Requirements¶
Proxmox VE¶
- Proxmox VE 8.x or 9.x.
- x86-64-v3 CPUs (AVX2) on every host — RHCOS 10 needs it. The VMs use
cpu: host. -
An API token for Ansible.
./ocp.sh tokencreates it for you on the host, or by hand: -
ISO storage that allows content
iso(localby default). With VMs spread over several hosts it must be shared, or all VMs must run onproxmox_iso_node. - A bridge for the cluster network (
proxmox_bridge, e.g.vmbr1). For autoscaling the Proxmox host (or the runner LXC) also needs an IPv4 address on that network.
DNS¶
Checked by the playbook before anything is built:
| Record | HA / compact | install_method: external |
SNO |
|---|---|---|---|
api.<cluster>.<domain> |
api_vip |
external_lb_address |
master-0 IP |
api-int.<cluster>.<domain> |
api_vip |
external_lb_address |
master-0 IP |
*.apps.<cluster>.<domain> |
ingress_vip |
external_lb_address |
master-0 IP |
No wildcard on the cluster domain itself
A record like *.<cluster>.<domain> — or a parent wildcard such as
*.<domain> at your DNS provider — makes the installer refuse to start
(dns-wildcard-not-configured). The preflight catches this. A wildcard on
*.apps.<cluster>.<domain> is fine. See
Troubleshooting.
Autoscaled workers need no DNS records: they get their name from DHCP and
an /etc/hosts MachineConfig.
Pull secret¶
OpenShift (ocp) needs a pull secret from
https://console.redhat.com/openshift/install/pull-secret. OKD doesn't.
Sizing¶
| Role | Minimum | Default in cluster.yml |
|---|---|---|
| master | 4 vCPU / 16 GB / 120 GB | 8 vCPU / 24 GB / 120 GB |
| worker | 2 vCPU / 8 GB / 120 GB | 8 vCPU / 16 GB / 120 GB |
| SNO | 8 vCPU / 16 GB / 120 GB | — |
Controller¶
Only if you don't use ocp.sh on the Proxmox host (which installs all of
this itself): Linux x86_64 with
ansible-core≥ 2.16,python3-proxmoxer,python3-requests,python3-requests-toolbelt(ISO upload > 256 MB)nmstate(nmstatectl) — the agent installer validates static network config with itdig, ~10 GB free in.work/