Motivation

Vite and create-react-app both have support for environnement variables (in vite/in cra) but they are bundled at build time when yarn build is run.
If we want to change anything like the URL of the backend the app should connect to, we have to rebuild, we can't ship customizable Docker image of our SPA apps.

The solution would be to be able to do:

 docker run --env FOO="xyz" my-org/my-vite-app

Then access FOO:

• In the code like process.env.FOO or import.meta.env.FOO
• In index.html like <title>%FOO%</title>

spa-env does just that, in a secure, performant and type safe way.

Features

• ✅ No impact on the startup time.
• ✅ No impact on the Docker image size.
• ✅ Require no network connection at container startups.
• ✅ Extends the existing standard in place without breaking them, your dev environnement won't be affected.
• ✅ Secure: It only injects the env declared in the .env file.
• ✅ It works like you are already used to. It just changes when the env are injected.
  This enables for example to conditionally preload one font or another
• ✅ Type safe: An env getter is generated so you know what env are available.
• ✅ CRA only: EJS support in public/index.html (did you know?).

Why this instead of import-meta-env?

Because spa-env is fully compatible with the way Vite and CRA manage environments variable.
It doesn't overwrite the existing standard, it extends it by enabling you to take control over when the ENV are injected.
This means that your development environnement, your storybook, your testing framework won't be affected at all by the fact you are using this module.

Project starter

Vite

spa-env + Vite + Nginx + Docker

👉 https://github.com/codegouvfr/spa-env-vite-starter

Create React App

spa-env + create-react-app + Nginx + Docker

👉 https://github.com/codegouvfr/spa-env-cra-starter

Manual setup

Start by installing the tool

yarn add spa-env

Then declare all the allowed environment variables into the .env file of your project

Example with Vite:
.env

VITE_FOO="Default value of foo"
VITE_BAR=
VITE_BAZ=
VITE_FIZZ=

Example with CRA: .env

REACT_APP_FOO="Default value of foo"
REACT_APP_BAR=
REACT_APP_BAZ=
REACT_APP_FIZZ=

Once it's done run the script npx generate-env-getter ( Use npx generate-env-getter js if you your project don't use TypeScript)

It will generate src/env.ts ( or src/env.js) looking like:

/*
 * Automatically generated by spa-env.
 * If you wish to declare a new environment variable declare it in the .env file (prefixed by REACT_APP_)
 * then run 'npx generate-env-getter' at the root of your project.
 * This file will be updated.
 */
import { getEnvVarValue } from "spa-env";

export const envNames = ["FOO", "BAR", "BAZ", "FIZZ"] as const;

export type EnvNames = (typeof envNames)[number];

let env: Record<EnvNames, string> | undefined = undefined;

export function getEnv() {
    if (env === undefined) {
        env = {} as Record<EnvNames, string>;
        for (const envName of envNames) {
            env[envName] = getEnvVarValue(envName);
        }
    }

    return env;
}

(This file should be gitignored)

Now let's test it by creating a .env.local file like:

VITE_BAR="Value of bar defined in .env.local" # Or REACT_APP_BAR= if in CRA

And let's do this somewhere in our code:

import { getEnv } from "./env.ts";

console.log(getEnv());

Now if we run REACT_APP_BAZ="Value of baz passed inline" yarn start we get this in the console:

{
    "FOO": "Default value of foo",
    "BAR": "Value of bar defined in .env.local",
    "BAZ": "Value of baz passed inline",
    "FIZZ": ""
}

Now if you run yarn build then BAZ="Value of baz on the server" npx embed-env the value of BAZ will be injected in build/index.html (or html/index.html) so that if you start statically serving the build/ dir, for example with serve -s build you will get this in the console:

{
    "FOO": "Default value of foo",
    "BAR": "Value of baz on the server",
    "BAZ": "",
    "FIZZ": ""
}

Note that on the server the environment variable names don't need to be prefixed with REACT_APP_ (they can though). Also note that the script runs very fast and thus represent virtually no overhead when starting your container.
By default embed-env does not embed variables defined in .env.local, if you want to include them use: --includes-.env.local or -i.

The next step is to set up a clean Dockerfile where there is both node and Ngnix available.
Node for being able to run npx embed-env and Ngnix for serving the app.
It is also important to make sure spa-env is not bootstrapped by npx in the entrypoint.

Usecase example

Onyxia-web is a create-react-app distributed as a Docker image.

Sysadmins that would like to deploy Onyxia on their infrastructure can simply use the official Docker image and provide relevant environnement variable to adjust the theme/branding of the website to their usecase.

Here are two deployment example: