I recently migrated from my previous server recently, and I took this opportunity to clean up my blog deployment routine. As I have said previously, this blog is powered by pelican and rst files, which means I am maintaining it in a private git repo.

Since I migrated to a more powerful server, I can now run gitea without worries, which allows me to use webhooks to notify new commits to a daemon.

I wrote a minimal HTTP server using aiohttp; as the blog is on the same machine as gitea, I don’t have to bother with secrets, nginx config, or whatever, and can open a port on localhost instead. (worst case: someone builds my blog)

import os
import sys
from aiohttp import web
from subprocess import run
from asyncio import get_event_loop
from os.path import abspath, dirname

os.chdir(dirname(abspath(sys.argv[0])))

def update():
    run(['git', 'pull'])
    run(['./deploy.sh'])

async def handle(request):
    loop = get_event_loop()
    await loop.run_in_executor(None, update)
    return web.Response(text='OK')

app = web.Application()
app.add_routes([web.get('/deploy', handle)])
web.run_app(app, host='localhost', port=2345)

This daemon runs with a systemd service, taking care of reboots and failures:

[Unit]
Description=Blog deployment
After=network.target
[Service]
ExecStart=/usr/bin/python -OO /home/blog-deploy/blog/server.py
User=blog-deploy
Restart=on-failure

[Install]
WantedBy=multi-user.target

I do the build + deploy in a bash script because that’s the quickest way to write it. Pelican clears the directory first, which means the renaming dance is required to minimize possible errors while the blog is building.

#!/usr/bin/env bash
pushd $(dirname $(realpath $0))
mkdir -p tmp_build public_html
make OUTPUTDIR=$PWD/tmp_build html
chown blog-deploy:http -R tmp_build
mv public_html to_delete
mv tmp_build public_html
rm -r to_delete