Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Kubernetes Nginx Ingress not finding service endpoint

I'm having some trouble getting the Nginx ingress controller working in my Kubernetes cluster. I have created the nginx-ingress deployments, services, roles, etc., according to https://kubernetes.github.io/ingress-nginx/deploy/

I also deployed a simple hello-world app which listens on port 8080

apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: hello-world
  namespace: default
spec:
  selector:
    matchLabels:
      name: hello-world
  template:
    metadata:
      labels:
        name: hello-world
    spec:
      containers:
      - name: hello-world
        image: myrepo/hello-world
        resources:
          requests:
            memory: 200Mi
            cpu: 150m
          limits:
            cpu: 300m
        ports:
          - name: http
            containerPort: 8080
            protocol: TCP

And created a service for it

kind: Service
apiVersion: v1
metadata:
  namespace: default
  name: hello-world
spec:
  selector:
    app: hello-world
  ports:
    - name: server
      port: 8080

Finally, I created a TLS secret (my-tls-secret) and deployed the nginx ingress per the instructions. For example:

apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  annotations:
    kubernetes.io/ingress.class: nginx
  name: hello-world
  namespace: default
spec:
  rules:
    - host: hello-world.mydomain.com
      http:
        paths:
        - path: /
          backend:
            serviceName: hello-world
            servicePort: server
  tls:
      - hosts:
          - hello-world.mydomain.com
        secretName: my-tls-cert

However, I am unable to ever reach my application, and in the logs I see

W0103 19:11:15.712062       6 controller.go:826] Service "default/hello-world" does not have any active Endpoint.
I0103 19:11:15.712254       6 controller.go:172] Configuration changes detected, backend reload required.
I0103 19:11:15.864774       6 controller.go:190] Backend successfully reloaded.

I am not sure why it says Service "default/hello-world" does not have any active Endpoint. I have used a similar service definition for the traefik ingress controller without any issues.

I'm hoping I'm missing something obvious with the nginx ingress. Any help you can provide would be appreciated!

like image 652
cookandy Avatar asked Jan 03 '19 19:01

cookandy


4 Answers

I discovered what I was doing wrong. In my application definition I was using name as my selector

  selector:
    matchLabels:
      name: hello-world
  template:
    metadata:
      labels:
        name: hello-world

Whereas in my service I was using app

  selector:
    app: hello-world

After updating my service to use app, it worked

  selector:
    matchLabels:
      app: hello-world
  template:
    metadata:
      labels:
        app: hello-world
like image 80
cookandy Avatar answered Nov 12 '22 16:11

cookandy


Another situation when it may happen is when ingress class of the ingress controller does not match ingress class in the ingress resource manifest used for your services.

Nginx installation command, short example:

  helm install stable/nginx-ingress \
  --name ${INGRESS_RELEASE_NAME} \
  --namespace ${K8S_NAMESPACE} \
  --set controller.scope.enabled=true \
  --set controller.scope.namespace=${K8S_NAMESPACE} \
  --set controller.ingressClass=${NGINX_INGRESS_CLASS}

ingress resource spec. , excerpt:

apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  labels:
  annotations:
    # folowing line is not valid for K8s or Helm, 
    # but reflects the values must be the same
    kubernetes.io/ingress.class: ${NGINX_INGRESS_CLASS}
like image 28
Rostislav Matl Avatar answered Nov 12 '22 18:11

Rostislav Matl


In our case, this was caused by having the ingress resource definition on a different namespace then the services.

kind: Ingress
apiVersion: networking.k8s.io/v1beta1
metadata:
  name: nginx-ingress-rules
  namespace: **default**       #<= make sure this is the same value like the namespace on the services you are trying to reach
like image 2
user2995891 Avatar answered Nov 12 '22 16:11

user2995891


In my case, I included an "id" directive in my Service selector that was missing from the Deployment metadata and this prevented the endpoints controller from finding the correct Pod. I expect the reverse case would also fail:

---
apiVersion: v1
kind: Service
metadata:
  name: some-service
spec:
  ports:
    - name: port-name
      port: 1234
      protocol: TCP
  selector:
    app: some-app
    id: "0"  ## include in both or neither
like image 1
kerlyn Avatar answered Nov 12 '22 16:11

kerlyn