How did kubernetes nginx ingress choose the right pods to balance without specific configurations?

Question

Architecture

• I have a namespace in my multi-node cluster in Google Kubernetes Engine (GKE), let's call it project-namespace
• In the project-namespace there is a single frontend application (replicated), and some backend applications (each one replicated too)
  • frontend: fe1 -> fe1-r1, fe1-r2, fe1-r3
  • backend1: be1 -> be1-r1, be1-r2, be1-r3
  • backend2: be2 -> be2-r1, be2-r2, be2-r3
  • backendN: ...
• be1 is a an exception, it needs to be accessible from the outside so i exposed it via the google cloud default load balancer:
  • GKE-BL-be1 -> be1-r1, be1-r2, be1-r3
• Every other application, including fe1 is backed by ClusterIP services

Objective

I exposed the fe1 via nginx ingress instead of the GKE default ingress

What i did

• I downloaded this operator file in the "getting started" documentation of nginx-ingress:

curl https://raw.githubusercontent.com/kubernetes/ingress-nginx/controller-v0.43.0/deploy/static/provider/cloud/deploy.yaml \
    > ingress.yaml

• I replaced the namespace used by the operator

sed -i 's/namespace: ingress-nginx/namespace: project-namespace/g'

• I deployed the operator file

kubectl apply -f ingress.yaml

• This created all the resources and the nginx ingress load balancer, making available port 80 and 443 from outside the cluster without k8s labels used by my other application pods.

What happened

From the browser i could access the fe1 using the external ip of the nginx ingress, and from the fe1 pod logs i saw that the requests were actually served by those pods. I scaled down to 0 replicas the fe1 pods and no one responded anymore to my browser requests.

So actually the objective is achieved but i had no control on it, i didn't configure anything in the nginx ingress, i only used the operator file. Moreover the gui is exposed only through ClusterIP and i didn't configure any proxying rules between the nginx ingress and the fe1 ClusterIP.

Why is this working? Which configuration am i missing?

EDIT:

After some troubleshooting i found out that in the "Ingress" section i already deployed a GKE Ingress pointing to the gui pods, the issue is that the nginx ingress service took the same ip address as the GKE ingress, so actually my browser was sending request to the old ingress. Now my doubt is: how is that possible that the new nginx ingress controller took the same ip of the GKE old one?

Answer 1

In order to deploy a NGINX Ingress Controller on Google Kubernetes Engine, you may find useful the following link. Bear in mind this:

The default configuration watches Ingress object from all namespaces. To change this behavior use the flag --watch-namespace to limit the scope to a particular namespace.