Skip to content

GitHub Actions

Deploy and destroy from a workflow, on a self-hosted runner in an LXC on the Proxmox host. The deploy has to reach the cluster nodes on the private bridge (port 8090 on master-0 during install, 6443 after), so a GitHub-hosted runner can't do it. The runner only makes outbound HTTPS connections to GitHub: nothing has to be opened.

flowchart LR
    GH[GitHub] <-- outbound HTTPS only --- R
    subgraph host[Proxmox host]
        R["runner LXC<br/>192.168.50.5"]
        API[Proxmox API :8006]
        N["cluster nodes<br/>:6443 / :8090"]
    end
    R --> API
    R --> N

Keep the repo private

On a public repo anyone could open a pull request that runs code on your runner.

1. Prerequisites

  • Root SSH from your machine to the Proxmox host, and paramiko (pip install paramiko) — Ansible configures the LXC with pct exec through the host, so the container needs no SSH server.
  • The host's address in inventory/hosts.yml (pve01).
  • Runner settings in group_vars/all/runner.yml:
Variable Default
runner_lxc_vmid 5099
runner_lxc_ip 192.168.50.5 (on machine_network, outside the node range)
runner_lxc_cores / _memory / _disk_gb 2 / 4096 / 40
runner_lxc_template_pattern newest Debian 13/12 standard template

2. Create the LXC and register the runner

# token: repo › Settings › Actions › Runners › New self-hosted runner
ansible-playbook runner.yml --ask-vault-pass
# or: export GITHUB_PAT=...  (fine-grained, Administration: read/write on this repo)

This creates an unprivileged LXC (nesting on, starts on boot) on the cluster network and installs git, python3-venv, nmstate, dig, a ghrunner user, the latest Actions runner as a systemd service (labels self-hosted, linux, proxmox) and /var/lib/ocp-deploy for cluster state. Re-running is safe.

3. Secrets

Repo › Settings › Environments › New environment proxmox:

Secret Value
PROXMOX_API_TOKEN_SECRET the Proxmox API token secret
OCP_PULL_SECRET your pull secret (JSON) — not needed for OKD
OCP_SSH_PUBKEY public SSH key for the core user
ANSIBLE_VAULT_PASSWORD optional, only with a committed encrypted vault.yml
REDFISH_BMC_PASSWORD for the autoscale stage: same as the BMC's redfish_bmc_password

Tip

Add yourself as required reviewer on the environment to get an approval click before every deploy or destroy.

4. Run

Actions › Deploy OpenShift on Proxmox › Run workflow:

Input Use
action = deploy, stage = all full install (~1–2 h, job timeout 4 h)
stage = iso / vms only that stage
stage = wait re-attach to an install that is running
stage = autoscale autoscaling, cluster part (see below)
rebuild_iso = true after changing cluster.yml before the VMs exist
action = destroy, confirm = <cluster_name> delete VMs, ISO and state

Autoscale stage: the runner user can't install system services, so install the Redfish BMC once from your machine (ansible-playbook autoscale.yml --tags bmc) and store its password as REDFISH_BMC_PASSWORD.

Notes

  • Deploys never start from a push. lint.yml (syntax check, ansible-lint, ruff, shellcheck) runs on every push/PR on GitHub-hosted runners, without secrets. docs.yml builds and publishes this site.
  • Cluster state, kubeconfig and kubeadmin password stay on the runner in /var/lib/ocp-deploy/<cluster>/; they're never uploaded as artifacts: sudo -u ghrunner oc --kubeconfig /var/lib/ocp-deploy/<cluster>/kubeconfig get nodes
  • Only one deploy runs at a time (concurrency group).
  • Lock down the API: restrict 8006 on the Proxmox firewall to your own IPs plus the runner. If the runner can't reach the public API name (no hairpin NAT), set a repository variable PROXMOX_API_HOST with an address that works from the LXC.