How it works¶
Install flow¶
flowchart LR
C[cluster.yml] --> T["templates<br/>install-config + agent-config"]
T --> I["openshift-install<br/>agent create image"]
I --> ISO[agent ISO]
ISO -->|proxmox_template| S[Proxmox ISO storage]
S -->|proxmox_kvm| V[VMs boot the ISO]
V --> W["agent wait-for<br/>install-complete"]
- Preflight — Proxmox API and token, node names, DNS records, no DNS wildcard.
- Tools —
openshift-installandocare downloaded and verified against the mirror'ssha256sum.txt(OKD: GitHub releases). - Agent ISO —
install-config.yamlandagent-config.yamlare rendered fromcluster.yml(static IPs via nmstate, NICs mapped by MAC); for Calico the Tigera manifests are added. The ISO is cached unlessagent_iso_rebuild=true. - Upload + VMs — UEFI, q35,
cpu: host, virtio-scsi, boot orderscsi0thenide2: on first boot the disk is empty so the agent ISO starts; afterwards RHCOS boots from disk. - Wait —
agent wait-for bootstrap-complete, theninstall-complete; prints the console URL and kubeadmin password.
Notes¶
-
Name — the project was called
openshift5before it became Shiftmox (OpenShift + Proxmox). Identifiers that live in your cluster keep the old prefix on purpose: the labelsopenshift5/role/openshift5/cluster, the MachineConfig99-worker-openshift5-hostsand the/etc/hostsblock markers. Renaming them would make the playbook lose track of hosts that a worker runs on and leave duplicate MachineConfigs behind. -
Idempotent — VMs are only created if their VMID doesn't exist; the ISO only if missing. Changing the node list of an installed cluster isn't handled: add workers with autoscaling,
oc adm node-image create, or rebuild. - Install state lives in
<ocp_state_dir>/<cluster>/build/(.work/locally,/var/lib/ocp-deploy/in CI). Keep it until install-complete;agent wait-forneeds it. Rendered configs are kept in.work/<cluster>/config/for review. - Mirror path —
ocp_mirror_basepoints atopenshift-v5/...; check the exact path for your version on mirror.openshift.com. - Storage — none is configured. For a lab, LVM Storage (SNO) or the NFS subdir provisioner is quickest; ODF needs 3 extra disks per node.
- Boot image vs cluster OS — new workers start from the release's boot image (e.g. RHCOS 9.8) and update themselves to the cluster OS (e.g. 10.2) on first boot, with one reboot.
Redfish BMC for Proxmox¶
roles/proxmox_redfish/files/proxmox_redfish.py is a system driver for the
OpenStack sushy-tools Redfish
emulator:
| Redfish | Proxmox |
|---|---|
Systems |
VMs named worker-auto-N.<cluster> + VM-less slots |
| power On / ForceOff / GracefulShutdown / reset | status/start, stop (overrule-shutdown), shutdown, reset |
| boot source override | boot: order=… |
| boot mode UEFI / Legacy | bios: ovmf / seabios |
| VirtualMedia InsertMedia | download ISO → upload to proxmox_iso_storage → ide2 |
| EthernetInterfaces, memory, CPUs, disks | VM config (or the slot definition) |
It runs under gunicorn with one worker process (the VM cache, the list of just-created VMs and the reaper live in-process) and 8 threads.