DigitalOcean Kubernetes负载均衡器多部署共存流量异常求助

我在DigitalOcean Kubernetes集群部署Java/Spring Boot微服务，同时部署Eureka Server和Google hello-app测试环境时遇到问题：配置两个Deployment和一个LoadBalancer Service后，同时部署两者时LoadBalancer流量状态显示0/1，无法连接Droplet；删除任意一个Deployment再重新部署，LoadBalancer即可正常工作。

部署配置 #

1. hello-deployment-1 #

apiVersion: apps/v1
kind: Deployment
metadata:
  name: hello-deployment-1
spec:
  replicas: 1
  selector:
    matchLabels:
      app: hello-deployment-1
      role: metrol-network-service
  template:
    metadata:
      labels:
        app: hello-deployment-1
        role: metrol-network-service
    spec:
      containers:
      - name: hello
        image: gcr.io/google-samples/hello-app:1.0
        ports:
        - containerPort: 8080
        resources:
          requests:
            memory: "400Mi"
            cpu: "200m"
          limits:
            memory: "500Mi"
            cpu: "250m"
        readinessProbe:
          httpGet:
            path: /
            port: 8080
          initialDelaySeconds: 60
          periodSeconds: 10
        livenessProbe:
          tcpSocket:
            port: 8080
          initialDelaySeconds: 20
          periodSeconds: 10

2. metrol-eureka-server-deployment #

apiVersion: apps/v1
kind: Deployment
metadata:
  name: metrol-eureka-server-deployment
spec:
  replicas: 1
  selector:
    matchLabels:
      app: metrol-eureka-server
      role: metrol-network-service
  template:
    metadata:
      labels:
        app: metrol-eureka-server
        role: metrol-network-service
    spec:
      containers:
      - name: 'metrol-eureka-server'
        image: crivera3/metrol-eureka-server:0.0.2
        ports:
        - containerPort: 8761
        resources:
          requests:
            memory: "1000Mi"
            cpu: "500m"
          limits:
            memory: "1100Mi"
            cpu: "600m"
        readinessProbe:
          httpGet:
            path: /actuator/health
            port: 8761
          initialDelaySeconds: 30
          periodSeconds: 10
        livenessProbe:
          tcpSocket:
            port: 8761
          initialDelaySeconds: 20
          periodSeconds: 10

LoadBalancer Service配置 #

apiVersion: v1
kind: Service
metadata:
  name: metrol-network-service
  annotations:
    service.beta.kubernetes.io/do-loadbalancer-healthcheck-protocol: "http"
    service.beta.kubernetes.io/do-loadbalancer-healthcheck-path: "/actuator/health"
    service.beta.kubernetes.io/do-loadbalancer-healthcheck-port: "8761"
spec:
  externalTrafficPolicy: Local
  type: LoadBalancer
  ports:
  - port: 8761
    targetPort: 8761
    name: metrol-eureka
  - port: 8080
    targetPort: 8080
    name: hello-deployment-1
  selector:
    role: metrol-network-service

节点资源使用情况 #

同时部署两个Deployment时 #

单独部署一个Deployment时 #

故障排查与解决方案 #

核心问题分析 #

你的LoadBalancer配置了针对8761端口的/actuator/health路径的健康检查，但Service的selector匹配了所有带有role: metrol-network-service标签的Pod——包括hello-app的Pod，而hello-app的Pod并没有暴露8761端口和/actuator/health端点。当两个Deployment同时运行时，DigitalOcean的负载均衡器会尝试对所有匹配到的Pod执行健康检查，hello-app的Pod无法通过检查，导致负载均衡器认为节点不可用（0/1状态）。

具体修复步骤 #

1. 拆分Service，为每个Deployment单独配置Service

不要让单个LoadBalancer同时代理两个不同的服务，而是分别创建两个LoadBalancer Service，各自匹配对应的Pod标签：

针对Eureka Server的Service：

apiVersion: v1
kind: Service
metadata:
  name: metrol-eureka-service
  annotations:
    service.beta.kubernetes.io/do-loadbalancer-healthcheck-protocol: "http"
    service.beta.kubernetes.io/do-loadbalancer-healthcheck-path: "/actuator/health"
    service.beta.kubernetes.io/do-loadbalancer-healthcheck-port: "8761"
spec:
  externalTrafficPolicy: Local
  type: LoadBalancer
  ports:
  - port: 8761
    targetPort: 8761
    name: metrol-eureka
  selector:
    app: metrol-eureka-server

针对hello-app的Service：

apiVersion: v1
kind: Service
metadata:
  name: hello-app-service
  annotations:
    service.beta.kubernetes.io/do-loadbalancer-healthcheck-protocol: "http"
    service.beta.kubernetes.io/do-loadbalancer-healthcheck-path: "/"
    service.beta.kubernetes.io/do-loadbalancer-healthcheck-port: "8080"
spec:
  externalTrafficPolicy: Local
  type: LoadBalancer
  ports:
  - port: 8080
    targetPort: 8080
    name: hello-app
  selector:
    app: hello-deployment-1

这样每个负载均衡器只会检查对应服务的Pod，避免健康检查不匹配的问题。

2. 调整资源配置（可选）

从资源使用数据看，同时部署时CPU请求已达节点的79%，内存请求达77%，接近节点资源上限。如果节点资源不足，可能导致Pod启动缓慢或健康检查超时，可考虑：

• 升级DigitalOcean Droplet的节点规格，增加CPU和内存资源
• 优化Pod的资源请求/限制值，避免资源过度占用

3. 验证健康检查配置

确保每个Service的健康检查配置与对应Pod的端点完全匹配：

• Eureka Server的健康检查路径是/actuator/health，端口8761，对应Pod的readinessProbe配置
• hello-app的健康检查路径是/，端口8080，对应Pod的readinessProbe配置

验证方法 #

1. 删除原有的metrol-network-service
2. 应用上述两个新的Service配置
3. 同时部署两个Deployment
4. 查看LoadBalancer的状态，确认流量状态变为1/1（或对应节点数），并测试两个服务的外部访问是否正常

内容的提问来源于stack exchange，提问作者Camilo Rivera