You should always keep development and production separate my friend, but how do you pull that off when everyone has your back and tries to keep your app secure?

wait what ๐Ÿคจ

Here’s the problem ๐Ÿ‘‰

You’ve built a cool app. Static JAM on the frontend, Serverless functions on the backend, nearly free to host, super fast to load. It’s amazing and you’re proud as heck.

Time to show off!

show_off giphy

You know dev and production should be separate. Wouldn’t want users to see a bug next time you mess around with code. Sure as heck don’t want to mix test and production data.

Prod vs. Dev on your serverless backend

Separating development and production on your backend is easy as pie. Change the stage: in serverless.yml and voila.

Using the demo app from ServerlessReact.Dev for example:

Click through for source

This configures a /graphql endpoint served by an AWS Lambda and a DynamoDB table to store your data. Both the DynamoDB table and the GraphQL endpoint use stage as part of their name.

Change that to stage: dev to stage: prod, run yarn deploy, and you get a whole new copy of your backend infrastructure. A new AWS Lambda, new DynamoDB table, everything.

perfect giphy

Making your stage easy to change

Changing serverless.yml on deploy is tedious. You’ll commit the wrong version to git, too.

Define your stage with command line arguments instead!

Like this:

Click through for source

Now you can add a helpful little bash script โ€“ย because I couldn’t figure out how to do this with Node and Bash is just fine. ๐Ÿ˜‡

Click through for source

Add to your package.json

Click through for source

And you can run yarn deploy prod to deploy to production, or yarn deploy to get development by default.


Prod vs. Dev on your JAMStack frontend

Separating production and development on the JAM is easy in theory ๐Ÿ‘‰ push to a branch and you get a deploy preview. Press a button or push to a special branch and you get production.

Both Netlify and Zeit do that. I’m sure others do, too.

That was easy.

Now, how do you tell your app that production data lives on and development should talk to

think_hard giphy

That’s where it gets tricky.

I’ve found 3 approaches with varying levels of success and different tradeoffs.

1. The Gatsby way

Gatsby comes with built-in support for .env files. Put your environment variables in .env.development or .env.production, access through process.env and let Gatsby handle the rest.

Say you want to tell Gatsby which GraphQL endpoint to use when building your JAMStack app.

Go in gatsby-config.js and add your server as a source

Click through for source

You can now configure the GATSBY_GRAPHQL_URL environment variable in 2 files to differentiate production and development.

Click through for source
Click through for source

Gatsby will choose production when you run gatsby build and development when you run gatsby development.

You can even use these in the browser!

Click through for source


The tradeoff

The Gatsby way breaks down when you deploy.

Both Zeit and Netlify helpfully strip your .env files during deploy. Wouldn’t want to store secrets in plaintext on their servers.

This is great from a security perspective.

And it breaks Gatsby support for .env files. They aren’t available during build. ๐Ÿคฆโ€โ™‚๏ธ

2. The Zeit way

The Zeit way is more secure. You store secrets and configs in their secure storage. Encrypted, easy to rotate, not part of your code.

Now you gotta jump a couple more hoops.

First you add your secret to Zeit

Click through for source

Then you create a now.json file

Click through for source

And then, you use it in gatsby-config

Click through for source

When you deploy to Zeit, the build process takes your now.json, replaces @variable with values from storage, loads the configs into process.env, and Gatsby can use them while building.

Sounds complicated … ๐Ÿคจ

But it works. Sort of

Netlify has a similar approach, by the way.

The tradeoff

Notice there’s no prod vs. dev. You get a config and that’s it. Zeit says they’ll have native support for environments soon.

Click through for source

The other tradeoff is that these configs aren’t available at runtime. You get them when compiling and can’t use them in the browser.

And you can’t import 'now.json' either. Zeit deletes that file โ˜น๏ธ

3. The way that actually works

In the end, here’s what works best:

  1. Put your configs in JSON files
  2. Import in gatsby-config.js
  3. Add to siteMetadata
  4. Access via static queries

Kinda bonkers but it works.

You start with 2 JSON files similar to the original .env approach.

Click through for source
Click through for source

Import in gatsby-config.js

Click through for source

Use the configs directly in your build-time code via config.X. Works just fine.

You can import JSON files in the browser, sure, but you won’t know which environment. Put them in siteMetadata.myConfig instead, so you can get configs with static queries anywhere in your app.

Like this

Click through for source

And you get …

yay giphy

The tradeoff

You’re storing configs in version control. In plaintext. They better not be secret ๐Ÿ˜‰

And you’ll have to re-deploy any time you change a value.

But hey at least it works.

In conclusion

There’s no perfect solution here yet, I’m afraid. The JSON approach combined with static queries seems best for now. If you’ve found something better, I’d love to know.

Happy hacking โœŒ๏ธ



PS: I like to use AWS SecretsManager on the backend for real secrets, maybe there’s an API you could use from the browser ๐Ÿค”

Learned something new? Want to improve your skills?

Join over 10,000 engineers just like you already improving their skills!

Here's how it works ๐Ÿ‘‡

Leave your email and I'll send you an Interactive Modern JavaScript Cheatsheet ๐Ÿ“–right away. After that you'll get thoughtfully written emails every week about React, JavaScript, and your career. Lessons learned over my 20 years in the industry working with companies ranging from tiny startups to Fortune5 behemoths.

PS: You should also follow me on twitter ๐Ÿ‘‰ here.
It's where I go to shoot the shit about programming.