blog: post on self hosting blog

This commit is contained in:
Thomas Bishop 2025-07-31 17:13:26 +01:00
parent a9440d6844
commit fb782e8cf2
4 changed files with 146 additions and 6 deletions

Binary file not shown.

Binary file not shown.

After

Width:  |  Height:  |  Size: 227 KiB

View file

@ -0,0 +1,136 @@
---
title: "Self-hosting this site"
slug: /self-hosting-this-site/
date: 2025-07-21
tags: ["self-hosting"]
---
[Previously](https://systemsobscure.blog/posts/how-I-deploy-this-site) this site
was deployed as follows. I used [Gatsby.js](https://gatsbyjs.com) to generate a
static website from React JavaScript. When changes to the `main` branch were
pushed to the remote GitHub repository, a GitHub action would execute, building
the source files and transferring them to an S3 bucket. The HTML generated was
then served using AWS Cloud Front.
As I am now trying to self-host as many services as I can, it was time remove
the dependency on AWS and GitHub and instead serve the site from my VPS and
manage deployment via my
[self-hosted Forgejo instance](https://forgejo.systemsobscure.net/thomasabishop).
As part of this process I went on a bit of a side-quest to rebuild the site
without Gatsby.
When I tried to update to the lastest version of Gatsby I faced several problems
mostly due to Gatsby trying to foist server-side rendering on every project. For
such a small blog site with next to zero traffic this is unnecessary. Moreover,
I realised that I could also do without all the bloat that Gatsby adds via its
plugin ecosystem. (Perhaps if frontend bundles weren't so large, SSR wouldn't be
needed lol.)
So I decided to simplify the blog and build it as a [Vite](https://vite.dev)
React application that gets its content from a JSON index that is generated via
a pre-build script.
I
[wrote a script](https://forgejo.systemsobscure.net/thomasabishop/systems-obscure/src/branch/main/scripts/generate-post-index.js)
in JavaScript that loops through all the blog posts in Markdown format from a
`/posts` directory at the project root. It creates a JSON array with an entry
for each post. For example, this post would be represented as follows:
```json
[
{
"slug": "self-hosting-this-site",
"title": "Self-hosting this site",
"date": "2025-07-22T00:00:00.00Z",
"tags": ["self-hosting"],
"html": "<p>Previously this site was..."
}
]
```
To convert the raw markdown to HTML I used
[marked](https://github.com/markedjs/marked). I also added the following steps:
- convert `<code>` tags into syntax highlighted blocks using
[shiki](https://shiki.style/)
- compress and resize images and then transfer them to the `public/` directory
where Vite sources static assets when deployed
- find and replace all image `src` attributes in the resulting HTML with the
`public/` file path
When I run `npm run build:posts` this generates the `post-index.json` file and
saves it to `/public` so that the React application can read from it when
rendering the site content.
<figure style="margin: 1rem 0">
<video controls>
<source src="./img/post-index-build-script.webm" type="video/webm">
Your browser does not support the video tag.
</video>
<figcaption>Running the pre-build script to generate the site content.</figcaption>
</figure>
I created a custom hook called `usePosts` that fetches the index and saves it to
the session storage to avoid unnecessary network requests. You can see this
being invoked in the
[template component for blog posts](https://forgejo.systemsobscure.net/thomasabishop/systems-obscure/src/commit/43eec03edd2e6330362cc5605a48a91387eb49d3/src/templates/BlogTemplate.tsx).
Having rebuilt the site without the dependency on Gatsby, the next step was to
deploy it.
I wanted to retain the "push and deploy" model that I previously achieved via
the GitHub Action. This was easy because the syntax for Forgejo Actions is
practically identical. You place the YAML declaration in the `.forejo/`
directory of the project and Forejo picks it up on each push to the remote.
Here's the file:
```yaml
name: Deploy Blog
on:
push:
branches: [main]
jobs:
deploy:
runs-on: ubuntu-latest
container: node:18
steps:
- uses: actions/checkout@v3
- run: npm install
- run: npm run build:posts
- run: npm run build
- run: |
echo "${{ secrets.SSH_FORGEJO_KEY }}" > /tmp/ssh_key
chmod 600 /tmp/ssh_key
ssh -i /tmp/ssh_key -o StrictHostKeyChecking=no ${{ vars.VPS_USER }} "bash -c 'rm -rf /var/www/systemsobscure.blog/*'"
scp -i /tmp/ssh_key -o StrictHostKeyChecking=no -r dist/* ${{ vars.VPS_USER }}:/var/www/systemsobscure.blog/
rm /tmp/ssh_key
```
Before compiling the build I first run the pre-build script to generate the post
index. Then I remove the existing source files for the site in `var/www/` over
`ssh` and then transfer the new source files using `scp`. (While I run most
third-party services on my VPS via Docker, I prefer to keep it simple with my
own applications and not use containers.)
![Forgejo Actions runner executions.](./img/systemsobscure-forgejo-runners.png)
In order for the site to be served from `var/www/systemsobscure.blog`, I had to
give my Docker instance of nginx running on the VPS access to this directory by
[adding it to the volume mappings](https://forgejo.systemsobscure.net/thomasabishop/self-host/commit/8c380b71735f278f4309bbd14ad96cb9a29104b7)
in the `docker-compose.`
Then I added an
[nginx `.conf` file for the blog](https://forgejo.systemsobscure.net/thomasabishop/self-host/src/branch/main/proxy/nginx/conf.d/systemsobscure.conf).
This file specifies the SSL certificate to use and sets `index.html` as the
document root. It also adds some default caching and compression, along with
security headers.
Finally I
[updated my SSL certificate generation script](https://forgejo.systemsobscure.net/thomasabishop/self-host/commit/61cdbe43c2041d5961ef74416270794eb6fb91c6)
to include `systemsobscure.blog`.
And that's it. My personal website is now self-hosted on my VPS and is
automatically deployed via pushes to my self-hosted Git forge. No more GitHub,
no more AWS.

View file

@ -15,23 +15,27 @@ const processBlogImages = async () => {
fs.mkdirSync(destDir, { recursive: true })
}
// Copy all images
const files = fs.readdirSync(srcDir)
const imageExtensions = [".jpg", ".jpeg", ".png", ".gif", ".webp", ".svg"]
const imageExtensions = [
".jpg",
".jpeg",
".png",
".gif",
".webp",
".svg",
".webm",
]
// Use for...of loop instead of forEach to work with async/await
for (const file of files) {
const ext = path.extname(file).toLowerCase()
if (imageExtensions.includes(ext)) {
const inputPath = path.join(srcDir, file) // Define inputPath and outputPath
const outputPath = path.join(destDir, file)
if (ext === ".svg") {
// SVGs just get copied
if (ext === ".svg" || ext === ".webm") {
fs.copyFileSync(inputPath, outputPath)
console.info(`📸 Copied ${file}`)
} else {
// Process other images
await sharp(inputPath)
.resize(1200, 800, {
fit: "inside",