核心 · Key Idea
In one line: vanilla K8s is heavy as a battleship; k3s packs the control plane into a single binary — 100 MB and one command to install. Home labs / edge / IoT / single-host production: all viable.
One-line k3s install#
# Master
curl -sfL https://get.k3s.io | sh -
# Auto-generated kubeconfig
sudo cat /etc/rancher/k3s/k3s.yaml
# Add a worker (get token from master)
curl -sfL https://get.k3s.io | K3S_URL=https://master:6443 \
K3S_TOKEN=xxx sh -kubectl get nodes just works — etcd / control plane / kubelet are all in one binary.
Analogy#
打个比方 · Analogy
Vanilla K8s is a full IT department — architects, DBAs, SREs, network engineers — fits big companies. k3s is a full-stack engineer — one person doing everything; small teams / personal / edge are actually more nimble.
Mainstream lightweight distros#
- k3s (Rancher / SUSE)
- Single binary + sqlite/etcd + Traefik / ServiceLB defaults. Most popular.
- k0s
- More 'zero-dependency' ethos; uses etcd or kine.
- MicroK8s (Canonical)
- snap install, Ubuntu's own.
- minikube / kind
- Local dev only — not production.
- Talos Linux
- Treats K8s as the only app on bare metal; the whole OS is immutable.
- RKE2 / EKS / GKE Autopilot
- Enterprise / cloud-managed.
How it works#
Practical notes#
- Batteries included: Traefik / metrics-server / local-path / ServiceLB — works out of the box on a single host. Disable any with
--disable=traefik. - Lightweight but really K8s: API-compatible — Helm / kubectl / CRDs / Operators all work.
- Datastore: < 30 nodes → embedded etcd is fine; more → external etcd / mysql / postgres.
- HA:
--cluster-init+ ≥3 server nodes for control-plane HA. - Edge fleets: Rancher Fleet / Cattle Drive manages hundreds of clusters via GitOps.
- Home lab: 3 Raspberry Pis + k3s + Tailscale = 100 % real production experience.
- Upgrade:
curl get.k3s.io | sh -s - --version v1.30.xor system-upgrade-controller.
Easy confusions#
k3s
Single binary, low memory.
Production / edge / personal all fine.
Production / edge / personal all fine.
Vanilla K8s (kubeadm)
Full component set — flexible but complex.
Big clusters / enterprise.
Big clusters / enterprise.