Add secure ingress to the Kubernetes API and access it anywhere

This blog post was initially published on the Cloud Native Computing Foundation blog—thanks to their team for featuring ngrok and the Kubernetes Operator!

A long while back, one of our customers asked if they could use ngrok to securely access the Kubernetes API of their clusters from anywhere instead of using VPC tunneling or VPNs.

The answer—an emphatic *yes`—recently got a lot easier with some new features in our Kubernetes Operator, which will let you:

1. Add ingress to your K8s API with a single CRD
2. Still use kubectl for administrative tasks, not compromise with kube proxy and curl
3. Use your existing RBAC/AuthN/AuthZ
4. Add observability and traffic management without deploying even more services to your cluster

You’ll start by setting up token-based RBAC (if you don’t already have that), then install the K8s Operator, apply an AgentEndpoint resource, and then taking your secure usage of the Kubernetes API to the next level.

Before you jump in: RBACCopy link to clipboard

I assume you already have a K8s cluster you’d like to access remotely. If not, or if you want to test this functionality out in a proof-of-concept fashion, minikube is always a great place to start.

Speaking of authentication, proper authentication (AuthN) and authorization (AuthZ) are also necessary to access the Kubernetes API remotely from both functional and security best practices perspectives. The default RBAC configuration for most K8s clusters won’t allow you to access API resources externally, as it assigns you the system:anonymous role, which lacks the privileges to even view or list the state of your cluster.

One way to bootstrap RBAC for remote access is a service account with the appropriate authorization.

kubectl create serviceaccount ngrok

Dive your new service account some privileges with a ClusterRoleBinding. For the sake of simplicity and setting up ngrok to its fullest, an admin ClusterRole is helpful—if you’re setting this up for others on your team, you may want to use edit or view.

kubectl create rolebinding ngrok-admin-binding --clusterrole=admin --serviceaccount=default:ngrok --namespace=default

Update your Kubectl config file with this new service account user and a token.

kubectl config set-credentials ngrok --token $(kubectl create token ngrok)

This isn’t the be-all-end-all for secure AuthN/AuthZ for your Kubernetes cluster, but it will get you through the basic setup.

Set up the ngrok Kubernetes OperatorCopy link to clipboard

Head over to ngrok and create an account for free if you don’t have one already, then reserve a domain—type in any string you’d like, then choose a TLD from the dropdown, like .ngrok.app. Keep note of this—I’ll refer to it as <NGROK_DOMAIN> later. ngrok also generates certificates and terminates TLS for you at the agent, which means one less thing to worry about.

Now you can install ngrok’s open-source Kubernetes Operator, which drops the agent into your cluster to provision ingress, observability, and traffic management with a single CRD and native tooling.

Set up a few environment variables to make using Helm a little easier—grab your ngrok authtoken in the dashboard, then create a new API key for this deployment, and drop those into environment variables.

export NGROK_AUTHTOKEN=<NGROK_AUTHTOKEN>
export NGROK_API_KEY=<NGROK_API_KEY>

Use Helm to install the ngrok Kubernetes Operator into the ngrok-operator namespace.

helm install ngrok-operator ngrok/ngrok-operator \
  --namespace ngrok-operator \
  --create-namespace \
  --set credentials.apiKey=$NGROK_API_KEY \
  --set credentials.authtoken=$NGROK_AUTHTOKEN

Give your cluster a few moments, then double-check you have both the ngrok-operator-manager and ngrok-operator-agent pods running and ready.

kubectl get pods -n ngrok-operator

NAME                                      READY     STATUS    RESTARTS   AGE
ngrok-operator-agent-576468c9f7-2crdm     1/1  Running   0        29s
ngrok-operator-manager-6d8bc67559-82zwt   1/1  Running   0        29s

Expose your cluster’s K8s API with ngrokCopy link to clipboard

Annotate the kubernetes service in the default namespace to declare the backend is an HTTPS upstream.

kubectl -n default annotate service kubernetes 'k8s.ngrok.com/app-protocols={"https":"https"}'

Create a file called expose-k8s-api.yaml with the below YAML, replacing <NGROK_DOMAIN> with the one you reserved earlier. This is our new AgentEndpoint custom resource, which you can use directly to create ingress into any Kubernetes service—even the API itself—without having to roll out a bunch of complex networking.

apiVersion: ngrok.k8s.ngrok.com/v1alpha1
kind: AgentEndpoint
metadata:
  namespace: default
  name: kubernetes-api-endpoint
spec:
  url: https://<NGROK_DOMAIN>
  upstream:
    url: http://kubernetes:443

Apply this CRD to your cluster.

kubectl apply -f expose-k8s-api.yaml

Edit your config to connect kubectl to your external clusterCopy link to clipboard

Your K8s API is technically ready for you to access remotely. If you curl the domain you reserved before, you’ll hit your cluster’s API—but won’t be allowed in thanks to Kubernetes’ built-in RBAC.

curl https://<NGROK_DOMAIN>/api/v1

{
	"kind": "Status",
	"apiVersion": "v1",
	"metadata": {},
	"status": "Failure",
	"message": "forbidden: User \"system:anonymous\" cannot get path \"/api/v1\"",
	"reason": "Forbidden",
	"details": {},
	"code": 403
}

Also, you probably would rather just continue to use kubectl to view your cluster’s resources.

Make a backup of your existing config for safekeeping.

cp ~/.kube/config ~/.kube/config.bak

Use kubectl to update the cluster’s server property.

kubectl config set-cluster <CLUSTER_NAME> --server https://<NGROK_DOMAIN>

And to remove the certificate-authority property from your cluster config, if that exists, as you’ll use your token instead.

kubectl config unset clusters.<CLUSTER_NAME>.certificate-authority

Create a new context that uses your service account and token with the appropriate privileges based on the cluster role you set before, then switch to it.

kubectl config set-context ngrok --cluster <YOUR_CLUSTER_NAME> --user ngrok --namespace=default && kubectl config use-context ngrok

Verify that you can access your cluster’s API remotely with kubectl get pods.

You’re set! But there is still plenty more functionality to get the most from external API access to your clusters.

View K8s API requests with Traffic Inspector observabilityCopy link to clipboard

As I mentioned earlier, kubectl is a wrapper around the Kubernetes API, which means that when you run a command like kubectl get pods, you’re actually generating an HTTP request to https://<NGROK_DOMAIN>/api/v1/namespaces/default/pods?limit=500. Since ngrok handles ingress through that public endpoint, you can also observe every API request through Traffic Inspector.

Here’s an example of what kubectl get pods actually looks like on the request level:

I’ve anonymized some of the details, but the request headers and response body contain valuable details about who is accessing your K8s APIs, using which user roles or service accounts and associated tokens, what actions they’re trying to perform, and how your cluster responds.

Traffic Inspector is built into every ngrok endpoint and doesn’t require you to deploy any other services to your cluster, like Prometheus, which then needs additional plumbing and AuthN protection.

Set access control with Traffic PolicyCopy link to clipboard

With this setup, you rely on ngrok for discoverability and ingress, and K8s’ RBAC for access control into your cluster, but you can also take other steps to control who is allowed to even `attempt* to fire off a request to your cluster.

ngrok's Traffic Policy engine lets you filter, orchestrate, and take action on traffic as it passes through the network. If you connect to these clusters from static IPs, you could restrict ingress to only those IPs with the restrict-ips action—and reject all other requests before they even hit your cluster.

Your expose-k8s-api.yaml file gets a new entry—trafficPolicy—and some extra config to only allow requests from the 1.1.1.1 IP. You can use CIDRs and multiple IPs, in IPv4 and IPv6, to allow just the right access.

apiVersion: ngrok.k8s.ngrok.com/v1alpha1
kind: AgentEndpoint
metadata:
  namespace: default
  name: kubernetes-api-endpoint
spec:
  url: https://<NGROK_DOMAIN>
  upstream:
    url: https://kubernetes.default:443
  trafficPolicy:
    inline:
      on_tcp_connect:
        - actions:
            - type: restrict-ips
              config:
                enforce: true
                allow:
                  - 1.1.1.1/32

Reapply the CRD with kubectl apply -f expose-k8s-api.yaml and you’re set.

What’s next?Copy link to clipboard

With ngrok and the Kubernetes Operator as your “pipe” between kubectl and the Kubernetes API, you can control your clusters independent of where you are or what networks you must traverse—while also adding additional security layers that you simply can’t get with the built-in options. This route is also more scalable, as you can deploy the same manifest multiple times, replacing only hostnames, to quickly add ingress to the K8s API across all your clusters.

All these features work on any plan—you only pay by usage for resources the Operator creates on your behalf—so sign up for a free plan today and get your Kubernetes ingress covered... even if that's to the Kubernetes API itself.

Looking for docs? We have you covered:

• Kubernetes Operator
• Traffic Policy

Ready to give feedback or file a bug report? Check out the ngrok/ngrok-operator project on GitHub.