OpenFeature Python SDK

Warnings

• breaking Python 3.8 support was dropped in OpenFeature Python SDK v0.8.0. Applications must be running Python 3.9 or newer to upgrade to v0.8.0 and beyond.
  Fix: Upgrade Python environment to 3.9 or higher. For example, use `python -m venv .venv` with Python 3.9+.
• breaking The signature of the `finally_after` hook stage changed in v0.7.5. It now requires `evaluation_details` as the second argument.
  Fix: Update any custom hooks implementing `finally_after` to accept `evaluation_details` as the second argument: `finally_after(flag_details: FlagResolutionDetails, evaluation_details: dict) -> None`.
• gotcha The core `openfeature-sdk` package does not include any flag providers by default. A provider, which connects to a feature flag management system (like Flagd, LaunchDarkly, Split.io), must be installed and configured separately.
  Fix: Install a specific provider package (e.g., `pip install openfeature-provider-flagd`) and then configure it using `api.set_provider(YourProvider())`.
• gotcha If no provider is explicitly set, OpenFeature will default to a 'No-Op' provider. This provider always returns the default value provided in the flag evaluation call, which can lead to unexpected behavior if you expect flags to be dynamic.
  Fix: Always ensure a provider is set via `api.set_provider()` early in your application's lifecycle, before attempting to evaluate flags, unless intentionally using the No-Op behavior.
• gotcha When providing an `EvaluationContext`, ensure a `targeting_key` is always included. Many providers rely on this for correct flag evaluation and targeting logic, and its absence can lead to incorrect or default flag values.
  Fix: Populate the `targeting_key` in the `EvaluationContext` with a unique identifier relevant to the user or entity being targeted, e.g., `EvaluationContext(targeting_key='user123')`.

Install

• pip install openfeature-sdk Install core SDK
• pip install openfeature-provider-flagd Example: Install a provider (e.g., Flagd)

Imports

• api

  from openfeature import api

• InMemoryProvider

  from openfeature.provider.in_memory_provider import InMemoryProvider

  Required for a basic, self-contained example without external providers.
• EvaluationContext

  from openfeature import EvaluationContext

  EvaluationContext is a top-level import from the openfeature package itself.

Quickstart

This quickstart demonstrates how to initialize the OpenFeature SDK with a basic in-memory provider, create a client, and evaluate a boolean and a string feature flag. It also shows how to provide an evaluation context for flag targeting. In a production environment, you would replace `InMemoryProvider` with a specific feature flag management system's provider (e.g., `FlagdProvider`, `LaunchDarklyProvider`).

from openfeature import api, EvaluationContext
from openfeature.provider.in_memory_provider import InMemoryFlag, InMemoryProvider
import os

# Configure a simple in-memory provider for demonstration
# In a real application, you would typically use an external provider (e.g., Flagd, LaunchDarkly)
my_flags = {
    "new-feature-enabled": InMemoryFlag("on", {"on": True, "off": False}),
    "welcome-message": InMemoryFlag("greeting", {"greeting": "Hello, OpenFeature!", "default": "Welcome!"})
}
api.set_provider(InMemoryProvider(my_flags))

# Create a client, optionally with a domain
client = api.get_client()

# Define an evaluation context for targeting
user_id = os.environ.get('USER_ID', 'default_user')
eval_context = EvaluationContext(targeting_key=user_id, attributes={"region": "us-east-1"})

# Evaluate a boolean flag
is_new_feature_enabled = client.get_boolean_value("new-feature-enabled", False, eval_context)
if is_new_feature_enabled:
    print(f"New feature is enabled for user {user_id}!")
else:
    print(f"New feature is disabled for user {user_id}.")

# Evaluate a string flag
message = client.get_string_value("welcome-message", "Fallback Message", eval_context)
print(f"Retrieved message: {message}")

# It's good practice to shut down providers gracefully
api.shutdown()