There is no single “Kubernetes installer”. There are many distributions, and the right one depends on what you want to do. The short version: use kind or minikube to learn, and a managed cloud cluster for production. This lesson shows each option so you can choose with confidence.
The options at a glance
| Option | Best for | Runs on | Multi-node |
|---|---|---|---|
| kind | Learning, CI pipelines, testing | Docker containers | Yes, with a small config file |
| minikube | Learning, trying add-ons such as Ingress | A VM or Docker | Yes |
| Docker Desktop | A quick look if you already use it | Its own VM | No |
| k3s | Edge devices, small servers, a cheap VPS | A real Linux host | Yes |
| kubeadm | Learning the internals, self-managed clusters | Real machines or VMs | Yes |
| EKS, GKE, AKS | Production | The cloud provider | Yes |
Whatever you choose, install kubectl first. It is the command line client you will use in every lesson, and it works the same with all of them.
For learning: kind or minikube
kind
kind (“Kubernetes in Docker”) runs every node as a Docker container. Clusters start in under a minute and are cheap to throw away, which makes kind ideal for experiments and CI.
$ kind create cluster --name lab
Creating cluster "lab" ...
✓ Ensuring node image (kindest/node:v1.34.0)
✓ Preparing nodes
✓ Starting control-plane
Set kubectl context to "kind-lab"
$ kubectl get nodes
NAME STATUS ROLES AGE VERSION
lab-control-plane Ready control-plane 45s v1.34.0
A multi-node cluster needs a small config file, kind-3nodes.yaml:
kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
- role: control-plane
- role: worker
- role: worker
$ kind create cluster --name lab --config kind-3nodes.yaml
Watch out: kind nodes have their own image store and cannot see the images in your local Docker. After docker build -t shop-api:dev ., run kind load docker-image shop-api:dev --name lab, or your Pods end in ImagePullBackOff.
minikube
minikube is more “batteries included”. Its add-on system turns on Ingress, metrics-server and a dashboard with one command each.
$ minikube start --cpus=4 --memory=8192
😄 minikube v1.37.0
🏄 Done! kubectl is now configured to use "minikube" cluster
$ minikube addons enable ingress
$ minikube addons enable metrics-server
$ minikube service web --url
http://192.168.49.2:31380
Two more commands you will use often: minikube tunnel gives LoadBalancer Services a real address on your machine, and minikube dashboard opens the web UI.
Which one should I pick?
Pick kind if you want speed, several nodes or a cluster inside CI. Pick minikube if you want Ingress and metrics without any setup. Both are good choices, and nothing in this course depends on one of them.
Docker Desktop
If Docker Desktop is already installed, open its settings and tick “Enable Kubernetes”. It is the least effort, but it has limits: one node only, a Kubernetes version tied to the Docker Desktop release and memory shared with all your other containers. It is fine for a first look and less good for learning scheduling.
k3s
k3s is a certified Kubernetes distribution packed into a single small binary. It replaces etcd with SQLite by default and ships with an ingress controller and a simple load balancer. It is a great fit for a home lab, a Raspberry Pi or a small VPS.
$ curl -sfL https://get.k3s.io | sh -
[INFO] Finding release for channel stable
[INFO] systemd: Starting k3s
$ sudo k3s kubectl get nodes
NAME STATUS ROLES AGE VERSION
vps-01 Ready control-plane,master 30s v1.34.1+k3s1
kubeadm
kubeadm is the official tool to build a cluster on machines you control. You install a container runtime on each machine, initialize the control plane, add a network plugin and join the workers:
$ sudo kubeadm init --pod-network-cidr=192.168.0.0/16
Your Kubernetes control-plane has initialized successfully!
$ kubectl apply -f https://raw.githubusercontent.com/projectcalico/calico/v3.30.0/manifests/calico.yaml
$ kubeadm token create --print-join-command
kubeadm join 10.0.0.10:6443 --token abc123.xyz --discovery-token-ca-cert-hash sha256:...
Doing this once is the best way to understand how the pieces from lesson 2 fit together. Running it in production is a different commitment: you own etcd backups, certificate rotation, control plane upgrades and node patching.
Managed clusters for production
For production, let a cloud provider run the control plane. EKS (AWS), GKE (Google Cloud) and AKS (Azure) handle the API server, etcd backups and control plane upgrades.
$ eksctl create cluster --name shop --region us-east-1 --nodes 3
$ gcloud container clusters create-auto shop --region us-central1
$ az aks create --resource-group shop-rg --name shop --node-count 3 --generate-ssh-keys
You still own the worker nodes (or their configuration), networking, add-on upgrades and who has access.
Resources your laptop needs
A local cluster needs at least 2 CPUs and 4 GB of free memory. Give it 4 CPUs and 8 GB if you plan to run Ingress and the monitoring stack from lesson 13. If your Pods stay Pending with Insufficient cpu or Insufficient memory, the cluster is too small: delete it and start it again with more resources.
Check that it works
Three commands tell you if a new cluster is healthy:
$ kubectl cluster-info
Kubernetes control plane is running at https://127.0.0.1:41235
CoreDNS is running at https://127.0.0.1:41235/api/v1/namespaces/kube-system/services/kube-dns:dns/proxy
$ kubectl get nodes
$ kubectl get pods -A
Every Pod in kube-system should be Running. Then run a real test from start to finish:
$ kubectl create deployment hello --image=nginx:1.27
deployment.apps/hello created
$ kubectl expose deployment hello --port=80
service/hello exposed
$ kubectl port-forward service/hello 8080:80
Forwarding from 127.0.0.1:8080 -> 80
Open http://localhost:8080 in a browser and you should see the nginx welcome page. Delete everything with kubectl delete deployment,service hello.
Quick reference
| Task | Command |
|---|---|
| Create a kind cluster | kind create cluster --name lab |
| Create a multi-node kind cluster | kind create cluster --config kind-3nodes.yaml |
| Load a local image into kind | kind load docker-image <image> --name lab |
| Start minikube with more resources | minikube start --cpus=4 --memory=8192 |
| Enable a minikube add-on | minikube addons enable ingress |
| Check the cluster | kubectl get nodes and kubectl get pods -A |
| Show the API endpoint | kubectl cluster-info |
| Delete the cluster | kind delete cluster --name lab or minikube delete |
Exercise
Which option fits best for a CI pipeline that needs a disposable multi-node cluster for every run?
Show answer
Answer: C. kind creates multi-node clusters as Docker containers in under a minute and deletes them just as fast, which is exactly what CI needs. Docker Desktop is single-node and the others are slow and costly to create per run.
Exercise
You move to a managed cluster such as EKS, GKE or AKS. What are you still responsible for?
Show answer
Answer: A. The provider runs the control plane, including etcd and its backups. Everything that runs your workloads and decides who can touch them is still yours.
Key takeaways
- Use kind or minikube for learning and a managed cluster for production.
- kind is fastest and great in CI; minikube’s add-ons make Ingress and metrics easy.
- Use a multi-node local cluster to see real scheduling behavior.
- kind cannot see your local Docker images: use
kind load docker-image. minikube tunnelmakes LoadBalancer Services work on a laptop.- Give a local cluster 4 CPUs and 8 GB when you can.
- Build a cluster with kubeadm once to understand it; production on kubeadm means owning etcd, certificates and upgrades.
FAQ
What is the easiest way to install Kubernetes for learning?
kind or minikube. Both create a working cluster with one command on Linux, macOS or Windows, and both are easy to delete and recreate.
Should I use the Kubernetes built into Docker Desktop?
It is fine for a quick look. Because it is single-node and shares resources with your other containers, kind or minikube is a better choice for following this course.
Can I run Kubernetes in production without a cloud provider?
Yes, with kubeadm or k3s on your own machines. You then take on etcd backups, certificates, upgrades, operating system patching and integrating storage and load balancers yourself.
Why are the CoreDNS Pods Pending after kubeadm init?
No network plugin (CNI) is installed yet. Apply one such as Calico, Flannel or Cilium, and make sure its Pod network range matches the --pod-network-cidr you gave kubeadm.