blog/content/posts/nixos-remote-auto-static-site-deploy.md
2025-08-14 15:16:21 +01:00

115 lines
3.6 KiB
Markdown

---
title: Over-engineered (?) nixos blog deployment setup
date: 2025-08-11
tags:
- nixos
draft: false
---
As is traditional with people hosting their own blog I'm going to do a post detailing EXACTLY how I'm hosting my blog.
Down to the last dirty detail.
I have nothing better to talk about.
Here is a diagram I edited to illustrate (credit to xkcd I think?).
![self-hosting](/image/self-hosting.png)
I host my site on a hetzner vps running nixos.
I also have a git repo where all the static files for my blog live.
I had previously been manually rsyncing the website up to my vps from my laptop.
Qute an easy, efficient solution; it worked well.
But not very nixos; far too simple, not sufficiently over-engineered.
So in true nixos fashion I decided I'd spend a couple of hours sorting the problem so I'd maybe save a minute once a year when I write a blog post.
### Remote Rebuilds
First, I'll show the fancy way to rebuild your remote nixos systems via ssh.
In my case, this means I can rebuild my hetzner box from my laptop.
You can read the wiki about it [here](https://wiki.nixos.org/wiki/Nixos-rebuild#Deploying_on_other_machines).
This sets up ssh with key-based authentication and lets our local user in.
This config belongs on the remote machine.
```nix
users.users.blog-king.openssh.authorizedKeys.keys = [
# ssh public key on computer you're deploying from
"ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIPzFa1hmmmmmPL5HvJZhXVEaWiZIMi34oR6AOcaaaaaaa"
];
nix.settings.trusted-users = [ "blog-king" ];
# ssh daemon
services.openssh = {
enable = true;
openFirewall = true;
settings = {
PasswordAuthentication = false;
PermitRootLogin = "no";
};
};
```
Once you have this going on your remote machine (in my case the hetzner vps) you should be able to rebuild the remote machine with `nixos-rebuild --target-host blog-king@remote-ip-here --ask-sudo-password switch`.
The `--ask-sudo-password` is not required if you ssh in as root though that would be a touch gauche.
### Caddy
You can do this with whatever your preferred webserver is.
I am a caddy stan.
This opens the necessary ports in the firewall and sets up caddy in file server mode pointing at `/etc/blog`.
```nix
networking.firewall.allowedTCPPorts = [
80
443
];
services.caddy = {
enable = true;
extraConfig = ''
blog.example.org {
root * /etc/blog
file_server
}
'';
};
```
### Getting the files from git
We have a web server pointing at `/etc/blog`.
The last piece of the puzzle is to get the static files from our git repo and spit them out in that directory.
I'm using the `fetchFromGitea` helper here which works for gitea and forgejo instances.
The `fetchFromGitHub` helper would look very similar.
You can get the `rev` and `sha256` of the commit using `nix-prefetch-git`.
Also note the little `/public` at the end of the `source` string.
That's the directory of the git repo that the website source lives.
```nix
environment.etc."blog" = {
enable = true;
target = "blog";
source = "${
pkgs.fetchFromGitea {
domain = "git.example.org";
owner = "james";
repo = "blog";
rev = "32d81f01388c88a259eed2ba52f4545dbcb1eb07";
sha256 = "173g99dj8y4sw1v7f1s5f7zgcrrlr6dly9n6ysr2i4jg095lkxw8";
}
}/public";
user = "caddy";
group = "caddy";
};
```
So now with all that setup the blog post work flow is:
- Commit rebuilt website to repo
- Update the `rev` and `sha256` to the new commit (this is annoying and I'm trying to work out a good way to automate it)
- Rebuild vps from laptop
Not necessarily faster than the old rsync method but it's pretty damn declarative, that's for sure.