Unit Testing with Fabra

Fabra is designed to be testable by default. You don't need to spin up Docker containers or mock complex database interactions to test your feature logic.

By using the default configuration (DuckDB + In-Memory), you can treat Fabra as a library rather than a service during your test suite execution.

The Strategy: "Trojan Horse" Testing

Instead of asking your infra team for a Kubernetes cluster, start by using Fabra to mock feature data in your local unit tests. This gives you immediate value (cleaner tests) with zero operational overhead.

Example: Testing a Fraud Model

Imagine you have a FraudDetector class that relies on features.

# fraud.py
from fabra import FeatureStore
import pandas as pd

class FraudDetector:
    def __init__(self, store: FeatureStore):
        self.store = store

    async def predict_risk(self, user_id: str) -> float:
        # Fetch features purely from Python/Mock data
        features = await self.store.get_online_features(
            "User", user_id, ["transaction_count_1h", "last_login_country"]
        )

        # Simple heuristic
        if features.get("transaction_count_1h", 0) > 10:
            return 0.9
        return 0.1

Writing the Test

You can write a standard pytest test that populates the in-memory store before running the logic.

# test_fraud.py
import pytest
from fabra import FeatureStore, entity, feature
from fraud import FraudDetector

@pytest.fixture
def store():
    # 1. Initialize ephemeral store (In-Memory by default)
    store = FeatureStore()

    # 2. Define entities/features inline for the test
    @entity(store)
    class User:
        user_id: str

    @feature(User)
    def transaction_count_1h(uid: str) -> int:
        return 0 # Default, or we override via set_online_features

    @feature(User)
    def last_login_country(uid: str) -> str:
        return "US"

    return store

@pytest.mark.asyncio
async def test_high_risk_scenario(store):
    # 3. Seed the "Online Store" with mock data
    # No Redis required! Just a Python dict under the hood.
    await store.online_store.set_online_features(
        entity_name="User",
        entity_id="u1",
        features={"transaction_count_1h": 15, "last_login_country": "US"}
    )

    detector = FraudDetector(store)
    risk = await detector.predict_risk("u1")

    assert risk == 0.9

Why this matters?

1. Speed: Tests run in milliseconds.
2. Isolation: No network calls to external DBs.
3. Portability: Runs in any CI environment (GitHub Actions) with just pip install fabra-ai.

Start small by using Fabra to organize your test data. When you're ready, the exact same code deploys to production with Redis.