Groups

Groups enable you to run mutually exclusive experiments.

Mutually exclusive experiments#

Let's say you have two experiments defined as firstFeature and secondFeature.

You want to run them both together in production, but you do not want to expose both of them together to any single user.

That's what "mutually exclusive" means here. They will never overlap for the same user.

You can take advantage of groups in Featurevisor to achieve this exclusions list.

Create a group#

We can create groups by creating a new file in the groups directory:

description: My exclusion group
slots:
  - feature: firstFeature # referring features/firstFeature.yml
    percentage: 50
  - feature: secondFeature # referring features/secondFeature.yml
    percentage: 50

The name of the group file is not used anywhere, so you can name it however you want.

Slots#

A group consists of multiple slots.

Each slot in a group defines a feature (by its key) and a percentage value (from 0 to 100). All the percentage values of slots in a group must add up to 100.

In the example above, we have two slots, each with a percentage value of 50. This means that any user once bucketed can only fall in to one of the slots, and not both.

The first 50% of the users will be exposed to firstFeature, and the other 50% will be exposed to secondFeature.

Impact on affected features#

The slot's percentage determines what's the maximum percentage value you can use in your Feature's rollout rules.

In the example above, the maximum percentage value you can use in firstFeature and secondFeature is 50.

Here's how the firstFeature would look like in that case:

# ...
rules:
  rules:
    - key: everyone
      segments: '*'
      percentage: 50 # can be any value between 0 and 50

Bucketing#

Whenever a feature is evaluated, it goes through a bucketing process where the user is assigned a number between 0 and 100. If the bucketed number falls into the range of any slot, it means that the user is exposed to that particular feature only in the group.

Read more in Bucketing.

Limitations#

• A feature can only belong to a maximum of one group at a time
• A feature cannot repeat in the same group's slots (this will be supported in a future version)
• Both the required and dependent features cannot coexist in the same group

Guides#

To maintain consistent bucketing, you are advised to:

• Keep mutual exclusions in mind before creating the feature and the group.
• Create them together, and then add the rollout rules to the feature.
• Plan the percentage distribution in your slots early when creating the group, and do not change those values afterwards.
• Start with a bigger slot for your feature, even if you do not want to use the full percentage value for your feature's rules. You can slowly increase it at feature's rule level later.
• You can set feature: false if you want to remove a feature from a group's slot without any replacement feature.

Rely on linting to catch any mistakes.

Archiving#

Groups cannot be archived. If you don't need them any more, you can delete their YAML files.