Understanding multiple Ingress in AKS

Last time, I covered the ins and outs of Ingress on AKS.

We looked at how to install nginx ingress controller and how the controller is deployed as a load balanced service. We did some URL based routing and domain name overloading.

This was in the spirit of clarifying the magic behind Ingress Controllers.

Another aspect of Ingress documentation I find could use more clarity is multiple ingress controllers.

There are many reasons why you might need to have multiple ingress controllers:

  • You need an internet facing one and a private network one
  • You need different controller implementations: for some workloads you need the NGinx controller while for others you need Traefik (for example)
  • Your cluster is used by many line of businesses and is segmented by namespace ; each line of business needs its own controller
  • For whatever reason, you do not want to have a single public / private IPs for all your workloads

The most documented case is the first one. I’ll cover that in a future article. Instead here, I want to be more general.

It’s quite simple actually. Once you understand how to work with multiple Ingress Controllers, the different scenarios will be a breeze to implement.

Scripts used in this article are on GitHub for convenience.


Our understanding of Ingress Controller is the following:

Conceptual Model

The most important part is the top. An Ingress Controller can have multiple Ingress rules. But there could be multiple Ingress Controller.

The way to attach a rule to a controller is with the name of the controller. Now for whatever reason the name of a controller is specified by its class.

That is the punchline.

That is what makes it a little confusing.

Let’s experiment a bit.

Installing controllers

We assume the setup we covered in our last article, i.e. the cluster and Helm is installed.

If we look at the node resource group, we should have:

Before multiple controllers

Here we have one public IP because we installed NGinx controller with a default configuration. If we didn’t we wouldn’t have one.

We know we need to distinguish them by class, so let’s look for that parameter in the chart:

$ helm inspect stable/nginx-ingress | grep class
  ## Name of the ingress class to route through this controller
To use, add the `kubernetes.io/ingress.class: nginx` annotation to your Ingress resources.
`controller.ingressClass` | name of the ingress class to route through this controller | `nginx`

So controller.ingressClass is the culprit.

Let’s install 3 controllers with classes first, second & third:

helm install stable/nginx-ingress --set controller.ingressClass=first --namespace kube-system --set controller.replicaCount=2 --set rbac.create=false
helm install stable/nginx-ingress --set controller.ingressClass=second --namespace kube-system --set controller.replicaCount=2 --set rbac.create=false
helm install stable/nginx-ingress --set controller.ingressClass=third --namespace kube-system --set controller.replicaCount=2 --set rbac.create=false

Deploying services

Let’s deploy a few services using the ingress controllers.

We’ll use the spec file multiple-controller.yaml:

kubectl apply -f multiple-controller.yaml

The file contains three (3) deployments of two (2) pods each: blue, white & red. For instance, the blue one:

# Blue deployment
apiVersion: apps/v1
kind: Deployment
  name: blue-deploy
  replicas: 2
      app:  blue
        app: blue
      - name: myapp
        image: vplauzon/get-started:part2-no-redis
        - name:  NAME
          value: Blue-Pod
        - containerPort: 80

The pod simply displays Blue Pod.

For each deployment we deploy a service exposing the pods in ClusterIP. For instance, the blue one:

# Blue Service
apiVersion: v1
kind: Service
  name: blue-svc
  type: ClusterIP
  - port: 80
    app: blue

Finally, for each ingress controller, we deploy an ingress. We expose the three services (blue, white & red) with each controller. A service isn’t dedicated to an ingress and we use that fact here.

Here is the ingress for the first controller. Notice the annotation kubernetes.io/ingress.class: this is the mapping between the ingress and the ingress controller:

# First Ingress
apiVersion: extensions/v1beta1
kind: Ingress
  name: first-ingress
    kubernetes.io/ingress.class: first
    nginx.ingress.kubernetes.io/rewrite-target: /
  - http:
      - path: /1-blue
          serviceName: blue-svc
          servicePort: 80
      - path: /1-white
          serviceName: white-svc
          servicePort: 80
      - path: /1-red
          serviceName: red-svc
          servicePort: 80


Before testing our configuration, we need to map the ingress controllers to public IPs.

If we look at the node resource group, we now find 3 new public ips:

After multiple controllers

One for each controller.

Public IPs were provisioned when we installed the Ingress Controllers. We could also provision them in advance and inject them by using set controller.service.loadBalancerIP="X.Y.Z.W". This is explained in the online documentation.

Since we didn’t provision the public IPs we need to do some queries to map them:

$ kubectl get svc --namespace kube-system -l component=controller
NAME                                           TYPE           CLUSTER-IP     EXTERNAL-IP       PORT(S)                      AGE
icy-chimp-nginx-ingress-controller             LoadBalancer   80:31611/TCP,443:31031/TCP   58m
illmannered-dolphin-nginx-ingress-controller   LoadBalancer    80:30482/TCP,443:30339/TCP   57m
rude-unicorn-nginx-ingress-controller          LoadBalancer    80:31333/TCP,443:30813/TCP   57m
stultified-puffin-nginx-ingress-controller     LoadBalancer     80:30737/TCP,443:32580/TCP   2d9h

This query gives use the mapping between Nginx Helm release name (e.g. icy-chimp) and the external IP.

Now we can see what each release is about:

$ helm get values icy-chimp
  ingressClass: first
  replicaCount: 2
  create: false

So we found out that, in our case, the first ingress controller corresponds to Helm release icy-chimp and has an external IP of We can now easily test it:

$ curl
<h3>Hello Blue-Pod!</h3><b>Hostname:</b> blue-deploy-64fd54844c-q7nhk<br/><b>Visits:</b> undefined

We can similarly test the other routes and other ingress.


We’ve seen it is quite simple to use multiple ingress controllers.

Each Ingress Controllers correspond to an Helm deployment. We can set the class of a controller using the *–set controller.ingressClass=* option. A public (or private) IP is associated with each controller.

We map an ingress with an ingress controller using the kubernetes.io/ingress.class annotation.

And that’s it! We can have as many ingress controllers as we want. We could deploy them to different namespaces if need be, by using the namespace override of Helm.

I hope this clarifies how to work with multiple ingress controllers!


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s