Introducing Environmental

6 minute read

Some people feel that shipping .json / .yml / .xml config files is an upgrade over using archaic environment variables.


Don't let your app load its config, inject it instead.

Unix environment vars are ideal for configuration and I have yet to encounter an application that wouldn't be better off with them. Why?

  • You can override a value at near-runtime without having to change/backup config files: DEBUG=*.* node run.js
  • You can inject environment variables (passwords, API keys) into the memory of a process belonging to a non-privileged user: source envs/ && sudo -EHu www-data node run.js without having to run / write any software for it.
  • You can inherit. Inside, just source, inside source
  • Your operating system is aware and provides tools to inspect, debug, optionally pass on to other processes, etc.
  • You can directly use config across languages, e.g. in supporting BASH scripts
  • You can directly use the config in a terminal yourself, e.g. cd ${MYAPP_DIR}

And as with any other type of config:

  • You can group/save them into files and keep them out of version control

One downside of environment variables is that there is little convention and syntactic sugar in the high-level languages. It doesn't feel atomic and you think it's more likely to let you down. This module attempts to change that.

Environmental doesn't:

Environmental does:

  • Impose one way of dealing with environment variables
  • Make vars available in nested format inside your app (e.g. MYAPP_REDIS_HOST) becomes
  • <3 unix
  • Interpret multiple inherited bash environment files in an isolated environment to capture them, and prepare them for exporting to Nodejitsu or Heroku.



Environmental tree:
│   └──

On disk:


You could make this super-DRY, but I actually recommend using mainly and, and duplicate keys between them so you can easily compare side by side. Then just use,, for tweaks, to keep things clear. Read 'Inheritance can be a bitch' to see why.

Inheritance can be a bitch

One common pitfall is re-use of variables:

export MYSQL_HOST=""
export MYSQL_URL="mysql://user:pass@${MYSQL_HOST}/dbname"

Then when you extend this and only override MYSQL_HOST, obviously the MYSQL_URL will remain unaware of your host change. Ergo: duplication of vars might be the lesser evil here compared to going out of your way to DRY things up.

Inject features

Instead of having your code make decisions based on environment:

if process.env.NODE_ENV == "production"
  # Install cronjobs

Keep that responsibility with your environment files:

$ cat envs/

$ cat envs/
if config.cronjobs.install == "1"
  # Install cronjobs

Mandatory and unprefixed variables

These variables are mandatory and have special meaning. There is no syntactic sugar for them, you are to access them via process.env.<var>:

export NODE_APP_PREFIX="MYAPP" # filter and nest vars starting with MYAPP right into your app
export NODE_ENV="production"   # the environment your program thinks it's running
export DEPLOY_ENV="staging"    # the machine you are actually running on
export DEBUG=*.*               # used to control debug levels per module

After getting that out of the way, feel free to start hacking, prefixing all other vars with MYAPP - or the actual short abbreviation of your app name. Don't use an underscore _ in this name.

In this example, TLS is our app name:

export TLS_REDIS_HOST=""
export TLS_REDIS_USER="jane"

Getting started

In a new project, type

$ npm install --save environmental

This will install the node module. Next you'll want to set up an example environment as shown in layout, using these templates:

cp -Ra node_modules/environmental/envs ./envs

Add envs/*.sh to your project's .gitignore file so they are not accidentally committed into your repository.
Having env files in Git can be convenient as you're still protoyping, but once you go live you'll want to change all credentials and sync your env files separately from your code.

Accessing config inside your app

Start your app in any of these ways:

source envs/
node myapp.js
source envs/
DEBUG=*.* node myapp.js
source envs/
# Following seems weird, but sudo will not preserve $PATH, regardless of -E
sudo -EHu www-data env PATH=${PATH} node myapp.js
source envs/ && node myapp.js
start myapp # see upstart example below

Inside your app you can now obviously already just access process.env.MYAPP_REDIS_HOST, but Environmental also provides some syntactic sugar so you could type instead. Here's how:

var config = require('environmental').config();

// This will return
//   { redis: { host: '' } }

Or in coffeescript if that's your cup of tea:

config      = require("environmental").config()
redisClient = redis.createClient(config.redis.port,

As you see

  • any underscore _ in env var names signifies a new nesting level of configuration
  • all remaining keys are lowercased

config takes two arguments: flat defaulting to process.env, and filter, defaulting to process.env.NODE_APP_PREFIX. Changing these allow you to inject or reload environment variables.

Capturing a specific config file

By default, environmental code will capture environment variables of the current process. However, you can also use it to capture variables in isolation as produced by a gives shell file (works with inherits too).

env = new Environmental
env.capture "#{__dirname}/envs/", (err, flat) ->
  expect(flat.MYAPP_REDIS_HOST, "")

Notice in does not nest the configuration. You can do that by using the config method and passing it flat environment vars:

config = Environmental.config flat, "MYAPP"

    host: ""

Exporting to Nodejitsu

Nodejitsu also works with environment variables. But since those are hard to ship, they want you to bundle them in a json file.

Environmental can create such a temporary json file for you from the command-line. In this example it figures out all vars from envs/ (even if it inherits from other files):

./node_modules/.bin/environmental --file=envs/ --format=json > /tmp/jitsu-env.json
jitsu --confirm env load /tmp/jitsu-env.json
jitsu --confirm deploy
rm /tmp/jitsu-env.json

Exporting to Heroku

heroku config:set $(./node_modules/.bin/environmental --file=envs/ --format=space)

Exporting to your own servers

To generate a single file that your server can source:

./node_modules/.bin/environmental --file=envs/ --format=newline

Note that this is different from:

source envs/ && env

As the output is cleansed from any environment variable that was not declared in env/ or one of it's ancestors.

You could use this list to inject into a process upon (re)starts, or save as a file so upstart can inject it into a non-privileged process, and use e.g. rsync to distribute it amongst privileged users:

for host in `echo ${MYAPP_SSH_HOSTS}`; do
  rsync \
   --recursive \
   --links \
   --perms \
   --times \
   --devices \
   --specials \
   --progress \
  ./envs/ ${host}:${MYAPP_DIR}/envs

Injecting into a non-privileged user process

When you deploy your app into production and you run the servers yourself, you might want to use upstart to respawn your process after crashes.

Here's how an upstart file (/etc/init/myapp) could look like, where the root user injects the environment keys into process memory of an unpriviliged user.

This has the big security advantage that you own program cannot even read its credentials from disk.

stop on runlevel [016]
start on (started networking)

respawn limit 10 5

limit nofile 32768 32768

pre-stop exec status myapp | grep -q "stop/waiting" && initctl emit --no-wait stopped JOB=myapp || true

  exec bash -c "cd /srv/myapp/current \
    && chown root envs/*.sh \
    && chmod 600 envs/*.sh \
    && source envs/ \
    && exec sudo -EHu www-data make start 2>&1"
end script

Leave a Comment Right Here