Published: October 18, 2026 · 7 min read

Kubernetes Cluster Setup — Minikube, kubeadm and EKS

We know what a cluster is and what is inside it. Now we build one.

There is more than one way to get a Kubernetes cluster. This class covers all of them in short, then sets up a real cluster on AWS.

Here is the whole class in short:

• A cluster can be self managed or managed.
• Self managed means you build it and look after it.
• Minikube gives you a single node cluster on your own machine.
• kubeadm gives you a multi node cluster on real servers.
• Managed means the cloud runs the control plane for you.
• AWS has EKS, Azure has AKS, Google Cloud has GKE, IBM Cloud has IKS.
• OpenShift is Red Hat's Kubernetes with extra tools on top.
• For EKS you need three tools: kubectl, AWS CLI and eksctl.
• Then one command builds the whole cluster, in about 13 minutes.

Self Managed Clusters

You build the cluster yourself. There are two common tools.

Minikube — single node. It runs on your own machine. The master and the worker are the same node. It starts in a minute and it is made for learning.

kubeadm — multi node. It runs on real servers. You install it on each machine and join them together. This is a proper cluster, and everything in it is yours to fix.

Managed Clusters

Here the cloud provider runs the control plane. You never see the master nodes.

The picture above is the whole difference. In a self managed cluster both halves are yours. In a managed cluster the control plane belongs to the cloud, and you only look after the worker nodes.

Red Hat OpenShift

OpenShift is Kubernetes with extra tools added by Red Hat.

You get a web console, an image registry, build pipelines and stricter security rules, all in one package.

Why companies pick it: paid support from Red Hat, and security rules that are on by default. Banks and large companies use it a lot. Underneath, it is still Kubernetes.

Which Kubernetes Cluster Setup Should You Pick?

For the rest of this class we use EKS.

Setting Up an EKS Cluster

Take one Ubuntu VM. Install three tools on it. Then run one command.

Step 1 — Install kubectl

# 1. install kubectl

sudo apt-get update
sudo apt-get install -y curl ca-certificates

curl -LO "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/linux/amd64/kubectl"

sudo install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl

kubectl version --client

What each line does:

• apt-get update refreshes the package list. It installs nothing.
• curl downloads files. ca-certificates lets it trust HTTPS sites.
• The long curl line reads the latest version number, then downloads that version of kubectl.
• install copies the file into /usr/local/bin and makes it runnable. That folder is in your PATH, so kubectl now works from anywhere.

Step 2 — Install the AWS CLI

# 2. install the AWS CLI

curl "https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip" -o "awscliv2.zip"
unzip -q awscliv2.zip
sudo ./aws/install

aws --version

# then give it your keys:
aws configure

Download the zip, unzip it, run the installer. Then aws configure asks for your access key and secret. Without this step eksctl has no permission to build anything.

Step 3 — Install eksctl

# 3. install eksctl

ARCH=amd64
PLATFORM=$(uname -s)_$ARCH          # becomes Linux_amd64

curl -sLO "https://github.com/eksctl-io/eksctl/releases/latest/download/eksctl_${PLATFORM}.tar.gz"
tar -xzf "eksctl_${PLATFORM}.tar.gz" -C /tmp
sudo install -m 0755 /tmp/eksctl /usr/local/bin/eksctl

eksctl version

Same pattern as kubectl. Work out which build you need, download it, extract it, put it in your PATH.

eksctl is worth it. Without it you would create the VPC, the IAM roles, the control plane and the node group yourself.

Step 4 — Create the Cluster

# 4. create the cluster

eksctl create cluster \
  --name srinu-desetti-eks \
  --region ap-south-1 \
  --node-type t3.small \
  --nodes 2 \
  --zones ap-south-1a,ap-south-1b

# go get a coffee. this takes about 13 minutes.

# 5. check it

kubectl get nodes

# NAME                        STATUS   ROLES    AGE   VERSION
# ip-192-168-12-34...         Ready    <none>   2m    v1.34.1
# ip-192-168-56-78...         Ready    <none>   2m    v1.34.1

Two nodes, both Ready. The cluster is live.

What Happens While You Wait

One command, but a lot happens behind it. eksctl sends two stacks to CloudFormation.

Control plane — about 9 minutes

1. VPC, subnets and routing are created.
2. IAM roles are created.
3. AWS starts the master VMs in its own account, across two zones.
4. AWS checks that the API server answers and etcd is in sync.
5. Add-ons go in: vpc-cni, kube-proxy, coredns.

Step 3 takes most of the time. Those master VMs are not in your account and you never see them.

Worker nodes — about 4 minutes

1. A launch template and an auto scaling group are created.
2. Two EC2 instances start, in your account this time.
3. Each one runs a script that points kubelet at the cluster.
4. The node joins the cluster and authenticates with IAM.
5. The health check passes and the node goes Ready.

At the end eksctl writes the cluster address and keys into ~/.kube/config. That is why kubectl get nodes worked straight away.

Delete It When You Are Done

# when you are done, delete it

eksctl delete cluster --name srinu-desetti-eks --region ap-south-1

# EKS charges by the hour. Do not leave it running.

EKS charges by the hour, for the control plane and for every node. Delete the cluster as soon as you are finished with it.

What's Next

You can now pick a setup and build one. Minikube to learn. Kubeadm for your own servers. EKS, AKS or GKE when you want the cloud to run the master. The official eksctl docs (opens in a new tab) and the EKS user guide (opens in a new tab) have the full list of options. Next we put an app on this cluster.

← Back to Kubernetes Architecture

Continue to Kubernetes Pod →

Frequently Asked Questions

What are the ways to set up a Kubernetes cluster?

Two ways. Self managed means you build the cluster and look after it. Minikube gives you a single node on your own machine. Kubeadm gives you a multi node cluster on real servers. Managed means a cloud provider runs the control plane for you. AWS has EKS, Azure has AKS, Google Cloud has GKE and IBM Cloud has IKS. It is the same Kubernetes either way.

What is the difference between minikube and kubeadm?

Minikube runs a single node cluster on one machine, usually your laptop. It is made for learning and testing, and it starts in a minute. Kubeadm builds a real multi node cluster across several servers. You install it on each machine yourself and you look after everything afterwards. Use minikube to learn, kubeadm when you want a real cluster on your own servers.

What is a managed Kubernetes cluster?

A managed cluster is one where the cloud provider runs the control plane for you. You never see the master nodes and you never fix them. You only create and pay for the worker nodes that run your app. AWS calls it EKS, Azure calls it AKS, Google Cloud calls it GKE, and IBM Cloud calls it IKS. It costs a little more, but nobody has to wake up when the control plane breaks.

What is Red Hat OpenShift?

OpenShift is Kubernetes with extra tools added on top by Red Hat. It brings a web console, a built in image registry, build and deploy pipelines, and stricter security defaults. Companies pick it for two reasons: paid support from Red Hat, and the security rules it enforces by default. It is common in banks and large enterprises. Underneath it is still Kubernetes.

What tools do I need to create an EKS cluster?

Three tools on one machine. kubectl is how you talk to the cluster once it exists. The AWS CLI is how you talk to AWS. It also holds your access keys after you run aws configure. eksctl is the tool that builds the cluster. Without eksctl you would build the VPC, the IAM roles, the control plane and the node group by hand.

How long does it take to create an EKS cluster?

About 13 minutes. The control plane takes around 9 minutes. AWS creates the VPC and the IAM roles first. Then it starts the master VMs in its own account, across two zones. Last it checks that the API server answers. The worker nodes take around 4 minutes. EC2 instances have to start, run their setup script, join the cluster and pass a health check. Most of the wait is the control plane.

Why do my EKS worker nodes fail to launch?

The usual reason is an account limit. On new or free tier accounts the instance type you asked for may not be allowed. The node group stack then fails, even though the control plane worked. Check the CloudFormation events for the node group to see the real error. Try a different instance type or region, or use a normal paid account. Also remember to delete the cluster afterwards, because EKS charges by the hour.