airflow-provider-openfeature¶
Feature flags for Apache Airflow. Ramp a change across your DAGs, measure it, and revert with a flag, not a redeploy.
Feature flags let you control how your Airflow platform behaves at runtime, across many DAGs, without editing them or redeploying. A platform team can move a subset of tasks to a different pool, queue, or executor, ramp a worker or executor migration, or flip a kill switch during an incident, centrally and without touching anyone's DAG. A DAG author can gate a code path or A/B a model inside a task. Both go through OpenFeature, so it works with the flag backend you already run: flagd, LaunchDarkly, GrowthBook, Unleash, Statsig, or an in-house engine.
Install¶
pip install airflow-provider-openfeature
It is a standard PyPI package, so uv works the same way:
uv pip install airflow-provider-openfeature.
Pick your path¶
I run the Airflow platform. Install the provider, enable the policy, and move a subset of DAGs to a different pool, queue, or executor from a flag, without touching any DAG. → Getting started, about 5 minutes on a local Airflow.
I write DAGs. Gate a code path or A/B a model inside a task with one call, no platform access needed. → Use cases and the example DAGs.
More¶
- Running a rollout: the day-to-day ramp, measure, and revert loop.
- Case study: canary a faster pipeline step end to end, with a real backend.
- Measurement: close the loop with your experiment platform or warehouse.
- Architecture: the flow and the surfaces it registers.
- Extending: add a backend.
The README on GitHub has the demos and the full picture.