Homelab

Motivation

A few things converged: multiplayer game servers cost real money to rent and the hardware you get is usually disappointing; streaming services kept raising prices while I had no ownership of anything I paid for; and I needed somewhere to store and process the terabytes of spatial data that come with GIS development work. Building a server solved all three at once.

Hardware

The server is headless - no monitor, no keyboard. SSH is the only way in. It runs continuously.

The 5800X3D’s 3D V-Cache makes it exceptional for cache-sensitive workloads. Factorio runs noticeably better than the clock speed alone would suggest. The GPU is too weak to assist Plex transcoding in any meaningful way, was simply included because the X3D chip doesn’t have integrated graphics.

Infrastructure

Containerisation: All services run in Docker, defined as Compose stacks and managed through Portainer. Stack definitions are version-controlled in a private GitHub repository. Secrets live in host-side env files excluded from Git and injected at deploy time.

Reverse proxy: Rather than manually maintaining proxy config, I use an nginx container that reads hostname and port environment variables off sibling containers and generates configuration automatically. Adding a new service to the LAN requires two environment variables and a restart - no manual nginx config involved - and an mDNS advertisement on the host.

Local DNS: I deliberately chose not to run a DNS server. A dedicated resolver is a network-wide single point of failure - if it goes down, nothing resolves. Instead, I use mDNS via Avahi to advertise .local hostnames on the LAN. Services like plex.local just work in any browser on the network with no client configuration required.

Service Stack

Backup

Media files and other re-downloadable content are excluded from backups - there’s no value in backing up what can be re-acquired. What gets backed up is configuration, databases, and irreplaceable personal data. Live databases require pre-dump snapshots rather than filesystem snapshots to avoid backing up mid-write state. I use Restic to create backup snapshots.

Destroying an entire RAID Array

After a power supply failed, I replaced it and learned mixing cables from modular power supplies, although the pins physically fit, will destroy equipment. I discovered this after frying an entire RAID array of disk drives during the drive shortage in 2026. That hurt to learn - and my backups were a year old! Fortunately I recovered data from most sources. I knew I needed to do backups more often, but I was lazy and the effort was high.

RAID protects against a drive failure. It doesn’t protect against human error, accidental deletion, or, apparently, mixing modular PSU cables. Who’d have thought?

Well, that pushed me to create a bash script run on every SSH connection to check if a backup was needed. If it has been a while since my last backup, I’m now prompted with a simple 31 days since last backup and Backup now? [y/n]. That mounts my external drive (or complains the drive is unplugged) and begins the backup sequence. This way, I don’t need to remember to do anything except occasionally SSH in (I do that anyway) and plug in the backup drive.

I’ve said for a long time for my career: Make it so doing the right thing is the easy thing to do. I’m no exception to the rule.