Files
notes/docs/home-server/ideas-next-steps.md
T
2026-10-26 12:10:24 +00:00

230 lines
12 KiB
Markdown

---
schema_version: 1
id: home-server.ideas-next-steps
type: plan
scope: [jupiter, mercury, private-access, recovery, downloads, power, services]
sensitivity: private-infrastructure
last_reviewed: "2026-10-26"
sources:
- kind: owner-report
reference: "AirVPN downloads live; GBP 56.06 three-year plan ending 2029-10-25; Ultra.cc/rclone retired; imports/hardlinks/seeding/removal and public UI confirmed; full isolation checks and backups outstanding, 2026-10-26"
- kind: owner-report
reference: "Access, device-loss recovery, hardware constraints, broadband and VPN preferences interview, 2026-10-22"
- kind: documentation
reference: "Planning discussion based on jupiter-stacks and home-server reference notes, 2026-10-22; proposals only, no live inspection or deployment"
related:
- home-server.reference
- home-server.networking
- home-server.authentication
- home-server.backups
- home-server.operations
- home-server.seedbox
- home-server.downloads
- home-server.storage
- home-server.host
update_triggers: [implementation, priority-change, access-requirements-change, pricing-change, power-measurement]
unknowns:
- live-routing-firewall-and-wireguard-configuration
- work-laptop-local-dns-and-lan-access-policy
- measured-wall-power-and-electricity-tariff
- airvpn-renewal-terms-and-complete-isolation-results
- immich-original-photo-backup-coverage
---
# Ideas / next steps
Planning backlog from the 2026-10-22 discussion, updated with the owner's
2026-10-26 AirVPN cutover confirmation. Unchecked proposals remain
**recommendations, not implemented changes**; the completed downloads work
below is owner-reported deployed state.
## Requirements confirmed with the owner
| Area | Requirement / preference |
| --- | --- |
| Private administration | Portainer, Beszel, Dozzle, Backrest, Prowlarr, Radarr and Sonarr should stop being publicly reachable |
| Public applications | Keep normal user-facing applications public, including Vaultwarden |
| Personal devices | Owner-only admin access; VPN clients or certificates can be installed on personal phone/laptop |
| Work laptop | Browser-only admin access needed on home Wi-Fi; cannot install VPN clients or certificates |
| Device loss | Vaultwarden access remains essential; admin tools can wait until returning home |
| Vaultwarden recovery | Owner knows the master password; TOTP is backed up/restorable through Microsoft, with independently recoverable Microsoft access; owner report, not an observed recovery exercise |
| Private-access tooling | Prefer minimal infrastructure and no vendor-hosted coordination service; existing Cloudflare DNS/SSH use is acceptable |
| Hardware | Software/BIOS tuning only; no hardware purchases or removals |
| Broadband | Reported 500 Mbps download and upload; public trackers only |
| Download VPN | AirVPN selected and live; three-year plan purchased for GBP 56.06, expires Thursday, 25 October 2029 |
| New services | Interested in reliability, everyday tools, media, development and privacy |
## Priority order
1. Prepare independent recovery access, then introduce private admin ingress.
2. Make backups recoverable and add failure notifications.
3. Finish download-VPN isolation checks and recovery coverage; seedbox replacement is complete.
4. Measure and tune power use.
5. Add services that solve an actual day-to-day problem.
## Private admin access
### Preferred approach: plain WireGuard through Mercury
Reuse Mercury's public address and existing WireGuard connection to Jupiter.
Give each personal device its own peer/key, narrowly scoped routes and firewall
permissions. No Tailscale account or self-hosted coordination service is needed.
| Client | Intended access path |
| --- | --- |
| Work laptop on home Wi-Fi | LAN directly to private admin ingress; browser only |
| Personal device away from home | WireGuard to Mercury, then routed private access to Jupiter |
| Public internet | Mercury nginx to explicitly public applications only |
Use split DNS: admin hostnames resolve to private addresses on the LAN and
through the VPN. Keep HTTPS certificates using the existing DNS-based challenge;
certificate issuance does not require a public website.
Before choosing exact addresses/listeners, inspect live WireGuard `AllowedIPs`,
routing, firewall rules and Docker port publication. Check that the work laptop's
corporate DNS/proxy policy permits local access; do not bypass employer controls.
### Implementation checklist
- [ ] Prepare recovery access and a rollback route before restricting ingress.
- [ ] Add per-device WireGuard peers and a way to revoke a lost device's key.
- [ ] Configure LAN/VPN DNS for the private admin hostnames.
- [ ] Create private-only Traefik ingress; public requests must not reach admin
backends simply by supplying their hostnames.
- [ ] Replace or constrain Mercury's wildcard forwarding in
`jupiter-stacks/mercury/nginx/umbra.mom`, retaining all intended public apps.
- [ ] Apply the same boundary to Authentik's separate outpost callback routes
for Backrest and the Arr applications.
- [ ] Remove or restrict Portainer's direct `9123:9000` publication in
`stacks/portainer/portainer_start.sh`.
- [ ] Retain application authentication and Authentik MFA behind the private
boundary; VPN connectivity must not imply application administrator rights.
- [ ] Confirm private apps and authentication callbacks work from LAN and VPN,
but cannot be reached through public ingress or alternative published ports.
- [ ] Confirm public applications, especially Vaultwarden, remain accessible.
Removing public DNS records or adding a login page is not sufficient isolation.
The network/proxy boundary must enforce the restriction.
### Alternatives
| Option | Assessment |
| --- | --- |
| Plain WireGuard | Preferred: reuse existing infrastructure, few devices, no hosted control service |
| Tailscale | Convenient, but adds the vendor dependency the owner prefers to avoid |
| Headscale | Self-hosted coordination, but more infrastructure to maintain than currently needed |
| mTLS | Device-certificate authentication; adds certificate distribution/recovery and cannot be used directly on the work laptop |
mTLS rejects clients without a certificate but leaves the HTTPS endpoint
internet-facing. Mercury terminates public TLS, so browser client-certificate
enforcement would need to happen there, not merely between Mercury and Jupiter.
## Recovery and reliability
Keep Vaultwarden public with application login and MFA. Do not make access
depend on the new admin VPN or Authentik.
- [ ] Maintain an encrypted recovery pack outside Jupiter: Mercury provider
console/SSH access, WireGuard re-enrolment instructions, Authentik recovery,
retained local admin credentials, backup credentials, restic key, stack
secrets and essential host configuration.
- [ ] Ensure retrieving/decrypting that pack does not depend on Jupiter,
Portainer or secrets available only inside Vaultwarden.
- [ ] Retain the existing Cloudflare SSH route during migration, with its
effective Access policy reviewed. It still depends on Jupiter's connector
and networking; it is not an out-of-band console.
- [ ] Restore an important service into an isolated location and record the
dependencies, time taken and acceptable data loss.
- [ ] Review backup coverage for Backrest configuration, host configuration,
Docker named volumes and Mercury.
- [ ] Establish whether Immich contains irreplaceable originals and where those
originals and the associated database are backed up.
- [ ] Make failed backup restart hooks visible; a successful snapshot must not
hide stopped Authentik, Portainer or Wardrowbe services.
- [ ] Add service, disk-health, capacity and backup-failure notifications using
Beszel and the existing Mercury uptime service where suitable.
- [ ] Add an external missed-backup heartbeat so Jupiter going offline cannot
prevent notification of its own failure.
Private connectivity does not fix Authentik downtime. The documented backup plan
stops Authentik and ignores some restart failures; independent SSH/local recovery
remains necessary. See [Backups](backups.md) and [Operations](operations.md).
## Local downloads instead of the seedbox
**Completed by owner report on 2026-10-26:**
[local qBittorrent/Gluetun downloads through AirVPN](downloads.md) are live.
This is both a **cost saving** and an **architecture simplification**: local
storage and hardlink imports replace the remote seedbox and SFTP/FUSE dependency.
The purchased **three-year AirVPN plan cost GBP 56.06** and expires
**Thursday, 25 October 2029**. Amortised over 36 months, that is approximately
GBP 1.56/month versus Ultra.cc's former GBP 4.50/month: a subscription saving of
**GBP 35.31/year**, or **GBP 105.94 over three years**, before extra electricity.
The former GBP 2.50/month VPN estimate is superseded by this actual purchase.
### Completed work and remaining follow-up
- [x] Deploy local qBittorrent behind Gluetun/AirVPN.
- [x] Confirm encrypted VPN egress uses the home ISP directly, not Mercury.
- [x] Confirm Sonarr and Radarr imports, hardlinks and intended seeding/removal.
- [x] Confirm public Web UI through Mercury/Traefik/Authentik plus native
qBittorrent login; this is an intentional change from the private-only proposal.
- [x] Retire Ultra.cc, handle retained files and remove the rclone-seedbox
service and FUSE mount.
- [ ] Complete peer-ingress, DNS/IPv6 leak and VPN-failure/startup/reboot checks.
- [ ] Add qBittorrent state/defaults and AirVPN secrets to encrypted off-host backups.
A hardlink gives the same file data two names without storing a second full
copy. Hardlinks are confirmed for both import workflows; they are not backups.
The separate private-admin-access proposal above remains outstanding.
## Electricity and compute efficiency
Reported average CPU usage is approximately 1%, but CPU utilisation does not
measure wall-socket power. No power measurement or actual electricity tariff
was supplied.
Illustrative costs at **25p/kWh**, using 365 days/year divided by 12:
| Continuous average draw | Approximate monthly cost |
| --- | --- |
| 40 W | GBP 7.30 |
| 60 W | GBP 10.95 |
| 80 W | GBP 14.60 |
At that illustrative tariff, saving 10 W continuously saves about GBP 18/year.
An extra average draw of about 16 W would consume the GBP 35.31/year subscription saving.
Count the incremental download workload, not the server's existing baseline.
- [ ] Measure wall power and obtain the actual unit rate when practical.
- [ ] Inspect CPU idle states and frequency-scaling policy before BIOS tuning;
low CPU limits do not automatically save energy.
- [ ] Inspect RAM pressure and swapping during Immich processing and Wardrowbe
Ollama jobs; the host has 16 GB RAM.
- [ ] Schedule heavy jobs where useful and avoid unnecessary repeated scans.
- [ ] Review effective container log rotation; daemon defaults require activation
and recreation, and explicit Compose settings override them.
- [ ] Avoid aggressive HDD spindown while `/data` hosts active databases.
- [ ] Avoid consolidating unrelated application databases solely to reduce
container count; shared failures/upgrades may outweigh uncertain savings.
No hardware changes are proposed. Leave spare CPU capacity idle unless a useful
workload justifies the added power and maintenance.
## Optional service ideas
These are candidates, not a deployment list.
| Area | Idea | Constraint / reason |
| --- | --- | --- |
| Reliability | Notifications and external backup heartbeats | Start with existing monitoring rather than a second metrics stack |
| Privacy | AdGuard Home or Pi-hole | Choose one; household DNS availability becomes an operational dependency |
| Files | Syncthing for selected folders | Sync is not backup; check client support on the owner's devices |
| Bookmarks / reading | Linkding; FreshRSS | Small tools with a clear everyday purpose |
| Development | Renovate automation for existing Gitea workflows | Propose updates; avoid unattended production upgrades |
| Media | Profilarr for supported Arr profiles/custom formats | Add only if manual profile maintenance is a real problem |
Defer Kubernetes, a large metrics stack, another identity provider or a general
cloud suite unless a concrete need emerges.