Creating Highly Available clusters with kubeadm

This page explains two different approaches to setting up a highly available Kubernetes cluster using kubeadm:

With stacked control plane nodes. This approach requires less infrastructure. The etcd members and control plane nodes are co-located.
With an external etcd cluster. This approach requires more infrastructure. The control plane nodes and etcd members are separated.

Before proceeding, you should carefully consider which approach best meets the needs of your applications and environment. This comparison topic outlines the advantages and disadvantages of each.

If you encounter issues with setting up the HA cluster, please provide us with feedback in the kubeadm issue tracker.

Before you begin

For both methods you need this infrastructure:

Three machines that meet kubeadm's minimum requirements for the control-plane nodes
Three machines that meet kubeadm's minimum requirements for the workers
Full network connectivity between all machines in the cluster (public or private network)
sudo privileges on all machines
SSH access from one device to all nodes in the system
kubeadm and kubelet installed on all machines. kubectl is optional.

For the external etcd cluster only, you also need:

Three additional machines for etcd members

First steps for both methods

Create load balancer for kube-apiserver

Note: There are many configurations for load balancers. The following example is only one option. Your cluster requirements may need a different configuration.

Create a kube-apiserver load balancer with a name that resolves to DNS.
- In a cloud environment you should place your control plane nodes behind a TCP forwarding load balancer. This load balancer distributes traffic to all healthy control plane nodes in its target list. The health check for an apiserver is a TCP check on the port the kube-apiserver listens on (default value :6443).
- It is not recommended to use an IP address directly in a cloud environment.
- The load balancer must be able to communicate with all control plane nodes on the apiserver port. It must also allow incoming traffic on its listening port.
- Make sure the address of the load balancer always matches the address of kubeadm's ControlPlaneEndpoint.
- Read the Options for Software Load Balancing guide for more details.
Add the first control plane nodes to the load balancer and test the connection:
```
nc -v LOAD_BALANCER_IP PORT
```
- A connection refused error is expected because the apiserver is not yet running. A timeout, however, means the load balancer cannot communicate with the control plane node. If a timeout occurs, reconfigure the load balancer to communicate with the control plane node.
Add the remaining control plane nodes to the load balancer target group.

Stacked control plane and etcd nodes

Steps for the first control plane node

Initialize the control plane:
```
sudo kubeadm init --control-plane-endpoint "LOAD_BALANCER_DNS:LOAD_BALANCER_PORT" --upload-certs
```
- You can use the --kubernetes-version flag to set the Kubernetes version to use. It is recommended that the versions of kubeadm, kubelet, kubectl and Kubernetes match.
- The --control-plane-endpoint flag should be set to the address or DNS and port of the load balancer.
- The --upload-certs flag is used to upload the certificates that should be shared across all the control-plane instances to the cluster. If instead, you prefer to copy certs across control-plane nodes manually or using automation tools, please remove this flag and refer to Manual certificate distribution section below.
Note: The kubeadm init flags --config and --certificate-key cannot be mixed, therefore if you want to use the kubeadm configuration you must add the certificateKey field in the appropriate config locations (under InitConfiguration and JoinConfiguration: controlPlane).

Note: Some CNI network plugins require additional configuration, for example specifying the pod IP CIDR, while others do not. See the CNI network documentation. To add a pod CIDR pass the flag --pod-network-cidr, or if you are using a kubeadm configuration file set the podSubnet field under the networking object of ClusterConfiguration.
- The output looks similar to:
```
...
You can now join any number of control-plane node by running the following command on each as a root:
    kubeadm join 192.168.0.200:6443 --token 9vr73a.a8uxyaju799qwdjv --discovery-token-ca-cert-hash sha256:7c2e69131a36ae2a042a339b33381c6d0d43887e2de83720eff5359e26aec866 --control-plane --certificate-key f8902e114ef118304e561c3ecd4d0b543adc226b7a07f675f56564185ffe0c07

Please note that the certificate-key gives access to cluster sensitive data, keep it secret!
As a safeguard, uploaded-certs will be deleted in two hours; If necessary, you can use kubeadm init phase upload-certs to reload certs afterward.

Then you can join any number of worker nodes by running the following on each as root:
    kubeadm join 192.168.0.200:6443 --token 9vr73a.a8uxyaju799qwdjv --discovery-token-ca-cert-hash sha256:7c2e69131a36ae2a042a339b33381c6d0d43887e2de83720eff5359e26aec866
```
- Copy this output to a text file. You will need it later to join control plane and worker nodes to the cluster.
- When --upload-certs is used with kubeadm init, the certificates of the primary control plane are encrypted and uploaded in the kubeadm-certs Secret.
- To re-upload the certificates and generate a new decryption key, use the following command on a control plane node that is already joined to the cluster:
```
sudo kubeadm init phase upload-certs --upload-certs
```
- You can also specify a custom --certificate-key during init that can later be used by join. To generate such a key you can use the following command:
```
kubeadm certs certificate-key
```
Note: The kubeadm-certs Secret and decryption key expire after two hours.

Caution: As stated in the command output, the certificate key gives access to cluster sensitive data, keep it secret!
Apply the CNI plugin of your choice: Follow these instructions to install the CNI provider. Make sure the configuration corresponds to the Pod CIDR specified in the kubeadm configuration file if applicable.

In this example we are using Weave Net:
```
kubectl apply -f "https://cloud.weave.works/k8s/net?k8s-version=$(kubectl version | base64 | tr -d '\n')"
```
Type the following and watch the pods of the control plane components get started:
```
kubectl get pod -n kube-system -w
```

Steps for the rest of the control plane nodes

Note: Since kubeadm version 1.15 you can join multiple control-plane nodes in parallel. Prior to this version, you must join new control plane nodes sequentially, only after the first node has finished initializing.

For each additional control plane node you should:

Execute the join command that was previously given to you by the kubeadm init output on the first node. It should look something like this:
```
sudo kubeadm join 192.168.0.200:6443 --token 9vr73a.a8uxyaju799qwdjv --discovery-token-ca-cert-hash sha256:7c2e69131a36ae2a042a339b33381c6d0d43887e2de83720eff5359e26aec866 --control-plane --certificate-key f8902e114ef118304e561c3ecd4d0b543adc226b7a07f675f56564185ffe0c07
```
- The --control-plane flag tells kubeadm join to create a new control plane.
- The --certificate-key ... will cause the control plane certificates to be downloaded from the kubeadm-certs Secret in the cluster and be decrypted using the given key.

External etcd nodes

Setting up a cluster with external etcd nodes is similar to the procedure used for stacked etcd with the exception that you should setup etcd first, and you should pass the etcd information in the kubeadm config file.

Set up the etcd cluster

Follow these instructions to set up the etcd cluster.
Setup SSH as described here.

Copy the following files from any etcd node in the cluster to the first control plane node:

export CONTROL_PLANE="ubuntu@10.0.0.7"
scp /etc/kubernetes/pki/etcd/ca.crt "${CONTROL_PLANE}":
scp /etc/kubernetes/pki/apiserver-etcd-client.crt "${CONTROL_PLANE}":
scp /etc/kubernetes/pki/apiserver-etcd-client.key "${CONTROL_PLANE}":

Replace the value of CONTROL_PLANE with the user@host of the first control-plane node.

Set up the first control plane node

Create a file called kubeadm-config.yaml with the following contents:

apiVersion: kubeadm.k8s.io/v1beta3
kind: ClusterConfiguration
kubernetesVersion: stable
controlPlaneEndpoint: "LOAD_BALANCER_DNS:LOAD_BALANCER_PORT"
etcd:
    external:
        endpoints:
        - https://ETCD_0_IP:2379
        - https://ETCD_1_IP:2379
        - https://ETCD_2_IP:2379
        caFile: /etc/kubernetes/pki/etcd/ca.crt
        certFile: /etc/kubernetes/pki/apiserver-etcd-client.crt
        keyFile: /etc/kubernetes/pki/apiserver-etcd-client.key

Note: The difference between stacked etcd and external etcd here is that the external etcd setup requires a configuration file with the etcd endpoints under the external object for etcd. In the case of the stacked etcd topology this is managed automatically.

Replace the following variables in the config template with the appropriate values for your cluster:

- `LOAD_BALANCER_DNS`
- `LOAD_BALANCER_PORT`
- `ETCD_0_IP`
- `ETCD_1_IP`
- `ETCD_2_IP`

The following steps are similar to the stacked etcd setup:

Run sudo kubeadm init --config kubeadm-config.yaml --upload-certs on this node.
Write the output join commands that are returned to a text file for later use.

Apply the CNI plugin of your choice. The given example is for Weave Net:

kubectl apply -f "https://cloud.weave.works/k8s/net?k8s-version=$(kubectl version | base64 | tr -d '\n')"

Steps for the rest of the control plane nodes

The steps are the same as for the stacked etcd setup:

Make sure the first control plane node is fully initialized.
Join each control plane node with the join command you saved to a text file. It's recommended to join the control plane nodes one at a time.
Don't forget that the decryption key from --certificate-key expires after two hours, by default.

Common tasks after bootstrapping control plane

Install workers

Worker nodes can be joined to the cluster with the command you stored previously as the output from the kubeadm init command:

sudo kubeadm join 192.168.0.200:6443 --token 9vr73a.a8uxyaju799qwdjv --discovery-token-ca-cert-hash sha256:7c2e69131a36ae2a042a339b33381c6d0d43887e2de83720eff5359e26aec866

Manual certificate distribution

If you choose to not use kubeadm init with the --upload-certs flag this means that you are going to have to manually copy the certificates from the primary control plane node to the joining control plane nodes.

There are many ways to do this. In the following example we are using ssh and scp:

SSH is required if you want to control all nodes from a single machine.

Enable ssh-agent on your main device that has access to all other nodes in the system:
```
eval $(ssh-agent)
```
Add your SSH identity to the session:
```
ssh-add ~/.ssh/path_to_private_key
```
SSH between nodes to check that the connection is working correctly.
- When you SSH to any node, make sure to add the -A flag:
```
ssh -A 10.0.0.7
```
- When using sudo on any node, make sure to preserve the environment so SSH forwarding works:
```
sudo -E -s
```

After configuring SSH on all the nodes you should run the following script on the first control plane node after running kubeadm init. This script will copy the certificates from the first control plane node to the other control plane nodes:

In the following example, replace CONTROL_PLANE_IPS with the IP addresses of the other control plane nodes.

USER=ubuntu # customizable
CONTROL_PLANE_IPS="10.0.0.7 10.0.0.8"
for host in ${CONTROL_PLANE_IPS}; do
    scp /etc/kubernetes/pki/ca.crt "${USER}"@$host:
    scp /etc/kubernetes/pki/ca.key "${USER}"@$host:
    scp /etc/kubernetes/pki/sa.key "${USER}"@$host:
    scp /etc/kubernetes/pki/sa.pub "${USER}"@$host:
    scp /etc/kubernetes/pki/front-proxy-ca.crt "${USER}"@$host:
    scp /etc/kubernetes/pki/front-proxy-ca.key "${USER}"@$host:
    scp /etc/kubernetes/pki/etcd/ca.crt "${USER}"@$host:etcd-ca.crt
    # Quote this line if you are using external etcd
    scp /etc/kubernetes/pki/etcd/ca.key "${USER}"@$host:etcd-ca.key
done

Caution: Copy only the certificates in the above list. kubeadm will take care of generating the rest of the certificates with the required SANs for the joining control-plane instances. If you copy all the certificates by mistake, the creation of additional nodes could fail due to a lack of required SANs.

Then on each joining control plane node you have to run the following script before running kubeadm join. This script will move the previously copied certificates from the home directory to /etc/kubernetes/pki:

USER=ubuntu # customizable
mkdir -p /etc/kubernetes/pki/etcd
mv /home/${USER}/ca.crt /etc/kubernetes/pki/
mv /home/${USER}/ca.key /etc/kubernetes/pki/
mv /home/${USER}/sa.pub /etc/kubernetes/pki/
mv /home/${USER}/sa.key /etc/kubernetes/pki/
mv /home/${USER}/front-proxy-ca.crt /etc/kubernetes/pki/
mv /home/${USER}/front-proxy-ca.key /etc/kubernetes/pki/
mv /home/${USER}/etcd-ca.crt /etc/kubernetes/pki/etcd/ca.crt
# Quote this line if you are using external etcd
mv /home/${USER}/etcd-ca.key /etc/kubernetes/pki/etcd/ca.key

Last modified October 26, 2021 at 12:09 AM PST : kubeadm: use git.k8s.io to link to kubeadm repo (0162445d1)

Creating Highly Available clusters with kubeadm

Before you begin

First steps for both methods

Create load balancer for kube-apiserver

Stacked control plane and etcd nodes

Steps for the first control plane node

Steps for the rest of the control plane nodes

External etcd nodes

Set up the etcd cluster

Set up the first control plane node

Steps for the rest of the control plane nodes

Common tasks after bootstrapping control plane

Install workers

Manual certificate distribution

Feedback