You’ve resolved to build your company’s Next Big Thing in Phoenix and Elixir. That’s great! You’re facing a problem though – all user authentication and access concerns are performed on your Rails system, and the work to reimplement this in Phoenix is significant.
Fortunately for you, there is a great Phoenix plug to share session data between Rails and Phoenix. If you pull this off, you’ll be able to build your new API on your Phoenix app, all while letting Rails handle user authentication and session management. Let’s get started!
Before we begin
In this scenario, you want to build out a new API in Phoenix that is consumed by your frontend single-page application, whose sessions are hosted on Rails. We’ll call the Rails app
rails_app and your new Phoenix app
Additionally, each app will use a different subdomain. The Rails app will be deployed at the
www.myapp.com subdomain. The Phoenix app will be deployed at the
We are going to take Chris Constantin‘s excellent
PlugRailsCookieSessionStore plug and integrate it into our Phoenix project. Both apps will be configured with identical cookie domains, encryption salts, signing salts, and security tokens.
In the examples that follow, I’ll be using the latest versions of each framework at the time of writing, Rails 4.2 and Phoenix 1.2.
Cookie-based session storage
Our session data is stored on the client in a secure, encrypted, validated cookie. We won’t cover the basics of cookies here, but you can read more about them here.
Our approach will only work if your current Rails system utilizes cookie-based sessions. We will not cover the use case with a database-backed session store in SQL, Redis, or Memcache.
Step 1: Configure Rails’ cookie store
Let’s set up your Rails app to use a JSON cookie storage format:
Your app may not be configured with a
SESSION_ENCRYPTED_SIGNED_COOKIE_SALT. You may generate a pair with any random values.
Some speculate that Rails does not require the two salts by default because the
SECRET_KEY_BASE is sufficiently long enough to not require a salt. In our example, we choose to supply them anyways to be explicit.
Another important value to note here is that we have chosen a key for our session cookie –
_rails_app_session. This value will be the shared cookie key for both apps.
Step 2: Configure the plug for Phoenix
Turning our attention to our Phoenix app, add the library to
mix deps.get to fetch the new library.
Now in your
web/phoenix_app/endpoint.ex file, remove the configuration for the existing session store and add the configuration for the Rails session store.
We set a
DOMAIN environment variable with the value
myapp.com. The goal is for these two apps to be able to be deployed at any subdomain that ends in
myapp.com, and still be able to share the cookie.
secure flag configures the app to send a secure cookie, which only is served over SSL HTTPS connections. It is highly recommended for your site; if you haven’t upgraded to SSL, you should do so now!
Our cookies are signed such that their origins are guaranteed to have been computed from our app(s). This is done for free with Rails (and Phoenix’s) session libraries. The signature is derived from the
encrypt flag encrypts the contents of the cookie’s value with an encryption key derived from
encryption_salt. This should always be set to
key_digest are configurations that dictate how the signing and encryption keys are derived. These are configured to match Rails’ defaults (see also: defaults). Unless your Rails app has custom configurations for these values, you should leave them be.
Step 3: Configure both apps to read from the new environment variables
Be sure your development and production versions of your app are configured with identical values for
SESSION_ENCRYPTED_SIGNED_COOKIE_SALT. You’ll want to make sure your production apps store identical key-value pairs.
Step 4: Change Phoenix controllers to verify sessions based on session data.
Now when the Phoenix app receives incoming requests, it can simply look up user session data in the session cookie to determine whether the user is logged in, and who that user is.
In this example, our Rails app implements user auth with Devise and Warden. We know that Warden stores the user ID and a segment of the password hash in the
warden.user.user.key session variable.
Here’s what the raw session data looks like when the
PlugRailsCookieSessionStore extracts it from the cookie:
Two keys are important to remember here – the
warden.user.user.key used to store a reference to the logged-in user, and the
session_id token used to uniquely identify this session. This session ID can be used as a unique reference to this session if you choose to store additional session data in an external store.
A very naive plug implementation simply renders a 401 if the session key is not found in the session, otherwise it allows the request through.
What if I’m not using Warden?
No worries. You’ll want to modify your session fetching code to search for the user key in the session. You may need to dig deep into your Rails app to discover where this is implemented, but typically you only need to find the path to the user’s ID in the Rails session Hash, then implement the finder code in
Step 5: Move session concerns into its own module
Let’s move session concerns around session parsing out of the controller into its own
Session module. Additionally, we include two helpers,
This leaves the controller looking skinnier, implementing only the Plug. Extracted methods are delegated to the new
Finally, we implement some nice helpers for your APIs:
This gives you the ability to call
current_user(@conn) from within your views, should you desire to.
Step 6: Fetching additional information from the backend
Let’s enhance our
Session module with the capability to fetch additional information from another resource.
In this case, we’ll model a call an external User API to fetch extended data about the User, potentially with some sensitive information (that’s why we didn’t want to serialize it into the session).
Session can be extended to return the proper
User, which may provide more utility to us as we implement our Phoenix feature.
Here’s the two apps in action:
Caveats and considerations
Heroku deployment gotchas
If you are deploying this to Heroku with the popular Heroku Elixir buildpack, please be aware that adding or changing environment variables that are required at build time require that the new environment variables outlined here are added to your
elixir_buildpack.config file in your repository.
At the time of this writing, Phoenix and Rails overwrite each others’ session CSRF tokens with incompatible token schemes. This means that you are not able to make remote POST or PUT requests across the apps with CSRF protection turned on. Our current approach will work best with read-only APIs, or sessions that primarily make requests to the same host app.
Be judicious about what you store in a cookie
Cookies themselves have their own security strengths and drawbacks. You should be judicious about the amount of data you store in a session (hint: only the bare minimum, and nothing sensitive).
The OWASP guidelines also provide some general security practices around cookie session storage.
Moving beyond session sharing
Even though this scheme may work in the short run, coupling our apps at this level in the long run will result in headaches as the apps are coupled to intricate session implementation details. If, in the long run, you wanted to continue scaling out your Phoenix app ecosystem, you may want to look into the following authentication patterns, both of which move your system toward a microservices architecture.
1) Develop an API gateway whose purpose is to be the browser’s buffer to your internal service architecture. This one gateway is responsible for identity access and control, decrypting session data and proxying requests to an umbrella of internal services (which may be Rails or Phoenix). Internal services may receive user identities in unencrypted form.
2) Consider implementing a JWT token implementation across your apps, in which all session and authorization claims are stored in the token itself, and encrypted in the client and server.. This scheme may still rely on cookies (you may store the token in a cookie, or pass it around in an HTTP header). The benefits of this scheme is the ability for your app(s) to manage identity and authentication claims on their own without having to verify against a third-party. Drawbacks of this scheme are the difficulty around revoking or expiring sessions.
Each of these approaches is not without overhead and complexity; be sure to do your homework before your proceed.
That’s it! I hope I’ve illustrated a quick and easy way to get a working Phoenix app sharing sessions with Rails app(s), should you decide to prototype one in your existing system. I’ve also pushed up a sample app if you want to cross-reference the code. Good luck!