Progressive Delivery

Progressive Delivery is an advanced software release strategy that extends upon Continuous Delivery to involve techniques like feature flags, A/B testing, canary releases, and dark launches.

It allows you to roll out new features gradually to a subset of your users rather than a big bang release to everyone. This enables safer deployments, better user experience, and a more controlled approach to releasing software.

Benefits

• Risk mitigation: By rolling out features to a small group of users initially, you can detect issues early without impacting your entire user base.
• User experience: You can gather user feedback during the early stages of the release, making necessary adjustments before the full-scale rollout.
• Performance monitoring: Gives you room for monitoring of how new updates impact system performance and user behavior before you decide for wider rollout.
• Rapid iteration: Faster feedback loops allow for quick iterations, letting you adapt to user needs swiftly.
• Compliance & security: Easier to manage features in compliance with regulatory requirements (think of countries with strict data privacy laws like GDPR).

Defining gradual rollouts

Featurevisor allows you to define your features' rollout rules declaratively including their targeting conditions, rollout percentage, and more.

A very simple feature flag can be defined as follows:

# features/myFeature.yml
description: My feature's description here
tags:
  - all

bucketBy: userId

environments:
  production:
    rules:
      - key: "1"
        segments: "*" # everyone
        percentage: 10

In above example, we are rolling out the feature to only 10% of all our users in production environment.

We could have narrowed it down even further by using segments to target only a specific group of users.

# features/myFeature.yml
description: My feature's description here
tags:
  - all

bucketBy: userId

environments:
  production:
    rules:
      - key: "1"
        segments: germany
        percentage: 10

Here we are defining that the feature should be exposed to 10% of users in Germany only, and not targeting 10% of all users worldwide like we did in previous example.

Defining experiments

We could also achieve something similar using A/B tests, where we can define multiple variations of a feature and assign a percentage to each variation.

A/B testing would be more appropriate if we wish to measure two or more different variations of our feature, rather than a simple on/off toggle.

# features/myOtherFeature.yml
description: My other feature's description here
tags:
  - all

bucketBy: userId

variations:
  # default behaviour
  - value: control
    weight: 50

  # new behaviour to test
  - value: treatment
    weight: 25

  - value: anotherTreatment
    weight: 25

environments:
  production:
    rules:
      - key: "1"
        segments: "*" # everyone
        percentage: 100

In above example, we are rolling out the feature to 100% of all our users in production environment, but 50% of them will see the default behaviour (control), and the other 50% will see two new different behaviour (treatment and anotherTreatment).

The weight distribution of the variations can be tweaked to your liking, and you can add as many variations as you want.

Conclusion

Featurevisor provides the mechanism you need to implement a Progressive Delivery strategy effectively, both with simple feature flags and also with A/B tests.

Its blend of feature management, instant updates, and GitOps workflow makes it an approachable tool for engineering teams looking to improve their release processes, reduce risk, and accelerate time-to-market.