Kubernetes on Edge, without the platform tax

Deploy

A six-node k3s cluster in five steps

Two custom startup scripts do the install. Attach every VM to one private network from the Networking tab; k3s keeps retrying until the private address answers.

~/acme-cluster — zsh

$ edge compute scripts create \

--name k3s-server --file k3s-server.sh

$ edge compute create --name cp-1 \

--size s-2vcpu-4gb --image ubuntu-24 \

--region london --script k3s-server

✓ vm-3c91d0 running · 10.0.0.2

$ kubectl get nodes

NAME STATUS ROLES

cp-1 Ready control-plane,etcd

worker-1 Ready <none>

Control-plane fee$0

k3s · londonHealthy

Nodes6 Ready

Control plane3 · embedded etcd

1. 01

   Write the bootstrap scripts

   Startup scripts substitute ${param} values before they run. The first server starts etcd; the rest join it.

   k3s-server.sh · k3s-agent.shCopy

   #!/bin/bash
   # k3s-server.sh: empty join_ip starts a new cluster
   set -e
   ARGS="--cluster-init"
   [ -n "${join_ip}" ] && ARGS="--server https://${join_ip}:6443"
   curl -sfL https://get.k3s.io | K3S_TOKEN="${k3s_token}" \
     sh -s - server $ARGS
   
   #!/bin/bash
   # k3s-agent.sh
   set -e
   curl -sfL https://get.k3s.io | K3S_URL="https://${server_ip}:6443" \
     K3S_TOKEN="${k3s_token}" sh -

2. 02

   Bring up the control plane

   Register both scripts, then create three servers. cp-1 initialises etcd and the other two join it.

   shellCopy

   $ edge compute scripts create --name k3s-server --file k3s-server.sh
   $ edge compute scripts create --name k3s-agent --file k3s-agent.sh
   $ TOKEN=$(openssl rand -hex 32)
   
   $ edge compute create --name cp-1 --size s-2vcpu-4gb \
       --image ubuntu-24 --region london --script k3s-server \
       --script-param k3s_token=$TOKEN
   $ for n in 2 3; do edge compute create --name cp-$n \
       --size s-2vcpu-4gb --image ubuntu-24 --region london \
       --script k3s-server --script-param k3s_token=$TOKEN \
       --script-param join_ip=10.0.0.2; done

3. 03

   Add the workers

   Performance VMs give pods 160 GB of local NVMe each. Copy the kubeconfig and swap 127.0.0.1 for cp-1's address.

   shellCopy

   $ for n in 1 2 3; do edge compute create --name worker-$n \
       --size s-4vcpu-8gb --image ubuntu-24 --region london \
       --script k3s-agent --script-param k3s_token=$TOKEN \
       --script-param server_ip=10.0.0.2; done
   
   $ scp root@<cp-1-ip>:/etc/rancher/k3s/k3s.yaml ~/.kube/config
   $ kubectl get nodes

4. 04

   Lock it down and front the ingress

   Only HTTPS is public; the API server answers on the private range. The CDN caches what it can and passes the rest to Traefik.

   shellCopy

   $ edge compute groups create --name k8s-nodes
   $ edge compute groups add-rule grp-abc123 --port 443 \
       --protocol tcp --source 0.0.0.0/0
   $ edge compute groups add-rule grp-abc123 --port 6443 \
       --protocol tcp --source 10.0.0.0/8
   
   $ edge cdn create --name acme-cluster
   $ edge cdn domains add cdn-a1b2c3 \
       --domain app.acme.io --origin https://<worker-ip>

5. 05

   Back up the cluster to Edge Storage

   Velero's AWS plugin speaks to any S3 endpoint. Schedule nightly backups and test a restore before you need one.

   shellCopy

   $ edge storage create k8s-backups
   $ S3=https://storage.edge.network
   $ velero install --provider aws --bucket k8s-backups \
       --plugins velero/velero-plugin-for-aws:v1.12.0 \
       --secret-file ./edge-credentials \
       --use-volume-snapshots=false \
       --backup-location-config \
         region=us-east-1,s3ForcePathStyle=true,s3Url=$S3
   $ velero schedule create nightly --schedule="0 3 * * *"

Prefer to hand it off? Give the job to your AI agent or have our engineers do it.