I have a kubernetes service that works in a leader/follower fashion, so only one of 2 pods (for HA) can accept the connection. I want to make my service publicly available with traefik ingress controller.
One of the ways to achieve that - tweaks the readiness probe for the service, so slave pod will fail on readiness checks until it becomes a master. But I don't like this approach, because it will be confusing to distinguish if the pods are indeed failing, or just waiting to become a master?
Any thoughts are welcome
So you don't need to use the k8s readiness probe for the traefik ingress, you can use a traefik backend health check defined in your k8s ingress through annotations. This way you don't forward to backends that are not active. For example:
kind: Ingress
metadata:
name: specific-deployment
annotations:
traefik.backend.healthcheck.port: 8080
traefik.backend.healthcheck.scheme: http
traefik.backend.healthcheck.path: /health
spec:
rules:
- host: specific.minikube
http:
paths:
- path: /
backend:
serviceName: stilton
servicePort: httpThis way you can use the k8s readiness probes for your pods.