Class: Gitlab::ExperimentFeatureRollout

Inherits:
Gitlab::Experiment::Rollout::Percent
  • Object
show all
Defined in:
lib/gitlab/experiment_feature_rollout.rb

Instance Method Summary collapse

Instance Method Details

#enabled? ⇒ Boolean

For this rollout strategy to consider an experiment as enabled, we must:

  • have a feature flag yaml file that declares it.

  • be in an environment that permits it.

  • have the feature flag in any state of enablement (can be fully or partially rolled out with percentages, actors, etc.)

Returns:

  • (Boolean) 


12
13
14
15
16
17
18
 
# File 'lib/gitlab/experiment_feature_rollout.rb', line 12

def enabled?
  return false unless feature_flag_defined?
  return false unless available?
  return false unless ::Feature.enabled?(:gitlab_experiment, type: :ops)

  feature_flag_instance.state != :off
end

#execute_assignment ⇒ Object

For assignment we first check to see if our feature flag is enabled for this rollout instance. Feature.enabled? calls below #flipper_id to get a unique identifier. By default #flipper_id returns ‘Experiment;#Gitlab::ExperimentFeatureRollout.experimentexperiment.id` where experiment.id is an anonymous SHA generated using details of the experiment. This ensures deterministic assignment - same actor always gets same result.

If the Feature.enabled? check is false, we return nil implicitly, which will assign the control. If it returns true, we call super to assign a variant based on the experiment’s distribution rules (if specified) or evenly across all non-control variants (default behavior of Gitlab::Experiment::Rollout::Percent).

Since we exclude :control from #behavior_names, the feature flag’s rollout percentage determines the split: e.g., 30% flag rollout means 70% get control (flag disabled) and 30% get variants (flag enabled).



34
35
36
 
# File 'lib/gitlab/experiment_feature_rollout.rb', line 34

def execute_assignment
  super if ::Feature.enabled?(feature_flag_name, self, type: :experiment)
end

#flipper_id ⇒ Object

This is what’s provided to the Feature.enabled? call that will be used to determine experiment inclusion. An experiment may provide an override for this method to make the experiment work on user, group, or projects.

For example, when running an experiment on a project, you could make the experiment assignable by project (using chatops) by implementing a flipper_id method in the experiment:

def flipper_id

context.project.flipper_id

end

Or even cleaner, simply delegate it:

delegate :flipper_id, to: -> { context.project }



54
55
56
57
58
 
# File 'lib/gitlab/experiment_feature_rollout.rb', line 54

def flipper_id
  return experiment.flipper_id if experiment.respond_to?(:flipper_id)

  "Experiment;#{id}"
end