Betterment / test_track

Server app for the TestTrack multi-platform split-testing and feature-gating system

TestTrack

TestTrack is an open-source split-testing and feature-toggling system written in Ruby on Rails.

Key features and design decisions

• Uses a stateful server to provide consistent experiences for customers across devices, and allow for bulk assignment overrides.
• Rich client libraries available for multiple platforms optimized to minimize time-to-glass and gracefully degrade if the server is unavailable.
• Designed to streamline developer and PM workflow - focusing on designing and running your tests, not the plumbing, and reducing incidence of implementation mistakes that can hurt your data.
• Provides an optional Chrome extension to allow your team to interactively override their split assignments.
• TestTrack is not an analysis tool. It focuses on fast, trustworthy, and robust split assignment and identity management, leaving analysis to the great tools that already exist, and those that are yet to come.

Check out the Rails at Scale talk for some background on why we built it and some of the key design decisions behind TestTrack:

The TestTrack system consists of the following components:

• the server (this repository)
• the Rails client
• the JS client
• the Dart client
• the Chrome extension

Getting Started

Requirements

The list of requirements to configure a TestTrack server are:

• Ruby 2.2.3+
• Postgresql 9.4+
• NodeJS 14.19.1+

Installation

Option 1: Deploy to heroku:

Option 2: Deploy to your machine:

1. git clone https://github.com/Betterment/test_track
2. bundle install
3. yarn install
4. bundle exec rake db:setup
5. bin/dev

Either way:

At this point, you've got a working installation and can proceed to setting up the Rails client in order to create your first split.

Deployment

TestTrack is designed to deploy as a conventional 12-factor application.

Required environment variables:

• DATABASE_URL -- the url to your Postgresql database server

Configuration for JS Client

In order to use the JS client, you will need to specify the set of hosts that will be making CORS requests to your TestTrack server. That can be set via the WHITELIST_CORS_HOSTS environment variable.

export WHITELIST_CORS_HOSTS=yoursite.example.org,othersite.example.org

Configuration for Chrome extension

In order to use the TestTrack Chrome extension, you will need to set up the BROWSER_EXTENSION_SHARED_SECRET environment variable. Details.

Running specs

To run all specs:

bundle exec rspec

If you are interested in hitting a binding.pry somewhere while playing around with your development environment, you can use binding.pry_remote in your code. When reloading the page you expect to have the breakpoint, the browser should freeze and you should be able to connect to your console via running in your terminal pry-remote. Pry away as normal!

If you are curious to see your integration tests running in a non-headless manner, run your test this way:

$ CAPYBARA_DEBUG=1 bundle exec rspec <path to file>/<file>s

Managing your installation

There are a few things that you will need to do in the TestTrack application:

• Create Apps -- client applications that will manage splits on your TestTrack server
• Create Admins -- users that can access the admin features of the TestTrack server
• Manage splits using the admin features

Creating Apps

In order to create spilts in your client applications, you will need to register that client application with your TestTrack server. Run the following in a rails console.

> App.create!(name: "[myapp]", auth_secret: SecureRandom.urlsafe_base64(32)).auth_secret
=> "[your new app password]"

This is the password that you should plug into your client application's TEST_TRACK_API_URL.

Seeding Apps For Local Development

At Betterment, we run TestTrack in every environment, including our laptops, which enables engineers to override splits with the Chrome extension while they code.

TestTrack provides a Rake task to make it easier to set up apps that automatically get reloaded whenever you recreate your TestTrack database. If you want to add a rails app called widget_maker to TestTrack, run:

rake seed_app[widget_maker]

This will do three things:

• Find or create a file called db/seed_apps.yml
• Find or create an entry in it for your app name and set a randomly generated auth_secret
• Run rake db:seed, which reloads your seed apps into the database

Note that db/seed_apps.yml is .gitignored so you can run TestTrack locally without having a private copy of the test_track repository or having uncommitted changes on your local checkout. That way it's easier to contribute to TestTrack, and stay on the latest version of the open source product.

You can use a configuration management tool like boxen to install TestTrack and inject a custom seed_apps.yml file for your team.

Creating Admins

In order to access the admin features of the TestTrack server, you must create an Admin in your database. Run the following in a rails console.

> Admin.create!(email: "[e-mail]", password: "[something secret]")

Enabling attachments

Variants can be associated with metadata to describe their effects in human-readable terms. To enable variant screenshots, you'll need some additional configuration in environment variables.

• To store uploads on the local file system, set ATTACHMENT_STORAGE to local, and:
  • LOCAL_UPLOAD_PATH (optional, defaults to :rails_root/public/system/:class/:attachment/:id_partition/:style/:filename)
• To store uploads in S3, set ATTACHMENT_STORAGE to s3, and:
  • AWS_ACCESS_KEY_ID (required)
  • AWS_SECRET_ACCESS_KEY (required)
  • AWS_REGION (optional, defaults to us-east-1)
  • S3_ATTACHMENT_BUCKET (required)
  • S3_ATTACHMENT_PERMISSIONS (optional, defaults to private; see AWS canned ACLs)
  • S3_ATTACHMENT_PATH (optional, defaults to :class/:attachment/:id_partition/:style/:filename)

For more on how paths are constructed, see the paperclip documentation.

Advanced topics

Feature Gate Experience Sampling Weight

Feature gates report to your analytics provider differently relative to experiments. Where an experiment is assigned once per visitor and the assignment is recorded on the server for later reuse, a feature gate assignment is recalculated on the client side at each interaction. As a result, the analytics provider receives a different kind of event - feature_gate_experienced instead of split_assigned. These experience events are much higher velocity than assignment events because they occur each time a customer or backend codepath encounters the feature gate.

While this experience event information is valuable to developers who are evaluating the number of customers who have experienced a feature, for example during a slow feature rollout to establish confidence that error rate remains low, there is little value in having comprehensive records. Sampling a fraction of customer interactions provides the same signal developers need at much lower cost.

To switch to a sampling strategy, set the EXPERIENCE_SAMPLING_WEIGHT env var to a non-negative integer value as follows:

• The default is 1, which means that every experience event will be reported to analytics
• To disable reporting of feature gate assignments altogether, set it to 0
• A value of 10 tells clients to report experience events to analytics probablistically one out of ten times. Conformant (>= v4) TestTrack clients will then reduce their reporting rate accordingly across all feature gates.

Concepts

App

A Rails application that manages Splits on the TestTrack server.

Admin

A member of your team that administers the weightings of splits, deciding a the winning variant of a split, and uploading one-off visitor assignments.

Visitor

A person using your application.

Split

A feature for which TestTrack will be assigning different behavior for different visitors.

Split names must be strings and should be expressed in snake_case. E.g. homepage_redesign_late_2015_experiment, signup_button_color_experiment, or invite_button_enabled.

Variant

One the values that a given visitor will be assigned for a split, e.g. true or false for a classic A/B test or e.g. red, blue, and green for a multi-way split. Variants may be strings or booleans, and they should be expressed in snake_case.

Weighting

Variants are assigned pseudo-randomly to visitors based on their visitor IDs and the weightings for the variants. Weightings describe the probability of a visitor being assigned to a given variant in integer percentages. All the variant weightings for a given split must sum to 100, though variants may have a weighting of 0.

Experiment

Experiments are the standard flavor of splits in TestTrack. They are intended to be used for A/B testing, and the TestTrack server records visitors' experienced variants so that those visitors will continue to experience the same variant regardless of subsequent changes to the weightings of those variants via the admin interface.

Storing the variant a visitor experienced for an experiment also allows TestTrack to provide a consistent UX to a customer who experienced a new-to-them experiment before logging in on a new device, only to be recognized as an existing visitor upon sign-in. TestTrack will merge all variant assignments from the anonymous visitor into the authenticated visitor at sign-in as long as the authenticated visitor doesn't have conflicting assignments. In that case, the authenticated visitor's previous assignments win.

Feature Gate

As of TestTrack version 1.2, splits with names ending in the _enabled suffix will be treated as feature gates. Feature gates differ from experiments in that they are not intended to be used for analysis, and therefore it is not important that the user remain in the same variant throughout the entire split lifecycle. Feature gates are meant to be slow-rolled (incrementally increasing the percentage of customers experiencing the new feature), released en masse, or instantly rolled back.

To facilitate these smooth transitions and rapid toggles, the TestTrack server will not record variant assignments when a visitor experiences a split. This means that every time a visitor experiences a split, they will be deterministically (pseudorandomly) assigned to a variant based on their visitor ID and the name of the split. This will provide the customer with a stable variant given a constant split weighting, but probablistically increase the percentage of visitors experiencing the the true variant as the split weightings are increased via the admin panel, giving an admin full control over the feature's release.

IdentifierType

A name for a customer identifier that is meaningful in your application, typically things that people sign up as, log in as. They should be expressed in snake_case and conventionally are prefixed with the application name that the identifier is for, e.g. myapp_user_id, myapp_lead_id.

How to Contribute

We would love for you to contribute! Anything that benefits the majority of test_track users—from a documentation fix to an entirely new feature—is encouraged.

Before diving in, check our issue tracker and consider creating a new issue to get early feedback on your proposed change.

Suggested Workflow

• Fork the project and create a new branch for your contribution.
• Write your contribution (and any applicable test coverage).
• Make sure all tests pass (bundle exec rake).
• Submit a pull request.