How to Set Up a Multi-Node Kubernetes Cluster with kubeadm

kubeadm is a tool that sets up a production-grade Kubernetes cluster on any Linux machines. It handles the control plane initialisation and generates the join commands for worker nodes. This guide walks through setting up a cluster with one control plane node and two worker nodes on Ubuntu.

This setup is useful for learning how Kubernetes works at a deeper level, or for running Kubernetes on bare metal or cloud VMs without a managed service.

What you will build

Three Ubuntu nodes:

• control-plane — runs the API server, scheduler, controller manager, and etcd
• node-1 — worker node
• node-2 — worker node

Requirements

Each node needs:

• Ubuntu 20.04 or 22.04
• 2 CPUs or more
• 2GB RAM or more
• Full network connectivity between all nodes
• Unique hostname, MAC address, and product_uuid per node
• Root or sudo access

Step 1: Run on all three nodes

The following steps must be completed on the control plane node and both worker nodes before initialising the cluster.

Load required kernel modules

1cat <<EOF | sudo tee /etc/modules-load.d/containerd.conf
2overlay
3br_netfilter
4EOF
5
6sudo modprobe overlay
7sudo modprobe br_netfilter

Configure networking for Kubernetes

1cat <<EOF | sudo tee /etc/sysctl.d/99-kubernetes-cri.conf
2net.bridge.bridge-nf-call-iptables = 1
3net.ipv4.ip_forward = 1
4net.bridge.bridge-nf-call-ip6tables = 1
5EOF
6
7sudo sysctl --system

Install containerd

1sudo apt-get update && sudo apt-get install -y containerd

Create the containerd configuration file:

1sudo mkdir -p /etc/containerd
2sudo containerd config default | sudo tee /etc/containerd/config.toml

Restart containerd:

1sudo systemctl restart containerd

Disable swap

Kubernetes requires swap to be disabled. The scheduler makes pod placement decisions based on available memory and swap can cause instability:

1sudo swapoff -a

Install kubeadm, kubelet, and kubectl

Install the dependency packages:

1sudo apt-get update && sudo apt-get install -y apt-transport-https ca-certificates curl

Add the Kubernetes package repository signing key:

1curl -fsSL https://pkgs.k8s.io/core:/stable:/v1.29/deb/Release.key | sudo gpg --dearmor -o /etc/apt/keyrings/kubernetes-apt-keyring.gpg

Add the Kubernetes repository:

1echo 'deb [signed-by=/etc/apt/keyrings/kubernetes-apt-keyring.gpg] https://pkgs.k8s.io/core:/stable:/v1.29/deb/ /' | sudo tee /etc/apt/sources.list.d/kubernetes.list

Update the package list and install:

1sudo apt-get update
2sudo apt-get install -y kubelet kubeadm kubectl

Prevent these packages from being automatically updated:

1sudo apt-mark hold kubelet kubeadm kubectl

Step 2: Initialise the control plane node

Run only on the control plane node:

1sudo kubeadm init --pod-network-cidr=10.244.0.0/16

The --pod-network-cidr value must match the CNI plugin you plan to install. The value above is correct for Flannel.

When the command completes, set up kubectl access for your user:

1mkdir -p $HOME/.kube
2sudo cp -i /etc/kubernetes/admin.conf $HOME/.kube/config
3sudo chown $(id -u):$(id -g) $HOME/.kube/config

Save the kubeadm join command from the output. You will need it in Step 4. It looks like this:

1kubeadm join <control-plane-ip>:6443 --token <token> --discovery-token-ca-cert-hash sha256:<hash>

Step 3: Install a CNI plugin

Without a CNI plugin, nodes will stay in NotReady state and pods will not be scheduled. Install Flannel on the control plane node:

1kubectl apply -f https://raw.githubusercontent.com/flannel-io/flannel/master/Documentation/kube-flannel.yml

Verify the Flannel pods are running:

1kubectl get pods -n kube-system

Wait until all pods show Running before proceeding.

Step 4: Join the worker nodes

Run the join command saved from Step 2 on each worker node:

1sudo kubeadm join <control-plane-ip>:6443 --token <token> --discovery-token-ca-cert-hash sha256:<hash>

Step 5: Verify the cluster is healthy

Run on the control plane node:

1kubectl get nodes

Expected output once all nodes have joined and the CNI is running:

1NAME            STATUS   ROLES           AGE     VERSION
2control-plane   Ready    control-plane   5m      v1.29.0
3node-1          Ready    <none>          2m      v1.29.0
4node-2          Ready    <none>          1m      v1.29.0

All three nodes should show Ready. If any node shows NotReady, check the CNI pods with kubectl get pods -n kube-system and check the kubelet logs with journalctl -u kubelet.

Frequently asked questions