Implement a user profile service for the Python SDK

3 min

Use a User Profile Service to persist information about your users and ensure variation assignments are sticky. For example, if you are working on a backend website, you can create an implementation that reads and saves user profiles from a Redis or Memcached store.

In the Python SDK, there is no default implementation. Implementing a User Profile Service is optional and is only necessary if you want to keep variation assignments sticky even when experiment conditions are changed while it is running (for example, audiences, attributes, variation pausing, and traffic distribution). Otherwise, the Python SDK is stateless and relies on deterministic bucketing to return consistent assignments.

If the User Profile Service does not bucket a user as you expect, then check whether other flags are overriding the bucketing. For information, see How bucketing worksHow bucketing works.

Implement a service

To implement a custom User Profile Service, refer to the following code samples. The service needs to expose two functions with the following signatures:

lookup – Takes a user ID string and returns a user profile matching The following schema.

save – Takes a user profile and persists it.

If you want to use the User Profile Service purely for tracking purposes and not sticky bucketing, you can implement only the save method (always return None from lookup).

from optimizely import user_profile
from optimizely import optimizely

class MyUserProfileService(user_profile.UserProfileService):
    def lookup(self, user_id):
        pass
        # Retrieve and return user profile

    def save(self, user_profile):
        pass
        # Save user profile

optimizely_client = optimizely.Optimizely(datafile, user_profile_service=MyUserProfileService())

The following code example shows the JSON schema for the user profile object.

Use experiment_bucket_map to override the default bucketing behavior and define an alternate experiment variation for a given user. For each experiment that you want to override, add an object to the map. Use the experiment ID as the key and include a variation_id property that specifies the desired variation. If there isn't an entry for an experiment, then the default bucketing behavior persists.

In The following example, ^[a-zA-Z0-9]+$ is the experiment ID.

{
  "title": "UserProfile",
  "type": "object",
  "properties": {
    "user_id": {"type": "string"},
    "experiment_bucket_map": {"type": "object",
                              "patternProperties": {
                                 "^[a-zA-Z0-9]+$": {"type": "object",
                                                    "properties": {"variation_id": {"type":"string"}},
                                                    "required": ["variation_id"]}
                               }
                             }
  },
  "required": ["user_id", "experiment_bucket_map"]
}

The Python SDK uses the User Profile Service you provide to override Optimizely Feature Experimentation's default bucketing behavior in cases when an experiment assignment has been saved.

When implementing your own User Profile Service, you should loading the user profiles into the User Profile Service on initialization and avoiding performing expensive, blocking lookups on the lookup function to minimize the performance impact of incorporating the service.

When implementing in a multi-server or stateless environment, you should using this interface with a backend like Cassandra or Redis. You can decide how long you want to keep your sticky bucketing around by configuring these services.