NAME
senv - secure 12-factor env vars for your apps, in any lang, local and remote
SYNOPSIS
~> senv @production run-you-server-in-the-production-environment.py
~> SENV=development senv your-development-server.rb
~> export SENV=test
~> senv run-my-tests.go
DESCRIPTION
senv is a command that lets you run other commands under named sets of environment variables. named sets are specified via the @name syntax, or by setting the meta-environment variable SENV. the named set, which is stored locally, but encrypted, will be unpacked before running your process, which can be written in any language, and all the variables for that SENV will be loaded into the process environment. senv stores it's environment files in your project's '.senv' directory so, given:
.senv/
├── development.enc.json
├── development.json
├── production.enc.json
└── production.json
one can run commands such as
# load all development environment variables, encrypted and non-encrypted,
# before running app.py
~> senv @development app.py
# load the encrypted/non-encrypted SENV named 'production' and run app.js
~> export SENV=production
~> senv app.js
# if no command is given, simply show the @named environment
~> senv @development
~> senv @production | grep DATABASE_URL
the '.senv' directory is searched for 'upwards', similarly to how git finds its '.git' directory, and will normally exist at your project's root. in it are text based config files in .json, .yaml, or .rb format, in both encrypted, and un-encrypted flavors. senv merges them into one set at load time, using the encryption key stored in '.senv/.key'
NOTE: you will never check in your .senv/.key file. ADD IT TO YOUR .gitignore
NOTE: see above^ note. #important
config files can be either non-encrypted, or encrypted. encrpted files are stored with '.enc' in the filename, obviously.
config files can be either non-encrypted, for example 'development.json', or encrypted, as for the filename 'development.enc.json'. for both 'json' and 'yaml' formats, the files must be simple dictionaries/hashes containing key=var pairs, which will be set in the process's environment
so given .senv/development.yaml containing
APP_ENV : development
USE_SSL : false
and .senv/development.enc.yaml containing (albeit as encrypted text)
API_KEY : very-sensitive-info-654321
running
~> senv @development
will show (or run another command) in this environment
---
APP_ENV : development
USE_SSL : false
API_KEY : very-sensitive-info-654321
of course, senv also supports management of these files
# encrypt a config file
~> cat /tmp/development.json | senv .write .senv/development.enc.json
# read a config file
~> senv .read .senv/development.enc.json
# edit a config file using the value of $EDITOR like a good unix gal
~> senv .edit .senv/production.enc.rb
note that, in addition to simple yaml, and json files, one can also load '.rb' files, which are parsed with full ruby syntax. and this can massively simplify managing complex sets of environment variables for any application.
this does what you'd expect:
# file : .senv/all.enc.rb
ENV['API_KEY'] = '1234-xyz'
# file : .senv/development.rb
Senv.load(:all)
ENV['DB_URL'] = 'db://dev-url'
# file : .senv/production.rb
Senv.load(:all)
ENV['DB_URL'] = 'db://prod-url'
~> senv @production
---
API_KEY : 1234-xyz
DB_URL : db://prod-url
and so does this
# file : .senv/all.enc.rb
('A' .. 'Z').each do |alpha|
ENV[alpha] = 'one for every letter in the alphabet'
end
one can access the environment as in the examples above, using the senv cli program to run another command, but the environment can be loaded programatically as well:
in your shell scripts via
# import all senv env vars into this script
eval $(senv init -)
or, in ruby programs via
require 'senv'
if ENV['APP_ENV'] == 'production'
Senv.load(:production)
else
Senv.load(:development)
end
learn more, and increase your ninja score, by installing senv and running
senv .help
INSTALL
senv supports 3 methods of installation
Stand Alone Binary Distribution
- grab the distribution for your platform at https://github.com/ahoward/senv/tree/main/dist
- if you don't know which flavor of linux you are on run
uname -a
and look at the output
- if you don't know which flavor of linux you are on run
- unpack the distribution
- make sure the ./bin/ directory in the distribution is your $PATH
Stand Alone, Dependency-less, Ruby Script
- grab the distribution here https://github.com/ahoward/senv/blob/main/dist/senv.rb
- drop it in, for example, ./lib/, and 'require "./lib/senv.rb"'
- the distribution is both the lib, and the command line script, so that same file can be saved both as './lib/senv.rb' and './bin/senv', a clever person might save just as './lib/senv.rb' and make a symlink from './lib/senv.rb' to './bin/senv'
RubyGems/Bundler
- in Gemfile, gem 'senv'
- or via rubygems, 'gem install senv'
BIKESHED
Q1:
isn't this an imaginary problem? aren't i still pretty professional dropping all my crap in a .env file and calling it a day?
A1:
no.
you will check it in. you will put it in your backups. but most of all you'll pass all the info inside it around in slack, email, text, and whatsapp because you don't have a better way to give it to people. and you'll do this for each of the 17 config settings your 12-factor app needs. as will the other 11 developers that touch the code over the next 2 years.
check it all in. drive all blames into git.
encrypt what is sensitive, and reduce your problem to merely needing to give the next girl one single key to unpack the whole lot. btw, for this, passing around an senv key, i recommend using
Q2:
i call b.s., the key and the program that uses it to encrypt are still on disk. i'm paranoid and want to solve this problem, along with rewriting the laws of thermodynamics to eliminate 'time t' from the equations. i program scala and haskell.
A2:
carlo has #2 solved for you, at least mostly - https://www.goodreads.com/book/show/36442813-the-order-of-time
now, about the key. there are two main approaches to keeping the key, the code that uses it, and the config values, separate:
this one really doesn't count as one of the two, but if you are not ultra paranoid it is worth mentioning that the difference between having files that are encrypted at rest, with a key that needs located, and a process that knows how to use that key to unpack a binary file full of garbage is light years more complex to exploit vs. having unencrypted credentials lying around in your repo for any backup, stray command, or git commit to reveal to the world.
symlinks are magic. use a thumb drive. keep all your keys there. symlink them into your project. unplug the drive when you are not running 'npm install' or anytime your paranoia compells you to do something. rest well knowing attackes need something you know, the credentials, and something you have, the thumb drive.
avoid complexity. learn your shell. set one damn envronment varialbe when you start working on a project. bask in the free time and lack of complexity. indeed, managing an single environment variable setting can, and does, befuddle many a programmer, but this is a great time to dicuss whether those programmers should have access to any sensitive information, let alone in unencrypted files lying around on thier personal machines. doing it the 'unix way' just isn't that hard:
# export the var, do the work
~> export SENV_ENV=teh-passwordz-y
# fancy
~> SENV_KEY=my-key exec $SHELL
# fancier
~> SENV_KEY=my-key tmuxinator
# ultra fancy and magic, also possibly broken...
~> SENV_KEY=teh-key visual-studio-magick-coding-ide-thing
all of the above a good solutions, execpt the last one. that won't work. acutally, it might. if you are uncomfortable on the command line, and manging environment variables, senv may not be for you. however, if that is the case then managing unencrypted files full of api keys is definitely not for you.
in the end. simple is better, and the power of plaintext endures across presidents and epidemics.
WHY?
so many tools exist to load a file full of environment variables. all of them expose these variables to arbitrary code anytime you:
- run
npm install
,bundle install
,pip install
- accidentally commit .env files to git/version control
- loose your laptop
- a trillion other sloppy ways of leaking unencrypted files full of sensitive information from your developer machine to the world wide web
this problem is neither theoritical nor FUD based:
- https://www.helpnetsecurity.com/2020/04/17/typosquatting-rubygems/
- https://codeburst.io/how-secure-is-your-environment-file-in-node-js-7c4d2ed0d15a
- https://www.twilio.com/blog/2017/08/find-projects-infected-by-malicious-npm-packages.html
- https://blog.sonatype.com/sonatype-spots-malicious-npm-packages
- and a million other reasons not to store unencrypted environment variables
- https://lmgtfy.app/#gsc.tab=0&gsc.q=malicious%20packages%20env
solutions to this problem exist, indeed, i authored the original solution for the ruby programming language:
this solution was eventually adapted and merged into 'Ruby on Rails':
- https://github.com/rails/rails/issues/25095#issuecomment-269054302
- https://guides.rubyonrails.org/4_1_release_notes.html
solutions exist for es6, javascript, go, and many languages. each operates 'cowboy differently':
however, all of these solutions, including my own, operate at the wrong level, which is to say at the language or framework level. this misses, entirely, the point of configuring applications via the environment in the ordained 12-factor way; by requiring tight integration, such as the addition of libraries and tooling into projects to manage, load, and set environment variables, we reduce significantly the simplicity of a pure 12-factor app that does only.
const DATABASE_URL = process.env.DATABASE_URL
it's easy to make this mistake and, so long as your project remains a monolith, it works just fine. until it doesn't.
as many of us know, a real, modern, web project is unlikely to remain a single process in a single framework and language. micro-services will be introduced. background jobs will get introduced. someone will split the main web app from the consumer app. everything will get re-written in node, and then go. finally, each developer in every language will solve the problem their own way, and pass around unencrypted text files full of sensitive information all day long like someone yelling 'b-crypt' very slowly in bounded time. someone will port the application deployment from heroku to gcp, and re-tool setting 100 confiuration variables instead of the one meta SENV_KEY to rule them all. he'll be a 'dev-ops' guy, kind of an asshole, and he'll grind deployments down from a 3 minute process to a 3 week fight about vpns.
don't be that guy.
REFMASTER, OF THE UNIVERSE
http://github.com/ahoward/senv
TL;DR;
it's at the bottom because everyone should RTfM ;-)
↟ senv[]@master $ senv .setup .
[SENV] setup /home/ahoward/git/ahoward/senv/.senv
- .senv/all.rb
- .senv/development.enc.rb
- .senv/development.rb
- .senv/production.enc.rb
- .senv/production.rb
↟ senv[]@master $ cat .senv/all.rb
ENV['A'] = 'one'
ENV['B'] = 'two'
ENV['C'] = 'three'
↟ senv[]@master $ cat .senv/production.rb
Senv.load(:all)
ENV['B'] = 'two (via production.rb)'
↟ senv[]@master $ cat .senv/production.enc.rb
0GWID?䱐xAǼdW)\ 1waxE͑!k
↟ senv[]@master $ senv .read .senv/production.enc.rb
Senv.load(:all)
ENV['C'] = 'three (via production.enc.rb)'
↟ senv[]@master $ cat .senv/.key
770db0fd-fddc-4c8c-a264-37d15766d0a5
↟ senv[]@master $ senv @production
---
A: one
B: two (via production.rb)
C: three (via production.enc.rb)
↟ senv[]@master $ rm .senv/.key
↟ senv[]@master $ senv @production
Senv.key not found in : /home/ahoward/git/ahoward/senv/.senv/.key
↟ senv[]@master $ export SENV_KEY=770db0fd-fddc-4c8c-a264-37d15766d0a5
↟ senv[]@master $ senv @proudction
---
A: one
B: two (via production.rb)
C: three (via production.enc.rb)
↟ senv[]@master $ cat a.sh
#! /bin/sh
echo $C
↟ senv[]@master $ senv ./a.sh
three (via production.enc.rb)
FINALLY
:wqa!