0

We are have enabled horizontal pod autoscaling in GKE, our pods are sitting behind a clusterIP type service and we are routing public traffic to that Service using NGINX Ingress controller. When monitoring the usages we have noticed that traffic is not equally distributed between pods. it's routing traffic to one single pod. but whenever we manually deleted that particular pod it will route traffic to another available pod.

Is there any way we can enable ingress rules to distribute traffic equally

Ingress

apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  annotations:
    kubernetes.io/ingress.class: nginx
    kubernetes.io/tls-acme: "true"
    nginx.ingress.kubernetes.io/load-balance: round_robin
    nginx.ingress.kubernetes.io/ssl-passthrough: "true"
    nginx.ingress.kubernetes.io/ssl-redirect: "false"
    nginx.org/proxy-connect-timeout: 30s
    nginx.org/proxy-read-timeout: 20s
  generation: 11
  name: test-ingress
  namespace: default
spec:
  rules:
  - host: gateway.example.com
    http:
      paths:
      - backend:
          serviceName: gateway-443
          servicePort: 443
        path: /
      - backend:
          serviceName: gateway-80
          servicePort: 80
        path: /

Service manifest

apiVersion: v1
kind: Service
metadata:
  annotations:
    cloud.google.com/neg: '{"ingress":true}'
    serviceloadbalancer/lb.cookie-sticky-session: "false"
    serviceloadbalancer/lb.host: gateway.example.com
    serviceloadbalancer/lb.sslTerm: "true"
  labels:
    name: gateway-default
    port: gateway-default-8243
  name: gateway-8243
  namespace: default
spec:
  clusterIP: 10.20.215.122
  ports:
  - name: pass-through-https
    port: 443
    protocol: TCP
    targetPort: 8243
  selector:
    name: gatway-default
  sessionAffinity: ClientIP
  type: ClusterIP
Samith Perera
  • 83
  • 9

2 Answers2

1

Your Ingress should have a serviceName which in your case is "gateway-443" and "gateway-80" but the actual name specified in the Service in metadata.name is "gateway-8243".

(If this is on purpose, please post the YAML of the other resources so I can take a look at the whole setup.)

Also please take a look at this page that has lots of good examples on how to achieve what you are looking to do.

Vi Pau
  • 337
  • 1
  • 4
  • Thank you for your reply and sorry I made a mistake when copying the ingress rule, the service name should be gateway-8243. – Samith Perera May 21 '21 at 12:56
1

Finally, I have figure out the issue, reason is setting sessionAffinity: ClientIP,

Only If you want to make sure that connections from a particular client are passing to the same Pod each time, you should set the session affinity to ClientsIP it will route traffic based on the Client's IP addresses. To equally distribute traffic between pods you can use the value "None" or remove this sessionAffinity parameter because the default value is none.

References,

https://kubernetes.io/docs/concepts/services-networking/service/#proxy-mode-userspace

How to use Session Affinity on requests to Kubernetes service?

Samith Perera
  • 83
  • 9