Skip to main content

TCP and TLS Edges

ngrok offers TCP and TLS Edges which can be used to provide ingress to TCP or TLS based services. Both are implemented as CRDs and function similarly in broad strokes, albeit with slightly different configuration options offered. Their CRD reference is a useful companion to this guide.

(TLS Only) Get a Domain

At least one hostports must be specified when creating a TLSEdge resource, which takes the form <fqdn>:443. The fully qualified domain name must first be reserved either via the ngrok dashboard or the Domain CRD.

Example:

apiVersion: ingress.k8s.ngrok.com/v1alpha1
kind: Domain
metadata:
name: tlsedgetest-ngrok-app
spec:
domain: tlsedgetest.ngrok.app

Create the Edge

Create the edge CRD. These resources are fairly similar, and both require you to specify a TunnelGroupBackend. This consists of a list of labels that determine which specific Tunnel should receive traffic from the edge. Both may also specify IP Policies for limiting access to the edge. At the time of writing, these policies must be provided as a reference in the form ipp_<id>.

On top of the options available to TCP Edges, TLS Edges support (and require) a few other options:

  • (required) hostports: A list of "<fqdn>:443" strings declaring the list of reserved domains for the edge to listen on.
  • tlsTermination: Configure the TLS Termination behavior. The terminateAt field may be set to upstream to pass the encrypted stream to the Tunnel backend, or edge to terminate the TLS stream at the ngrok edge, and pass plaintext bytes to the Tunnel.
  • mutualTls: Configure client certificate validation at the edge. Requires a reference to a Certificate Authority.

TCP Example:

apiVersion: ingress.k8s.ngrok.com/v1alpha1
kind: TCPEdge
metadata:
name: test-edge
spec:
backend:
labels:
app: tcptestedge

Because TCP Edges don't currently support providing a reserved TCP address. On edge creation, one will be allocated for them, and will be visible by checking the status of the resource:

$ kubectl get tcpedges test-edge
NAME ID HOSTPORTS BACKEND ID AGE
test-edge edgtcp_2Wg5AzVE878vQoNMP3Z8wONIr76 ["7.tcp.ngrok.io:27866"] bkdtg_2Wg5Amjb4GiQoV7SAnpEdM0Dg3n 2m35s

TLS Example:

apiVersion: ingress.k8s.ngrok.com/v1alpha1
kind: TLSEdge
metadata:
name: test-edge
spec:
hostports:
- tlstestedge.ngrok.app:443
backend:
labels:
app: tlstestedge
tlsTermination:
terminateAt: upstream

Start the Tunnel

Finally, create a Tunnel to receive and forward traffic for your edge.

Important fields:

  • forwardsTo: The <hostname>:<port> to forward traffic to. This can be any hostname resolvable and accessible from the ingress controller pod.
  • labels: a map of labels corresponding to the edge to receive traffic for. These must match the labels specified when creating your edge.
  • backend.protocol: The protocol understood by the backend service. TCP will forward connections to the backend as-is, while TLS will create a TLS connection to the backend first, and then forward the connection stream over that.

Example:

apiVersion: ingress.k8s.ngrok.com/v1alpha1
kind: Tunnel
metadata:
name: test-tunnel
spec:
backend:
protocol: TCP
forwardsTo: kubernetes.default.svc:443
labels:
app: tlsedgetest

Full Example

This is an example of using a TLS Edge to expose the kubernetes control plane via ngrok.

---
apiVersion: ingress.k8s.ngrok.com/v1alpha1
kind: Domain
metadata:
name: tlsedgetest-ngrok-app
spec:
# Reserve the tlsedgetest.ngrok.app domain.
domain: tlsedgetest.ngrok.app
---
apiVersion: ingress.k8s.ngrok.com/v1alpha1
kind: TLSEdge
metadata:
name: test-edge
spec:
hostports:
# Listen for connections on the domain we reserved
- tlsedgetest.ngrok.app:443
backend:
labels:
app: tlsedgetest
# Pass the TLS stream on to the backend - let the application do its own TLS
# handshake.
tlsTermination:
terminateAt: upstream
---
apiVersion: ingress.k8s.ngrok.com/v1alpha1
kind: Tunnel
metadata:
name: test-tunnel
spec:
# Forward the raw TCP stream to our backend.
# It will technically contain TLS, and the backend speaks TLS, but we don't
# want the Tunnel to terminate TLS before forwarding incoming connections.
# We don't want a TLS turducken.
backend:
protocol: TCP
# Forward to the kubernetes control plane.
forwardsTo: kubernetes.default.svc:443
# Listen for connections using the labels from our edge.
labels:
app: tlsedgetest

Check the status of your resources:

$ kubectl get domain
NAME ID REGION DOMAIN CNAME TARGET AGE
tlsedgetest-ngrok-app rd_2Wg986lvMqsiB1J5WV5lOcmT21a tlsedgetest.ngrok.app 4s
$ kubectl get tlsedge
NAME ID HOSTPORTS BACKEND ID AGE
test-edge edgtls_2Wg989BMmZLWXixStL8BjAxMcxW ["tlsedgetest.ngrok.app:443"] bkdtg_2Wg981gcSnxaX5cTL28LWwVg4xD 12s
$ kubectl get tunnel
NAME FORWARDSTO AGE
test-tunnel kubernetes.default.svc:443 52m

Our domain and edge both have IDs allocated, so we know they've been created successfully!

Edit your kubeconfig and replace the server with https://tlsedgetest.ngrok.app, comment out certificate-authority-data and add insecure-skip-tls-verify: true to your cluster config. This is needed because kubernetes is completing the TLS handshake with its own certificate, which won't be valid for your ngrok domain.

Use kubectl cluster-info to verify that everything is still working:

$ kubectl cluster-info
Kubernetes control plane is running at https://tlsedgetest.ngrok.app
CoreDNS is running at https://tlsedgetest.ngrok.app/api/v1/namespaces/kube-system/services/kube-dns:dns/proxy

To further debug and diagnose cluster problems, use 'kubectl cluster-info dump'.