petal-power/wiki-vault
3
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
main
wiki-vault/Sources/Homelab/homelab-ansible.md

5.8 KiB
Raw Permalink Blame History

created path project tags type
2026-05-26 Sources/Homelab homelab-ansible
ansible
proxmox
pbs
traefik
knot
docker
project-plan

Goal

homelab-ansible is the service + backup infrastructure repo for the PVE-hosted homelab. Today (v0.1.0) it configures the Traefik + DNS slice. This plan captures its expanding role coming out of the 2026-05-25 Ansible-repo reorg:

  1. Absorb the PBS backup stack as infrastructure-as-code (the backups already run and are restore-verified — this is IaC capture, not new infra).
  2. Adopt the new homelab-common shared-roles library to kill cross-repo drift.
  3. Coordinate two sibling cleanup tasks that fell out of the reorg (PBS codification here; CLI strip on the NAS).

This plan references the current repo state and the sibling projects; the live decision log lives in trellis thread #580 (herbys-dev-setup / project homelab-dev-ansible).

Locked Decisions

  • Repo topology (decided 2026-05-25): four Ansible repos, scoped by domain, sharing a thin baseline:
    • homelab-ansible (this) — service + backup infra.
    • homelab-dev-ansible (new) — herbys-dev dev workbench VM only (herbys-dev-setup).
    • herbylab-nas (GitHub trucktrav/herbylab-nas) — bare-metal NAS storage stack.
    • workstation-ansible (Gitea petal-power, formerly cli-standardization) — owns the CLI/shell concern (Fish/Starship/dotfiles) across machines (cli-standardization).
    • homelab-common (new) — shared baseline roles, consumed via requirements.yml.
  • Hosts owned by homelab-ansible: Knot DNS (LXC 103, 10.0.11.10), Pi-hole (LXC 104, 10.0.11.50), Traefik VM (VM 102, 10.0.21.181) + cloudflared + whoami, and PBS (LXC 110).
  • PBS lives here, not in homelab-dev-ansible (corrects an earlier draft that started PBS in the dev repo).
  • homelab-common is thin: baseline OS (timezone/locale/users/sudo/SSH keys/apt + unattended-upgrades/base packages) + docker-engine install + (optional) msmtp alert role + (optional) version_check. Shell, storage, service stacks, and docker-deploy patterns all stay domain-specific. Naming convention: noun-role style (homelab-ansible's), not the NAS verb-prefix style.
  • Adoption is deliberate, not big-bang: homelab-dev-ansible is homelab-common's first consumer (greenfield); homelab-ansible and herbylab-nas retrofit onto it as scheduled one-PR-per-repo follow-ups, when next touched.
  • Repo name homelab-ansible unchanged.

Open Items

  • Pick the canonical baseline for homelab-common: homelab-ansible's common vs the NAS manage_linux — they have already diverged, so this is a choice, not free extraction.
  • Confirm the seed-from-greenfield → deliberate-retrofit sequencing (proposed; pending Travis's explicit nod given his drift concern).
  • PBS codification must reflect reality: the off-box leg is rsync-on-a-timer to the NAS, NOT the originally-planned PBS native sync-job (NFS/SMB both failed against PVE+PBS).

Phases

Sibling task 1 — Codify PBS into homelab-ansible (IaC capture)

The backups already RUN and passed a two-leg restore test: PBS in LXC 110, datastore main/pbs-main on a single-disk ZFS pool on sda; PVE vzdump for VM 100 at 02:30 + 22:30; PBS prune-job pj-main at 03:00 (keep-daily 7 / weekly 4 / monthly 6); rsync-on-timer copying the datastore to the NAS. A proxmox-backup role is already on homelab-ansible's own v0.2.0 roadmap.

  • Write proxmox-backup role: PBS LXC + datastore, vzdump job, prune schedule, rsync-to-NAS timer.
  • Bake in the operational gotchas: rsync timer must fire in a QUIET window clear of the 02:30/22:30 backups and the 03:00 prune; run an occasional PBS verify on the source datastore (rsync does not verify chunk integrity in flight).
  • Reconcile with the locked backup decisions; record that off-box transport = rsync, not PBS sync.
  • Validate idempotent run against the live PBS (zero changes on second pass).

Sibling task 2 — Strip CLI out of herbylab-nas

manage_zsh (Oh-My-Zsh + Powerlevel10k) currently lives in herbylab-nas, but shell/CLI is workstation-ansible's domain. Remove it from the NAS repo.

  • Remove the manage_zsh role and its play wiring from herbylab-nas.
  • Confirm nothing else in the NAS playbook depends on it.
  • (Optional) If the NAS shell should be managed at all, add the NAS host to workstation-ansible's inventory rather than re-homing it in the NAS repo.

homelab-ansible's own homelab-common retrofit

  • Once homelab-common exists and is proven on homelab-dev-ansible, replace homelab-ansible's hand-rolled common with the shared baseline role (one focused PR).

Notes

  • Current repo state: homelab-ansible v0.1.0 (Gitea petal-power, owner herbyadmin, on disk /opt/projects/homelab/homelab-ansible). Roles: common, docker, compose_stack, knot, pihole, traefik, cloudflared, whoami. Terraform: bpg/proxmox modules (proxmox_vm, proxmox_lxc) + envs/homelab. Docs: README, CHANGELOG, ARCHITECTURE, REBUILD, RESTORE, SECRETS, CONVENTIONS. v0.2.0 roadmap already lists the deferred proxmox-backup role, inventory auto-render from Terraform output, and CI.
  • Related plans: herbys-dev-setup (the dev box → homelab-dev-ansible; trellis #580), repo-hygiene-remediation (trellis #272 — herbys-dev is its Phase-0 hard gate), cli-standardization (→ workstation-ansible).
  • Cross-remote sharing: herbylab-nas is on GitHub (trucktrav); homelab-ansible and workstation-ansible are on Gitea (petal-power). requirements.yml accepts git URLs, so homelab-common can be consumed across both remotes.
  • Cloudflare: pinned / out of scope for now.
  • Control node for push-model runs: deferred — repo is pushed off-prem, so clone-and-run on whatever target machine is chosen at the time.