Project ID: 1ad2e8d9-d8df-4cd2-aca2-4be43f7219d3
Environment ID: d579dbb9-e696-4beb-bbb5-84e1c176d53f

I'm currently experimenting on our backend deployments to try to improve build and deployment time and I'm noticing a huge difference between Nixpacks and Docker deployments.

Nixpacks-based deployment:

• Time to build: 3:32min

• Time to deploy: 1:34min

• Time between build log last line and pre-deploy first line: 1:03min

Docker-based deployment:

• Time to build: 1:00min (no cached layers)

• Time to deploy: 0:35min

• Time between build log last line and pre-deploy first line: 0:15min

This is a Turborepo monorepo with a Express-based backend and a few internal packages loaded.
The build time in Nixpacks is mostly spent outside of the actual Turborepo build and the yarn package install (which is actually pretty fast for that repo size).

My main thought would be around build image size, my Dockerfile being a multi-stage build with an output size of 405MB, including only the backend dependencies and extra packages for the database migration tools.
I don't know any way to see what's the situation over Nixpacks but I suppose it keeps all the monorepo packages in the final image, which Yarn indicates to be around 1.26GiB (thanks Expo/React Native).

From that, I have those questions:

• Is this a normal/expected behavior?

• Is my hypothesis correct/plausible?

• What's recommended between Nixpacks and Docker images in terms of integration with Railway (monitoring, serverless/sleep mode, …)?

Thanks!