K8s Quickly - Cheat Sheet
🔍 Context & Namespaces
List contexts — shows available kube contexts
kubectl config get-contexts
CURRENT NAME CLUSTER AUTHINFO * dev-cluster dev user-dev
Switch context
kubectl config use-context dev-cluster
→ Switched to context "dev-cluster".
List namespaces
kubectl get ns
NAME STATUS AGE default Active 10d monitoring Active 3d
Set default namespace
kubectl config set-context --current --namespace=app-prod
→ Context updated.
📦 Pods & Containers
List all pods
kubectl get pods -A
NAMESPACE NAME STATUS app-prod api-7d9df9b77b-xrjhg Running app-prod web-6cb5d49b87-l4ztz CrashLoopBackOff
Describe pod
kubectl describe pod web-6cb5d49b87-l4ztz -n app-prod
→ Shows events, restarts, image, node, reason for crash.
View logs
kubectl logs web-6cb5d49b87-l4ztz -n app-prod
Error: cannot connect to postgres at db-service:5432
Exec into container
kubectl exec -it web-6cb5d49b87-l4ztz -n app-prod -- /bin/bash
→ Opens interactive shell inside the container.
⚙️ Deployments
List deployments
kubectl get deploy -A
NAMESPACE NAME READY UP-TO-DATE AVAILABLE app-prod web 1/1 1 1
Update image
kubectl set image deploy/web web=myrepo/web:v2 -n app-prod
→ Updates running container image.
Restart rollout
kubectl rollout restart deploy/web -n app-prod
→ Forces a new pod rollout.
🧩 Nodes & GPUs
List nodes (with IPs)
kubectl get nodes -o wide
NAME STATUS ROLES INTERNAL-IP gpu-node-01 Ready worker 10.42.0.12
View node GPU capacity
kubectl describe node gpu-node-01 | grep -A5 "Capacity"
nvidia.com/gpu: 8
List pods using GPUs
kubectl get pods -A -o custom-columns='NS:.metadata.namespace,NAME:.metadata.name,GPU:.spec.containers[*].resources.limits.nvidia.com/gpu'
NS NAME GPU training trainer-abc123 4
🧾 Configs & Secrets
List ConfigMaps
kubectl get cm -n app-prod
NAME DATA web-config 3
View ConfigMap
kubectl get cm web-config -n app-prod -o yaml
→ Prints key-value data.
Decode a Secret
kubectl get secret db-secret -n app-prod -o jsonpath='{.data.password}' | base64 --decode
→ Reveals decoded secret value.
📊 Monitoring & Debug
Top nodes
kubectl top nodes
NAME CPU(%) MEM(%) gpu-node-01 72% 68%
Top pods
kubectl top pods -A
NAMESPACE NAME CPU(%) MEM(%) training trainer 1800m 5Gi
Cluster events
kubectl get events --sort-by=.metadata.creationTimestamp
→ Shows recent failures, restarts, scheduling delays.
Debug into pod
kubectl debug -it trainer-abc123 -n training --image=ubuntu
→ Spawns ephemeral container for inspection.
🧹 Cleanup
Delete pod
kubectl delete pod web-6cb5d49b87-l4ztz -n app-prod
→ Kills and lets Deployment recreate it.
Remove evicted pods
kubectl get pods -A | grep Evicted | awk '{print $2}' | xargs kubectl delete pod -n app-prod
→ Clears out evicted pods.
Delete old jobs
kubectl delete jobs --all -n training
→ Cleans up completed batch jobs.
💡 Tips: Use -o wide for extra info, --show-labels for labels, and watch kubectl get pods -A for live updates.
Check GPU plugin status: kubectl get ds -A | grep nvidia
Components
🧠 Core Components
API Server
Exposes the Kubernetes API — all kubectl commands and controllers communicate through it.
etcd
Cluster’s key-value database. Stores state, configs, and resource definitions.
Controller Manager
Watches resources and ensures actual state matches desired state (e.g. keeps replicas running).
Scheduler
Decides which node each Pod should run on based on resources, taints, and affinities.
Kubelet
Node-level agent. Talks to the API server and ensures containers are running as defined.
Kube Proxy
Handles Pod networking and load-balancing. Routes traffic between Services and Pods.
Container Runtime
Runs the actual containers. Common options: containerd, CRI-O, or legacy Docker.
📦 Workload Components
Pod
Smallest deployable unit — one or more containers sharing the same network and storage.
Deployment
Defines how many Pod replicas should run and handles rolling updates.
Service
Provides stable networking and load-balancing for a set of Pods.
Ingress
Manages external access to Services, usually HTTP/HTTPS. Popular options: NGINX, Traefik.
ConfigMap & Secret
Store environment variables, configs, or credentials separately from container images.
🗂️ Cluster & Storage
Node
Worker machine (virtual or physical) that runs Pods. Each has a Kubelet and runtime.
Namespace
Logical isolation inside the cluster — separate teams, environments, or apps.
Volume / PersistentVolume
Provides storage to Pods. PersistentVolumes (PVs) represent the actual storage, while PVCs claim it.
Storage Options
CSI drivers integrate external storage: EBS, NFS, Ceph, Longhorn, Portworx.
Cluster
The complete Kubernetes system — control plane (API, etcd, controller, scheduler) + worker nodes.
🔌 Common Add-Ons
Networking (CNI)
Manages Pod-to-Pod networking. Common: Calico, Cilium, Flannel.
Service Mesh
Adds traffic control, security, and observability. Common: Istio, Linkerd.
Monitoring
Prometheus and Grafana for metrics, Lens for visual cluster insights.
Security & Policy
Gatekeeper, Kyverno, and Falco for compliance and runtime protection.
Package Management
Helm or Kustomize for deploying apps; ArgoCD for GitOps workflows.
💡 Quick mental model: The control plane is the brain, the nodes are the hands, and Pods are the workers doing the job.
Cluster add-ons handle networking, storage, monitoring, and security layers.