cluster.yml¶
group_vars/all/cluster.yml describes the whole cluster. Change it, commit,
re-run.
Distribution and version¶
ocp_distribution: ocp # ocp = Red Hat OpenShift, okd = OKD (SCOS)
ocp_version: "5.0.0" # OpenShift: any z-stream on the mirror
okd_version: "5.0.0-okd-scos.6" # OKD: a tag from github.com/okd-project/okd/releases
cluster_name: ocp5
base_domain: lab.example.nl # API = api.ocp5.lab.example.nl
OpenShift (ocp) |
OKD (okd) |
|
|---|---|---|
| What | Red Hat OpenShift Container Platform | community distribution, same code base |
| OS on the nodes | RHEL CoreOS | CentOS Stream CoreOS (SCOS) |
| Pull secret | required | not needed (placeholder) |
| Tools from | mirror.openshift.com | github.com/okd-project/okd releases |
| Release image | quay.io/openshift-release-dev | quay.io/okd/scos-release |
| Support / updates | subscription, Red Hat support | community, stable OKD channel |
OKD tags ending in .ec.N are pre-releases. Switching distribution on an
existing cluster isn't supported: destroy.yml first.
Install method¶
agent boots every VM from an agent ISO; ipi lets the installer power and
install empty VMs through the Redfish BMC; external adds the Proxmox
cloud-controller-manager. Their settings are in group_vars/all/install.yml.
See Install methods.
Network¶
| Variable | Example | Meaning |
|---|---|---|
machine_network |
192.168.50.0/24 |
network of the node VMs |
gateway |
192.168.50.1 |
default gateway |
dns_servers |
[192.168.50.200] |
resolvers for the nodes |
dns_check_server |
10.0.3.98 |
DNS server the preflight asks (empty = local resolver) |
prefix_length |
24 |
prefix of machine_network |
api_vip / ingress_vip |
.10 / .11 |
VIPs (not used for SNO) |
cluster_network |
10.128.0.0/14 |
pod network |
service_network |
172.30.0.0/16 |
service network |
node_interface |
ens18 |
NIC name used in the agent network config (mapped by MAC) |
CNI: Calico Open Source¶
cni: Calico # or OVNKubernetes (the OpenShift default)
calico_version: "3.33.0"
calico_dataplane: Iptables # or BPF (replaces kube-proxy)
calico_encapsulation: VXLAN # or IPIP / None
The agent_iso role downloads the release's Tigera operator manifests and
hands them to the installer, so Calico runs from the first boot.
- Changing CNI or Calico settings needs a fresh cluster
(
destroy.yml, then deploy with-e agent_iso_rebuild=true). - OVN-only features (
EgressIP,EgressFirewall, OVN secondary networks) don't apply; use CalicoNetworkPolicy/GlobalNetworkPolicy. - VXLAN needs UDP 4789 between the nodes.
- Calico on OpenShift is supported through Tigera, not Red Hat.
Proxmox¶
| Variable | Default | Meaning |
|---|---|---|
proxmox_api_host / _port |
— / 8006 |
Proxmox API (ocp.sh uses 127.0.0.1 on the host) |
proxmox_api_user / _token_id |
ansible@pve / ocp |
API token; the secret is in vault.yml |
proxmox_validate_certs |
false |
check the API certificate (set true with a real cert) |
proxmox_iso_node |
first of proxmox_nodes |
node that receives the ISO |
proxmox_iso_storage |
local |
storage for ISOs |
proxmox_disk_storage |
local-zfs |
storage for VM disks |
proxmox_bridge / proxmox_vlan |
vmbr1 / "" |
network for the VMs |
Nodes — generated automatically¶
You don't list VMs by hand. Set counts, sizes and start values:
ocp_master_count: 3 # 1 = SNO, 3 = compact / standard
ocp_worker_count: 2
vmid_start: 5100
ip_start: 192.168.50.21
proxmox_nodes: [node02] # e.g. [pve01, pve02, pve03]
master_size: { cores: 8, memory: 24576, disk_gb: 120 }
worker_size: { cores: 8, memory: 16384, disk_gb: 120 }
| Field | Generated as |
|---|---|
| name | master-0..2, worker-0..N |
| VMID | vmid_start + n (masters first) |
| IP | ip_start + n |
| MAC | 52:54:00: + last 3 IP octets in hex — stable and unique per IP |
| Proxmox host | round-robin over proxmox_nodes |
| CPU / RAM / disk | master_size / worker_size |
| Masters / workers | Result | Platform |
|---|---|---|
| 1 / 0 | Single Node OpenShift | none |
| 3 / 0 | compact (masters schedulable) | baremetal + VIPs |
| 3 / N | standard | baremetal + VIPs |
Tweak one node with node_overrides (e.g. worker-1: { memory: 32768 }) or
list everything yourself in ocp_nodes_manual. The playbook refuses IPs that
land on .0/.255, a VIP or the gateway.
Other files¶
| File | What |
|---|---|
group_vars/all/autoscale.yml |
autoscaling settings |
group_vars/all/runner.yml |
runner LXC size, IP, VMID |
group_vars/all/vault.yml |
secrets, encrypted with ansible-vault |
group_vars/all/derived.yml |
computed values — don't edit |