Docs about RKE2 can be found here.

There are two install types agent for pure worker nodes and server for management nodes. There need to be at least three server nodes. there can be any number of agent nodes.

It is probably best to install the same version of Kubernetes as the cluster the node will be attached to. If needed the whole cluster can be upgraded after adding the nodes.

curl -sfL https://get.rke2.io | INSTALL_RKE2_TYPE="agent" INSTALL_RKE2_VERSION=v1.xx.yy+rke2zz sudo -E sh -
sudo mkdir -p /etc/rancher/rke2
sudo vi /etc/rancher/rke2/config.yaml

Create a /etc/rancher/rke2/config.yaml which will look something like:

node-name: leap-micro6
node-external-ip:
  - 10.3.6.55
node-ip:
  - 10.6.16.55
advertise-address: 10.6.16.55
# the following token can be found on 10.6.16.61 using
# sudo cat /var/lib/rancher/rke2/server/node-token
# comment this setting for the very first master node
token: K10<...>::server:<...>
# on all other servers after the initial setup on 10.6.16.61 is completed
server: https://10.6.16.61:9345
cluster-cidr: 10.42.0.0/16
service-cidr: 10.43.0.0/16
cni: calico
disable-kube-proxy: false
etcd-expose-metrics: false
etcd-snapshot-retention: 5
etcd-snapshot-schedule-cron: 0 */5 * * *
kube-controller-manager-arg:
  - cert-dir=/var/lib/rancher/rke2/server/tls/kube-controller-manager
  - secure-port=10257
kube-controller-manager-extra-mount:
  - >-
    /var/lib/rancher/rke2/server/tls/kube-controller-manager:/var/lib/rancher/rke2/server/tls/kube-controller-manager
kube-scheduler-arg:
  - cert-dir=/var/lib/rancher/rke2/server/tls/kube-scheduler
  - secure-port=10259
kube-scheduler-extra-mount:
  - >-
    /var/lib/rancher/rke2/server/tls/kube-scheduler:/var/lib/rancher/rke2/server/tls/kube-scheduler
kubelet-arg:
  - max-pods=250
kubelet-extra-mount:
  - >-
    /lib/modules:/lib/modules
node-label:
  - cattle.io/os=linux
protect-kernel-defaults: false

export PATH=$PATH:/opt/rke2/bin
sudo systemctl enable rke2-server.service
sudo systemctl start rke2-server.service
sudo journalctl -u rke2-server -f

export PATH=$PATH:/opt/rke2/bin
sudo systemctl enable rke2-agent.service
sudo systemctl start rke2-agent.service
sudo journalctl -u rke2-agent -f

As any service using local storage should implement restroing missing data themselves this describes how to create just the empty volumes/disks to do that.

As the local storage provisioner can not change the size of the volumes it will select the next larger volume for any claim. For example a 20 GB claim will select a 29.4 GiB volume, a 30 GB claim a 30.2 GiB volume etc.

For flatcar-linux we can follow the advice on the sig-storage-local-static-provisioner website: Mount formatted block storage on /mnt/local-disks/<UUID>. The UUID will be used to make sure mixing block devices will fail and not expose data to the wrong host.

1. Create a block device for an acdh-clusterX node in vCenter. Note that the size of the block device should be a little larger than the desired even number if GiB (example: for a 20 GiB volume create a 21 GiB disk) as there is a difference in how disk size is calculated
2. Format the volume on the respective flatcar node. Use ext4 or xfs depending on the needs of the service (for example elasticsearch/opensearch recommeds ext4)

sudo mkfs.ext4 /dev/sdd

1. reserved blocks for root are not very useful in kubernetes so set them to 0

sudo tune2fs -r 0 /dev/disk/by-uuid/<UUID>

1. Get the UUID. It is part of the output of mkfs.ext4. It is also for example available using using ls -l /dev/disk/by-uuid/*
2. Create a UUID directory in /mnt/local-disks/
3. Create a mount unit to mount the filesystem. The filename needs to match the mount point and is encoded.

sudo cp /etc/systemd/system/var-lib-rancher.mount "/etc/systemd/system/$(systemd-escape --path /mnt/local-disks/<UUID>).mount"
sudo vi /etc/systemd/system/"$(systemd-escape --path /mnt/local-disks/<UUID>).mount"
# change directory name and device name
# [Unit]
# Description=Mount local storage at /mnt/local-disks/<UUID>
# Before=local-fs.target
# [Mount]
# What=/dev/disk/by-uuid/<UUID>
# Where=/mnt/local-disks/<UUID>
# Type=ext4 or xfs
# [Install]
# WantedBy=local-fs.target
sudo systemctl daemon-reload
sudo systemctl enable "$(systemd-escape --path /mnt/local-disks/<UUID>).mount"

This is best done using the Rancher UI for cluster updates. If the version there and the version on the nodes get out of sync also all other settings cannot be changed anymore!. But for reference here is the very simple method of following the stable release channel for RKE2:

curl -sfL https://get.rke2.io | INSTALL_RKE2_CHANNEL=stable sudo -E sh -
sudo systemctl restart rke2-agent
# or
sudo systemctl restart rke2-server

Repeat on each node after the last one is showing as up and active in Rancher. Start with the management/server nodes then update the agent nodes.

Here you can see what version corresponds to stable at the moment. Kubernetes major versions are also channels. The channel latest refers to the very latest releases of K8s available.