpRetryTimes);
}
}
在选择节点时从server地址列表中自增选择下一个。
@Override
public String genNextServer() {
int index = currentIndex.incrementAndGet() % getServerList().size(); // 获取集群中下一个节点
return getServerList().get(index);
}
清理无效连接
详见:RpcScheduledExecutor#start();定时任务会清理无效连接
// 定时任务每3秒执行一次
RpcScheduledExecutor.COMMON_SERVER_EXECUTOR.scheduleWithFixedDelay(new Runnable() {
@Override
public void run() {
// 超过保鲜时间的连接,重新异步发起连接
connection.asyncRequest(clientDetectionRequest, new RequestCallBack() {
// 刷新激活时间
connection.freshActiveTime();
successConnections.add(outDateConnectionId);
//当失效时(保鲜时间超过20秒),注销关闭connection
unregister(outDateConnectionId);
}
}, 1000L, 3000L, TimeUnit.MILLISECONDS);
小结: 在临时节点注册时,客户端gRPC启动时会启动一个守护线程用户健康检查;健康检查的频率为5秒执行一次;当检查结果健康则刷新保鲜时间;检查结果不可用标记gRPC客户端状态为unhealthy;不健康节点在发起连接时会从server地址列表中选择下一个发起连接;服务端也会定时清理超过保鲜时间连接。
永久节点心跳检测在《Nacos2# 服务注册与发现客户端示例与源码解析(二)》HTTP心跳检测器有详细分析,这里把内容要点摘录如下:
HTTP心跳检测只适用于注册的节点持久节点,临时节点会使用grpc代理 心跳执行器通过每隔五秒中向Nacos Server发起HTTP请求 如果返回的server not found会向Nacos Server发起注册请求重新注册
持久节点探活支持HTTP、TCP和Mysql几种类型,下面以HTTP为例分析其运行逻辑。
探活示例
当注册时节点设置为不健康即Healthy设置为false,当服务端探活正常后将节点设置为健康即Healthy为true。
@SpringBootApplication
public class BootApplication {
public static void main(String[] args) throws Exception {
SpringApplication.run(BootApplication.class, args);
System.setProperty("serverAddr", "127.0.0.1:8848");
System.setProperty("namespace", "public");
Properties properties = new Properties();
properties.setProperty("serverAddr", System.getProperty("serverAddr"));
properties.setProperty("namespace", System.getProperty("namespace"));
Instance instance = new Instance();
instance.setClusterName("clusterDemo3");
instance.setHealthy(false);//默认健康检查通过后,才认为是真正的健康
instance.setIp(getIpAddress());
instance.setPort(8282);
instance.setWeight(100);
instance.setServiceName("AppNacosDemo3");
instance.setEphemeral(false);
Map<String, String> map = new HashMap<String, String>();
map.put("unit", "shunit");
instance.setMetadata(map);
NamingService naming = NamingFactory.createNamingService(properties);
naming.registerInstance("AppNacosDemo3", instance);
}
private static String getIpAddress() throws SocketException {
Enumeration<NetworkInterface> allNetInterfaces = NetworkInterface.getNetworkInterfaces();
while (allNetInterfaces.hasMoreElements()) {
NetworkInterface netInterface = allNetInterfaces.nextElement();
if (!netInterface.isLoopback() && !netInterface.isVirtual() && netInterface.isUp()) {
Enumeration<InetAddress> addresses = netInterface.getInetAddresses();
while (addresses.hasMoreElements()) {
InetAddress ip = addresses.nextElement();
final String hostAddress = ip.getHostAddress();
if (ip instanceof Inet4Address) {
return hostAddress;
}
}
}
}
return "";
}
}
@RestController
public class HelloController {
@RequestMapping(value = "/bike",method = RequestMethod.GET)
public void hello(){
System.out.println("receive message from server.");
}
}
探活地址设置
探活输出
receive message from server.
receive message from server.
小结: 启动时注册节点为非健康节点,Nacos通过检查路径请求返回200正确后将节点设置为健康状态。
源码分析
当持久节点注册时,会请求到InstanceController#register方法。
@Override
public void registerInstance(String namespaceId, String serviceName, Instance instance){
boolean ephemeral = instance.isEphemeral();
String clientId = IpPortBasedClient.getClientId(instance.toInetAddr(), ephemeral);
createIpPortClientIfAbsent(clientId, ephemeral); // 创建IpPortClient
Service service = getService(namespaceId, serviceName, ephemeral);
clientOperationService.registerInstance(service, instance, clientId);
}
public IpPortBasedClient(String clientId, boolean ephemeral) {
this.ephemeral = ephemeral;
this.clientId = clientId;
this.responsibleId = getResponsibleTagFromId();
if (ephemeral) {
beatCheckTask = new ClientBeatCheckTaskV2(this);
HealthCheckReactor.scheduleCheck(beatCheckTask);
} else {
// 持久节点创建HealthCheckTaskV2并定时任务调度
healthCheckTaskV2 = new HealthCheckTaskV2(this);
HealthCheckReactor.scheduleCheck(healthCheckTaskV2);
}
}
@Override
public void doHealthCheck() {
try {
// 获取Client的service列表
for (Service each : client.getAllPublishedService()) {
// 开启健康检查
if (switchDomain.isHealthCheckEnabled(each.getGroupedServiceName())) {
// 注册节点信息
InstancePublishInfo instancePublishInfo = client.getInstancePublishInfo(each);
ClusterMetadata metadata = getClusterMetadata(each, instancePublishInfo);
// 调用
ApplicationUtils.getBean(HealthCheckProcessorV2Delegate.class).process(this, each, metadata);
// ...
}
}
} catch (Throwable e) {
} finally {
if (!cancelled) {
// 下一轮调度
HealthCheckReactor.scheduleCheck(this);
if (this.getCheckRtWorst() > 0) {
// ...
}
}
}
}
}
}
备注:定时任务调度,不断向服务的注册节点发送探活请求。
探活处理器选择
备注:由运行时缓存情况可以看出,支持TPC、HTTP、MYSQL三种类型探活处理。
HTTP探活
代码坐标HttpHealthCheckProcessor#process
@Override
public void process(HealthCheckTaskV2 task, Service service, ClusterMetadata metadata) {
HealthCheckInstancePublishInfo instance = (HealthCheckInstancePublishInfo) task.getClient()
.getInstancePublishInfo(service);
if (null == instance) {
return;
}
try {
// TODO handle marked(white list) logic like v1.x.
if (!instance.tryStartCheck()) {
// ...
return;
}
Http healthChecker = (Http) metadata.getHealthChecker();
// 默认使用注册实例端口
int ckPort = metadata.isUseInstancePortForCheck() ? instance.getPort() : metadata.getHealthyCheckPort();
// 组织url请求
URL host = new URL("http://" + instance.getIp() + ":" + ckPort);
URL target = new URL(host, healthChecker.getPath());
Map<String, String> customHeaders = healthChecker.getCustomHeaders();
Header header = Header.newInstance();
header.addAll(customHeaders);
// 发送HTTP请求
ASYNC_REST_TEMPLATE.get(target.toString(), header, Query.EMPTY, String.class,
new HttpHealthCheckCallback(instance, task, service));
MetricsMonitor.getHttpHealthCheckMonitor().incrementAndGet();
} catch (Throwable e) {
instance.setCheckRt(switchDomain.getHttpHealthParams().getMax());
healthCheckCommon.checkFail(task, service, "http:error:" + e.getMessage());
healthCheckCommon.reEvaluateCheckRT(switchDomain.getHttpHealthParams().getMax(), task,
switchDomain.getHttpHealthParams());
}
}
@Override
public void onReceive(RestResult<String> result) {
instance.setCheckRt(System.currentTimeMillis() - startTime);
int httpCode = result.getCode();
// 返回200
if (HttpURLConnection.HTTP_OK == httpCode) {
// 将节点变更为健康
healthCheckCommon.checkOk(task, service, "http:" + httpCode);
healthCheckCommon.reEvaluateCheckRT(System.currentTimeMillis() - startTime, task,
switchDomain.getHttpHealthParams());
} else if (HttpURLConnection.HTTP_UNAVAILABLE == httpCode
|| HttpURLConnection.HTTP_MOVED_TEMP == httpCode) {
} else {
}
}
public void checkOk(HealthCheckTaskV2 task, Service service, String msg) {
// 节点设置为健康状态
healthStatusSynchronizer.instanceHealthStatusChange(true, task.getClient(), service, instance);
}
备注:向节点发起HTTP请求,返回状态码为200表示,将节点标志为健康状态;否则标记非健康状态。
TCP探活
TPC探活的代码详见TcpHealthCheckProcessor,不再详细分析。大体逻辑为通过与注册实例建立channel,不断ping 注册实例的端口是否可用,从而判断服务是否健康。
MYSQL探活
备注:主要检查当前节点为主库,不能访问到从库,可能在主从切换中使用。
总结: 本文就临时节点续约、持久节点心跳、持久节点的探活代码实现做了熟练。相信通过代码走查,对其使用场景和实现不再陌生。