K3s Lightweight Kubernetes on an Offshore VPS

Full Kubernetes installations (kubeadm, kops, k0s with full plugins) require multiple control-plane components, separate etcd, and often dozens of GB of disk for the full distribution. K3s replaces this with a single Go binary that bundles the API server, controller-manager, scheduler, kubelet, and a built-in storage backend (sqlite by default, etcd or other databases optional). The result runs comfortably on a single 2 GB RAM VPS. For self-hosted use cases - a personal Kubernetes lab, a small production cluster for a side project, an edge deployment with limited resources - K3s is the right choice. You get the same kubectl API, the same manifest format, and the same Helm chart compatibility as full Kubernetes. The ecosystem is identical from a user perspective. The differences are internal - K3s ships sane defaults for ingress (Traefik built in), storage (local-path-provisioner), and load balancing (klipper-lb). The single-binary install is one of the friendliest in the k8s ecosystem. `curl -sfL https://get.k3s.io | sh -` and you have a working single-node cluster in under 60 seconds. For multi-node, run the same command with `K3S_URL` and `K3S_TOKEN` environment variables on each additional node and the cluster forms automatically.

A single-node K3s on a single offshore VPS works great for personal use, dev environments, and small production workloads where some downtime is acceptable. A 4 GB RAM / 2 vCPU VPS comfortably runs the K3s control plane plus 10 to 30 application pods. The control plane itself uses about 500 MB RAM. For higher availability, K3s supports HA setups with three server nodes and a shared backend. The recommended HA pattern is three server nodes with embedded etcd (replacing the default sqlite). All three serve the API and store state in a Raft-replicated etcd. Add any number of agent nodes for workload capacity. Three offshore VPSes with 4 GB RAM each form a small but production-suitable HA cluster. K3s integrates well with cilium, calico, or the built-in flannel CNI. For network policy enforcement and observability, cilium is the strongest choice. For simple flat networking, the default flannel just works. The decision usually comes down to whether you need NetworkPolicy enforcement (cilium or calico) or basic L3 routing only (flannel).

On a fresh Ubuntu 22.04 VPS, install K3s with default settings: `curl -sfL https://get.k3s.io | sh -`. After about 30 seconds the cluster is up. Verify: `kubectl --kubeconfig /etc/rancher/k3s/k3s.yaml get nodes` shows your node as Ready. Set the kubeconfig path for convenience: `echo 'export KUBECONFIG=/etc/rancher/k3s/k3s.yaml' >> ~/.bashrc && source ~/.bashrc`. For HA with three server nodes, install on the first node with `curl -sfL https://get.k3s.io | sh -s - server --cluster-init`. Capture the join token from `/var/lib/rancher/k3s/server/node-token`. On the second and third nodes, install with `curl -sfL https://get.k3s.io | K3S_URL=https://NODE1_IP:6443 K3S_TOKEN=THE_TOKEN sh -s - server`. Verify all three nodes show as Ready. To add an agent (workload-only) node, install with `curl -sfL https://get.k3s.io | K3S_URL=https://ANY_SERVER_IP:6443 K3S_TOKEN=THE_TOKEN sh -`. The agent joins the cluster as a worker. Deploy a test workload: `kubectl create deployment nginx --image=nginx:alpine && kubectl expose deployment nginx --port=80 --type=NodePort && kubectl get svc nginx`. The NodePort exposes the service on a high port on every node. K3s ships with Traefik as the default ingress, so you can also use ingress resources for TLS-terminated routing through a single port.