inlets-operator

Get a Kubernetes LoadBalancer where you never thought it was possible.

In cloud-based Kubernetes solutions, Services can be exposed as type "LoadBalancer" and your cloud provider will provision a LoadBalancer and start routing traffic, in another word: you get ingress to your service.

inlets-operator brings that same experience to your local Kubernetes or k3s cluster (k3s/k3d/minikube/microk8s/Docker Desktop/KinD). The operator automates the creation of an inlets exit-node on public cloud, and runs the client as a Pod inside your cluster. Your Kubernetes Service will be updated with the public IP of the exit-node and you can start receiving incoming traffic immediately.

Who is this for?

This solution is for users who want to gain incoming network access (ingress) to their private Kubernetes clusters running on their laptops, VMs, within a Docker container, on-premises, or behind NAT. The cost of the LoadBalancer with a IaaS like DigitalOcean is around 5 USD / mo, which is 10 USD cheaper than an AWS ELB or GCP LoadBalancer.

Whilst 5 USD is cheaper than a "Cloud Load Balancer", this tool is for users who cannot get incoming connections due to their network configuration, not for saving money vs. public cloud.

Status and backlog

The inlets-operator automates cloud host provisioning to run inlets or inlets-pro to expose internal services to the Internet.

Backlog completed:

• Provision VMs/exit-nodes on public cloud
• Provision to Packet.com
• Provision to DigitalOcean
• Provision to Scaleway
• Provision to GCP
• Automatically update Service type LoadBalancer with a public IP
• Tunnel L7 http traffic
• In-cluster Role, Dockerfile and YAML files
• Raspberry Pi / armhf build and YAML file
• ARM64 (Graviton/Odroid/Packet.com) Dockerfile/build and K8s YAML files
• Ignore Services with dev.inlets.manage: false annotation
• Garbage collect hosts when Service or CRD is deleted
• CI with Travis (use openfaas-incubator/openfaas-operator as a sample)
• Automate inlets-pro for TCP traffic

Backlog pending:

• Automate wss:// for control-port using self-signed certs or LetsEncrypt and nip.io
• Move control-port and /tunnel endpoint to high port i.e. 31111 and make it configurable in the helm chart
• Provision to AWS EC2
• Provision to Civo

Inlets tunnels HTTP traffic at L7, so the inlets-operator can be used to tunnel HTTP traffic. A new project I'm working on called inlets-pro tunnels any TCP traffic at L4 i.e. Mongo, Redis, NATS, SSH, TLS, whatever you like.

Related projects

Inlets is listed on the Cloud Native Landscape as a Service Proxy

• inlets - open-source L7 HTTP tunnel and reverse proxy
• inlets-pro - L4 TCP load-balancer
• inlets-operator - deep integration for inlets in Kubernetes, expose Service type LoadBalancer
• inletsctl - CLI tool to provision exit-nodes for use with inlets or inlets-pro

Author

inlets and inlets-operator are brought to you by Alex Ellis. Alex is a CNCF Ambassador and the founder of OpenFaaS.

Note: inlets is made available free-of-charge, but you can support its ongoing development through GitHub Sponsors 💪

Video demo

This video demo shows a single-node VM running on k3s on Packet.com, and the inlets exit node also being provisioned on Packet's infrastructure.

See an alternative video showing my cluster running with KinD on my Mac and the exit node being provisioned on DigitalOcean:

• KinD & DigitalOcean.

Step-by-step tutorial

Try the step-by-step tutorial

Running in-cluster, using DigitalOcean for the exit node

Note: this example is now multi-arch, so it's valid for x86_64, ARMHF, and ARM64.

You can also run the operator in-cluster, a ClusterRole is used since Services can be created in any namespace, and may need a tunnel.

# Create a secret to store the access token

kubectl create secret generic inlets-access-key \
  --from-literal inlets-access-key="$(cat ~/Downloads/do-access-token)"

kubectl apply -f ./artifacts/crd.yaml

# Apply the operator deployment and RBAC role
kubectl apply -f ./artifacts/operator-rbac.yaml
kubectl apply -f ./artifacts/operator.yaml

Running in-cluster, using Google Compute Engine for the exit node using helm

Note: this example is now multi-arch, so it's valid for x86_64, ARMHF, and ARM64.

If you do not have helm installed and configured follow the instructions here

It is assumed that you have gcloud installed and configured on your machine. If not, then follow the instructions here

# Get current projectID
export PROJECTID=$(gcloud config get-value core/project 2>/dev/null)

# Create a service account
export SERVICEACCOUNT=$(gcloud iam service-accounts list | grep inlets | awk '{print $2}')

# Assign appropriate roles to inlets service account
gcloud projects add-iam-policy-binding $PROJECTID \
--member serviceAccount:$SERVICEACCOUNT \
--role roles/compute.admin

gcloud projects add-iam-policy-binding $PROJECTID \
--member serviceAccount:$SERVICEACCOUNT \
--role roles/iam.serviceAccountUser

# Create inlets service account key file
gcloud iam service-accounts keys create key.json \
--iam-account $SERVICEACCOUNT

# Create a secret to store the service account key file
kubectl create secret generic inlets-access-key --from-file=inlets-access-key=key.json

# Add and update the inlets-operator helm repo
helm repo add inlets https://inlets.github.io/inlets-operator/

helm repo update

# Install inlets-operator with the required fields
helm upgrade inlets-operator --install inlets/inlets-operator \
  --set provider=gce,zone=us-central1-a,gceProjectId=$PROJECTID

Running on a Raspberry Pi

Use the same commands as described in the section above.

There used to be separate deployment files in artifacts folder called operator-amd64.yaml and operator-armhf.yaml. Since version 0.2.7 Docker images get built for multiple architectures with the same tag which means that there is now just one deployment file called operator.yaml that can be used on all supported architecures.

Get a LoadBalancer provided by inlets

kubectl run nginx-1 --image=nginx --port=80 --restart=Always
kubectl run nginx-2 --image=nginx --port=80 --restart=Always

kubectl expose deployment nginx-1 --port=80 --type=LoadBalancer
kubectl expose deployment nginx-2 --port=80 --type=LoadBalancer

kubectl get svc

kubectl get tunnel/nginx-1-tunnel -o yaml

kubectl logs deploy/nginx-1-tunnel-client

Check the IP of the LoadBalancer and then access it via the Internet.

Example with OpenFaaS, make sure you give the port a name of http, otherwise a default of 80 will be used incorrectly.

apiVersion: v1
kind: Service
metadata:
  name: gateway
  namespace: openfaas
  labels:
    app: gateway
spec:
  ports:
    - name: http
      port: 8080
      protocol: TCP
      targetPort: 8080
      nodePort: 31112
  selector:
    app: gateway
  type: LoadBalancer

Annotations

By default the operator will create a tunnel for every loadbalancer service.

To ignore a service such as traefik type in: kubectl annotate svc/traefik -n kube-system dev.inlets.manage=false

You can also set the operator to ignore the services by default and only manage them when the annotation is true. dev.inlets.manage=true To do this, run the operator with the flag -annotated-only

Monitor/view logs

The operator deployment is in the kube-system namespace.

kubectl logs deploy/inlets-operator -n kube-system -f

Provider Pricing

Provider	Price per month	Price per hour	OS image	CPU	Memory	Boot time
Google Compute Engine	* ~$4.28	~$0.006	Debian GNU Linux 9 (stretch)	1	614MB	~3-15s
Packet	~$51	$0.07	Ubuntu 16.04	4	8GB	~45-60s
Digital Ocean	$5	~$0.0068	Ubuntu 16.04	1	512MB	~20-30s
Scaleway	2.99€	0.006€	Ubuntu 18.04	2	2GB	3-5m

* The first f1-micro instance in a GCP Project (the default instance type for inlets-operator) is free for 720hrs(30 days) a month

Contributing

Contributions are welcome, see the CONTRIBUTING.md guide.

Similar projects / products and alternatives

• metallb - open source LoadBalancer for private Kubernetes clusters, no tunnelling.
• inlets - inlets provides an L7 HTTP tunnel for applications through the use of an exit node, it is used by the inlets operator
• inlets pro - L4 TCP tunnel, which can tunnel any TCP traffic and is on the roadmap for the inlets-operator
• Cloudflare Argo - paid SaaS product from Cloudflare for Cloudflare customers and domains - K8s integration available through Ingress
• ngrok - a popular tunnelling tool, restarts every 7 hours, limits connections per minute, paid SaaS product with no K8s integration available