Roadmap
The focus is Foreman with Smart Proxies for provisioning and Remote Execution. Katello follows later. Stages are intended goals and may change. Open work is tracked in GitHub issues.
Done
| Stage |
Content |
State |
| Spike |
Manual manifests on k3s: dnsmasq next to an unprivileged proxy on Multus, UBI 9 images |
A test client got a lease and a boot file, see Testing |
| Foreman |
Foreman with the frontend sidecar, Dynflow, cert-manager, CloudNativePG, migration before rollout |
Released in 0.1.0 |
| Provisioning |
SmartProxy profile Provisioning: registration, DHCP and TFTP through dnsmasq, DNS through nsupdate, BMC |
Released in 0.1.0; E2E on a fake VLAN |
| Execution |
Profile Execution with Remote Execution and Ansible over SSH, hosts queue |
Released in 0.1.0; E2E runs both job types |
| Plugin catalog |
Activation through a migration; Remote Execution, Ansible, Templates, Webhooks, and Puppet |
Released in 0.1.0 |
| More plugins |
Boot Disk with its tools, Discovery with the proxy feature, KubeVirt, Proxmox, VMware |
On develop |
| Recurring tasks |
CronJobs for cron:hourly to cron:monthly |
On develop |
| Upgrades |
Preflight, fence, migration, roll forward guard, smoke checks; upgrade E2E from 0.1.0 |
On develop |
| Restore |
Restore order documented and drilled with a logical dump |
On develop |
| Releases |
semantic-release, signed images and charts in GHCR, SBOMs, source references |
Done |
| Version tracking |
Pinned Foreman and plugin RPMs, tracked by Renovate |
On develop |
Towards a beta
A beta means: a real host is built, the operator runs on OpenShift, and an operator can run it without reading the code. These items are needed for that, roughly in this order.
| Item |
Why |
Issue |
Release 0.2.0 |
Ships upgrades, recurring tasks, the plugin catalog, and the restore order, which are only on develop |
|
Sign develop images, publish the documentation |
The repository is public, so the reasons for sign: false and the manual docs workflow are gone |
|
| Scheduling for Smart Proxies |
nodeSelector, tolerations, and affinity. A provisioning proxy must run on a node that has the VLAN trunked |
|
| A complete PXE build |
No host has been built end to end. A KubeVirt VM on the E2E cluster, or a real VLAN in a homelab |
#38 |
| OpenShift qualification |
A dedicated SCC for dnsmasq, an insecure Route for /unattended, the E2E test on OpenShift |
#12 |
| E2E in CI |
The E2E tests run by hand on one shared cluster. A Kind job for Foreman and the Execution proxy, without Multus, would catch regressions on every pull request |
|
| Foreman 5.1 |
The first minor upgrade through a Renovate pull request and the upgrade flow, once Foreman 5.1 is released |
|
Operations
Part of the original concept or found missing while testing, not implemented yet. Each can be added without changing the API in an incompatible way.
| Item |
Purpose |
| PodDisruptionBudget for web |
Keeps one web pod during node drains |
| Default NetworkPolicy |
Limits Foreman and the proxies to the expected peers. Puma no longer needs it, since it has no TCP listener |
| Metrics |
Migration phase and duration, upgrade blocks, registration errors; a ServiceMonitor in the chart |
| Valkey authentication and TLS |
Needed for shared or managed Valkey instances |
| Backups in the database chart |
ScheduledBackup with the Barman Cloud plugin as optional values of foreman-db-cnpg |
| Off-cluster Smart Proxies |
Export a certificate bundle for a proxy outside the cluster, like the auth-bundle command of foremanctl |
Release |
A kind for compatible image sets of Foreman, frontend, and Smart Proxy, and the proxy upgrade after Foreman |
Later
| Item |
Notes |
| Migration from existing installations |
Plan in Migration; rehearsal against a production-sized copy |
| Katello |
Candlepin, Pulp, /pub (#33), and content proxies, see Katello |
| More Smart Proxy providers |
Infoblox for existing DHCP, dns_nsupdate_gss with a keytab, Remote Execution pull mode over MQTT |
| External authentication |
Kerberos or SAML in the frontend |
| More plugins |
foreman_openscap once scap-security-guide is available |
Comparison with the concept
The original design proposal for this operator, written as input for the foreman-kubernetes RFC, differs from the implementation in these points:
| Concept |
Implementation |
Reason |
Separate Frontend kind with its own replicas |
Apache as a sidecar in the web pods |
Puma has no TCP listener, so no pod can forge the client certificate headers (#7) |
| Dynflow only when a plugin needs it |
Orchestrator and one worker always run |
Foreman core uses Dynflow as its ActiveJob backend |
| ISC DHCP with OMAPI |
dnsmasq with smart_proxy_dhcp_dnsmasq |
ISC DHCP is end of life and not in UBI 9; one subnet per proxy as a consequence |
| PostgreSQL from CloudNativePG or external |
CloudNativePG only |
One path to test. An external path can be added later |
| Platform detection with a manual override |
Detection only |
No case needed an override yet |
| Smoker tests on Kind, k3s, and OpenShift |
Chainsaw tests on one k3s cluster |
Smoker targets a Katello installation. Kind and OpenShift are open, see above |
HPA, PDB, and NetworkPolicy from Foreman |
Not implemented |
See Operations |
| Upstream images |
Own UBI 9 images |
Random UID support and the provisioning tools were missing upstream |
Principles
- Spike before operator. Whether DHCP and TFTP work in a pod on Multus was the largest unknown. The spike answered it before the API was fixed.
- Check OpenShift behavior early. The operator never sets a UID for Foreman, frontend, or proxy containers, so they run under the UID the platform assigns.
- Test on a cluster. A feature counts as done when an E2E step passes on the test cluster, not when the controller tests pass.
Open decisions
- Whether the operator should stay separate from
foreman-kubernetes, or be proposed there
- Whether KubeVirt may be installed on the shared E2E cluster for #38
- Whether
Release is still needed once the chart pins all images to one version