Standalone mode for Gatsby

[tlgimenes]

Hi folks!

I work in a company that uses Gatsby for many of our websites. We are currently deploying it on our own Kubernetes infrastructure using Docker and Nginx. This setup has been working flawlessly for the past 2 years.
After the introduction of Gatsby v4, we decided to give it a try and speedup our build times using DSG and SSR. Using DSG and SSR is great for empowering our developers. This, however, makes it necessary to have a Node.JS/Gatsby capable server on our Dockerized infrastructure.

Our first idea on solving this problem was packing the website's node_modules into a docker image and use the gatsby serve command to generate this DSG and SSR pages lazily.
This solution worked, however, we saw a huge jump in our docker image sizes (going from a few 50Mb to 1.7Gb). This made our startup times skyrocket and made "scale to zero" impossible on these node.js pods. By analyzing the source of this increase in size, it became clear node_modules was the real villain in town.
As a counter measure, we tried to create an empty image containing only gatsby as a dependency. Our hopes were that we would only get Gatsby dependencies on this image, which would reduce node_modules sizes drastically. Turns out that the final node_modules had a size of ~500Mb.

Taking a look at how other frameworks solve this problem, I've come across this standalone mode on Next.JS. The idea is that they generate a folder called standalone with only the minimal dependencies you need to have a Next.JS server running.

By using this idea, and reverse engineering how the gatsby serve command works, I created a self containing script using Express and Webpack that outputs a file named server.js. By placing this file at the root of a Gatsby project, it uses the public and .cache folder to serve a Gatsby website without the need of any node_modules folder. According to my tests, the final docker image had 48Mb of files for serving the gatsby site. This was a nice decrease of the initial 1.7Gb to 48Mb.
The drawback I see on this solution is, since I had to code a new server, I'll also have to keep updating this script with new Gatsby versions.

I decided then, to share this with you and listen how thoughts about having a standalone mode on Gatsby. This could be supported via a Gatsby Plugin, like the Gatsby Fuctions Plugin or even entirely replace the current gatsby serve command.

cheers!