Adam Tuttle

Next.js Docker Single Entrypoint for Dev and Prod

Maybe I shouldn't be as excited about this as I am, but I love it when simple things do smart things.

I'm currently working on a Next.js application, which will be deployed in a Docker container on AWS Fargate. When you generate a new Next.js application, you get these scripts in your package.json to start:

"scripts": {
"dev": "next dev",
"build": "next build",
"start": "next start",
"lint": "next lint"
},

As you might expect, you use npm run dev to start a local development instance, which is not optimized for production (it assumes you're making lots of changes and is more eager to recompile stuff, basically). In production, you should be starting the application with npm run start. This assumes that you've already created a production-optimized build with npm run build, which can make assumptions like you're not going to be changing things, so it should pre-compile as much as possible.

So, your typical Dockerfile for a Next.js application should resemble this:

FROM node:14 as Builder
WORKDIR /usr/src/app
COPY . /usr/src/app
ENV NODE_ENV=production
RUN npm ci

FROM node:14.17.1-alpine
WORKDIR /usr/src/app
COPY --from=builder /usr/src/app /usr/src/app
RUN npm run build
CMD npm run start

This is all pretty standard stuff. We're using a multi-stage build to minimize final container size, and creating a production build. (For what it's worth, I initially had the npm run build step in the first stage, but the build artifacts weren't found when running npm run start in the final container, even with the full app directory copy. 🤷‍♂️)

This is where it gets interesting.

I've written in the past about my love of Makefiles. So now let's assume I have a Makefile with the following target, which I use to start my container for local development.

dev-up: .built
docker run --rm -d --name widgets-dev \
-p 3000:3000 \
-v `pwd`:/usr/src/app \
--env NODE_ENV=development \
myco/widgets:dev

This starts the container with the name widgets-dev, mapping host port 3000 to container port 3000, NODE_ENV set to development, and crucially, a volume of the current working directory mounted as /usr/src/app.

Initially I expected this to just work. Since the working directory is mounted inside the container at the application source directory, fast refresh and all of the usual development-environment goodies should still work out of the box, right?

Well, can you see what's missing? Yes, we're properly overriding the NODE_ENV value when we start the container for development, but the container has only one CMD entrypoint, and we want it to do different things depending on its current environment variables. In dev it should run npm run dev, but in production or qa it should run npm run start.

I found an easy and simple way to accomplish this with npm scripts, but it's not perfect...

"scripts": {
"dev": "next dev",
"build": "next build",
"prestart": "node -e 'process.env.NODE_ENV === \"production\" || process.exit(1);' || npm run dev",
"start": "next start",
"lint": "next lint"
},

Here I've added a script named prestart which npm will execute before it runs the start script. My prestart script uses the inline-evaluation feature of node.js (-e) to check the current value of NODE_ENV and if it's anything other than production the inline script exits with error code 1. After the inline script I have || npm run dev which will execute the npm run dev only if the previous command exited with an error code.

This is working reasonably well for me right now. The container always runs npm run start as its entrypoint, and it runs in dev mode in the dev environment, or production mode in the production environment.

Here's where it's not perfect:

In local development, when the dev server shuts down, npm has been waiting to spring into action and it follows suit by trying to execute npm run start. Remember that the dev server has been running as part of npm run prestart this entire time. I'm not sure yet if there's a way to stop this from happening, and to be fair it's not a big deal because the error happens inside the docker container which hides it from view.

I know that I can make the npm run dev portion of prestart exit with a nonzero code and that this would stop npm from running the start script, but it does then return an error:

"prestart": "node -e 'process.env.NODE_ENV === \"production\" || process.exit(1);' || (npm run dev && exit 1)"

I'm not sure if that's trading one bad thing for another, nor whether or not it's a good trade to make.