VM templates

Every cluster's nodes are clones of a Proxmox VE VM template. kubehenge template create makes the template a distribution needs, through the Proxmox VE API alone: no SSH to the Proxmox VE host and no build tools beside the kubehenge command. In the appliance, administrators build and delete them on the Administration page instead, with their own session (install.md); this page describes the command and what both do.

Distribution	Template	One per
Talos	the Talos Image Factory's image with the QEMU guest agent extension; made by the command only, ahead of Talos clusters, which no ClusterClass offers yet	Talos release
RKE2, k3s	Ubuntu's cloud image with every update and the QEMU guest agent; the distribution installs itself when a node boots	Ubuntu release
kubeadm	the same, plus containerd, the kubelet and kubeadm from pkgs.k8s.io as Kubernetes' installation guide sets them up, with the control-plane images pulled ahead	Ubuntu release and Kubernetes version

Ubuntu images are checked against Ubuntu's signed checksums before Proxmox VE downloads them, and the pkgs.k8s.io signing key is accepted only with its known fingerprint.

Create

export KUBEHENGE_PROXMOX_URL=https://pve.example.org:8006
export KUBEHENGE_PROXMOX_TOKEN='root@pam!kubehenge=...'
kubehenge template create talos --node pve1
kubehenge template create rke2 --node pve1 --storage shared-storage
kubehenge template create kubeadm --kubernetes-version v1.37.1 --node pve1

Ubuntu-based templates are prepared by booting the image once. That VM needs to reach Ubuntu's and Kubernetes' package mirrors: it uses DHCP on --bridge, or a static address given with --build-address, --build-gateway and --build-dns. Its cloud-init data comes on an ISO uploaded to --iso-storage (the node's first ISO storage by default) and removed afterwards.

Creating is idempotent: a template with the same tags in the template pool is found and reused. One with those tags in another pool is not, since clusters cannot clone from it. A failed run removes what it made; --keep-on-failure leaves the VM for inspection, and a later run refuses to start until it is removed.

How clusters find them

Each template carries a unique set of Proxmox VE tags, such as kubehenge;talos;talos-v1.14.4, and Cluster API's Proxmox VE provider selects it by that exact set. The template has no network card, one core and 512 MiB: each clone gets the network, size and disk its machine asks for.

Templates should be on shared storage in a Proxmox VE cluster, so that clones can start on any node.

Requirements

• The storage given with --import-storage (local by default) must allow the import content type.
• The API token needs to download to that storage, create VMs in the template pool (--pool, kubehenge-templates by default) and write to the ISO storage.

List and delete

kubehenge template list
kubehenge template delete kubeadm --kubernetes-version v1.37.1

Both work in the template pool, --pool as for create; --pool '' takes in every pool.