It’s been about eight years since I started running network services in my home. It all
started with Pi-hole, which I ran not on a Raspberry Pi, but an Atom-powered Dell Mini
“thin client” I acquired from eBay. Today I run a lot more and it’s become what is now
known as a Homelab.
I’m quite fond of it all, so thought I might as well write about it.
NAS
I built my NAS around five years ago. At the time my PC was in a Fractal Design Node 804
case, but I decided this was much more suitable for use as a NAS, so I transplanted my
PC into a smaller case and used the Node for my NAS. The specs are as follows:
Currently, “AI agents” like Claude Code are all the rage for software development.
These programs facilitate feeding context into LLMs directly from your codebase and
environment by empowering the LLM to read/write files and run arbitrary code on your
system.
If you’re at all security conscious, the above should make you very uncomfortable. An
intentionally non-deterministic model running arbitrary code and reading arbitrary
files? No thanks. What’s worse is programs like Claude Code are not even open source. As
a rule I go to lengths to avoid proprietary software, but I decided to cave on Claude
Code just to see what all the fuss is about.
The SQL standard includes a boolean type which, unsurprisingly, represents a Boolean
value. This can be useful, but I’m going to argue that it should almost never be used in
a normalised data model.
The Problems with Boolean
Here are few of the problems you’ll inevitably run into if you use boolean in a
normalised data model:
It’s not really binary
So you’re chatting to your domain expert and he says something like “books are either
hardback or paperback”. Something twigs in your mind: two options… binary… that’s a
Boolean! So you duly write:
If you are thinking about using an LLM to communicate with me, you will be glad to hear
I’ve just developed an exciting new technique which will save us both time. Rarely are
such wins possible, so this is surely cause for a celebration.
The technique goes as follows:
Type up your LLM prompt,
(Here’s where it gets really exciting), instead of sending your prompt to an LLM,
just send the prompt directly to me!
“What?” I hear you crying, “My prompt doesn’t contain all the information you need!”
Metabase is a really nice free software business intelligence tool. It’s quite easy to
deploy; they provide a Docker image and all you really need for a production instance is
your own database.
I use Flux to manage my cluster so I thought it would be nice to utilise
image update automation to keep up to date
with updates automatically. But Metabase makes this slightly difficult because it
doesn’t use proper semantic versioning on the image tags.
One of the most tried and tested techniques in programming is to build components that
map closely to the real-world problems you are trying to solve. In domain-driven design
(DDD), one of the key tenets is to have a domain model at the centre of your
application.
The domain model should be built in close collaboration with domain experts and,
crucially, should consist purely of high-level domain logic and be completely free of
low-level application logic like databases, GUIs, web frameworks etc.
Talos Linux is a minimal Linux distribution designed purely for running Kubernetes
nodes. It’s pretty neat, check out its website here: https://www.talos.dev/
I use XCP-ng in my homelab for running virtual machines and decided to set up a Talos
cluster there. I followed the excellent Getting Started guide but there are a few
additional things I wanted to do right from the start:
Have the Xen Guest Tools (aka Xen Guest Utilities,
xe-guest-utilities. xen-guest-agent etc.) installed on the nodes,
Have static IP addresses.
I also had to configure the installation disk as mentioned in the guide, to /dev/xvda.
In a previous article I wrote about adding maps to my blog, which is a static site. I
use these maps to write about walks I’ve done. These maps can be dragged, rotated and
zoomed, but there is no backend tileserver supporting it, it all comes from a single
static file. This is thanks to the PMTiles format and support by the frontend renderer
maplibre-gl.
In the original article I took a pre-made pmtile file of the world from the Protomaps
project and extracted a smaller part using the pmtiles CLI. Later I ended up making my
own pmtiles directly from OpenStreetMap data, which I’ll describe below. This provides a
lot of geographic data like coastlines, forests, roads etc., but what I really wanted
was hillshading, which I’ve now figured out.
The West Highland Way is a waymarked, long-distance path in Scotland. It starts just
north of Glasgow and ends at Fort William in the Highlands. It is probably the most
famous of Scotland’s Great Trails (similar to the National Trails in England and
Wales). In September 2024 I walked the West Highland Way over seven days with my partner
Vicki.
Planning
At 154 km the WHW will take most people between 5-10 days to complete. We didn’t fancy
any super long days, so we decided to do it in 7 days.
I’ve been using Kubernetes successfully for a while, but I felt like I still didn’t
fully understand the difference between an Ingress and a LoadBalancer. Whenever I tried
to find an explanation I’d find some vague thing like “they are sort of the same, but
not really”.
The problem is I was thinking about these things wrong. One is not a replacement for
the other; they exist in different planes of abstraction within k8s. For someone who
just wants to deploy a web service, there are three orthogonal concepts you need to
understand in k8s.
