Kubernetes — k3s
Version: k3s v1.34.4+k3s1
Kubeconfig: /root/.kube/config
k3s is a lightweight Kubernetes distribution developed by Rancher. It packages all Kubernetes components into a single binary (~70MB) — perfect for running a production-grade Kubernetes cluster on a single server without the overhead of full k8s.
Why k3s?
| Feature | k3s | Full Kubernetes |
|---|
| Binary size | ~70MB | Multiple large binaries |
| RAM usage | ~512MB | ~2GB+ |
| Install time | ~1 minute | 30+ minutes |
| Complexity | Low | High |
| Production-ready | Yes | Yes |
| Cloud-native compatible | Yes | Yes |
Cluster Info
# Check cluster status
kubectl cluster-info
# Check all nodes
kubectl get nodes -o wide
# Check all pods across all namespaces
kubectl get pods -A
Namespaces
| Namespace | Contents |
|---|
kube-system | k3s system components, kube-state-metrics |
coderz | Coderz application workloads |
Deployed Workloads
coderz-web (in coderz namespace)
A sample Nginx web application running with 2 replicas:
kubectl get deployments -n coderz
kubectl get pods -n coderz
kubectl get services -n coderz
kube-state-metrics (in kube-system)
Exposes Kubernetes object state as Prometheus metrics:
- Pod status (Running, Pending, Failed, CrashLoopBackOff)
- Node status (Ready, NotReady)
- Deployment replica counts (desired vs available)
- ReplicaSet and StatefulSet status
ClusterIP: 10.43.126.54:8080
Prometheus Integration
Prometheus scrapes the k3s cluster:
| Target | What It Collects |
|---|
k3s API server :6443 | API server health, request rates |
kube-state-metrics :8080 | Pod/node/deployment state |
Node Exporter :9100 | Host metrics for each node |
cAdvisor :8080 | Container metrics for k8s pods |
/opt/coderz/configs/prometheus/secrets/k3s-token
Kubernetes Dashboard in Grafana
The Kubernetes Overview dashboard shows:
- Node count and status
- Pod count (by namespace, by status)
- Deployment desired vs available replicas
- Container restart counts
- Resource usage per pod
Common kubectl Commands
# Deploy a new application
kubectl apply -f deployment.yaml
# Scale a deployment
kubectl scale deployment coderz-web --replicas=3 -n coderz
# Check pod logs
kubectl logs -f deployment/coderz-web -n coderz
# Exec into a pod
kubectl exec -it <pod-name> -n coderz -- /bin/sh
# Check resource usage
kubectl top pods -n coderz
kubectl top nodes
# Describe a pod (for debugging)
kubectl describe pod <pod-name> -n coderz
# Delete and redeploy
kubectl rollout restart deployment/coderz-web -n coderz
Prefect k8s Health Flow
Prefect automatically checks k3s health every 15 minutes:
- All nodes must be in
Ready state
- All pods in
coderz namespace must be Running
- No pods in
CrashLoopBackOff or OOMKilled
If any check fails → email alert is sent.
Uninstall k3s
/usr/local/bin/k3s-uninstall.sh
This will destroy the entire Kubernetes cluster and all workloads. Only run this if you intend to completely remove k3s from the server.