Zero Downtime Deploys, At Last

This is just a short note to document how my current blog is deployed with zero downtime. When I moved my blog from Cloudflare to my own server and Coolify I was somewhat dismayed to learn that Coolify can't do zero downtime (rolling release, blue/green) deployments using docker-compose files. Which is a pity because that is a great little format to configure a set of services and their relationship to each other.

My website had about 65 seconds of downtime just now, because coolify can’t do zero downtime deployments with a docker compose project. My blog consists of a db, a kv store, a node app and a server (for reverse proxying). I do not want to configure all of these services separately in this janky-er than the jankiest jank self hosted paas. I only have myself to blame as I didn’t recognize the downtime problem when moving the blog (and thought that I can live with the jank otherwise). Now I basically have to move again in some form or fashion.

(me on mastodon)

Well, I didn't leave coolify^[1], but I have given in - as it seems the main problem is that docker doesn't expose docker-compose on its API, making it hard for tooling to make use of compose files and have now stopped using docker-compose files.

I now have the following setup:

• One Dockerfile for the blog, the api and the caddy server in front of them
• An external db (postgres) - managed by coolify
• An external kv (redis) - managed by coolify

(With "external" I mean external to the Dockerfile.)

Since I only have one image instead of a bunch of services, I can now use the Dockerfile strategy instead of the docker-compose one, giving me free zero downtime deploys (coolify docs).

• ← Previous
  DailyDogo 1461 🐶