Dear Airflow Community, I am thrilled to announce the availability of Apache Airflow 3.0.0.alpha1 for testing! Airflow 3.0 marks a significant milestone as the first major release in over four years, introducing improvements that enhance user experience, task execution, and system scalability.
First, a few caveats: This is an alpha release, so do not run it in production. It may contain significant issues, and you will likely need to reset your database between this and subsequent alpha, beta or release candidate versions. (Consider yourself warned!) This release is intended for Airflow developers and early adopters to test the build and start preparing for Airflow 3.0.0. This is not an official releaseโthat will happen when we create a release candidate and hold a vote. The expected timeline for the first release candidate is the week of 2025-03-31, but we encourage early feedback to help stabilize the release. What's new in 3.0.0.alpha1? Airflow 3.0.0 introduces significant enhancements and breaking changes. Notable Features DAG versioning & Bundles Airflow now tracks DAG versions, offering better visibility into historical DAG changes and execution states. The introduction of DAG Bundles ensures tasks run with the correct code version, even as DAGs evolve. Modern Web Application The UI has been rebuilt using React and a complete API-driven structure, improving maintainability and extensibility. It includes a new component-based design system and an enhanced information architecture. A new React-based plugin system supports custom widgets, improved workflow visibility, and integration with external tools. Task Execution Interface Airflow 3.0 adopts a client / server architecture, decoupling task execution from the internal meta-database via API-based interaction. This allows for remote execution across networks, multi-language support, enhanced security, and better dependency management. The Edge Executor further enables seamless remote task execution without direct database connections. Data Assets & Asset-Centric Syntax Airflow 3.0 enhances dataset management by introducing Data Assets, expanding beyond tables and files to include ML models and more. Assets can be explicitly defined using the @asset decorator, simplifying tracking and dependencies. External Event-Driven Scheduling Airflow now supports event-driven DAG triggers from external sources like message queues. This builds upon dataset scheduling and enhances integration with the external data ecosystem. For a more comprehensive list of new features, please see the 3.0.0apha1 release notes: https://github.com/apache/airflow/blob/3.0.0a1/RELEASE_NOTES.rst#airflow-300alpha1-2025-01-31 For a list of the breaking changes, please see visit: https://cwiki.apache.org/confluence/x/9pCMEw Known issues in 3.0.0.alpha1: - AIP-72 - Task Execution Interface - DB isolation in tasks is not enforced yet - KubernetesExecutor does not yet use the TaskExecution Interface - The following does not work on LocalExecutor or CeleryExecutor: Dynamic Task Mapping, Skip based tasks (e.g. branch or skip operators), and Task callbacks. - OpenLineage listeners are not yet functional in this alpha release. - AIP-38 - Modern Web Application - The new UI has limited functionality at this time and is still being enhanced until GA. However, feedback on the UX flow is appreciated at this time. - The underlying FastAPI API server, including the new UI and public API, does not yet have auth. - UI may be slow if you have a lot of DAGs or very big DAGs - Notable areas of functionality that are not yet functional in the UI: DAG Versioning, Data Assets, Backfills, and the DAG Gantt View. - AIP-63 - DAG Versioning - Callbacks are not yet working on the DAG Processor (e.g. DAG level callbacks) - The concept of a separate `--subdir` in the DAG processor is being replaced with separate DAG bundles, but currently the DAG processor always parses all DAG bundles. - AIP-82 - Event Based Scheduling - Scheduling using some triggers, for example file-based, can lead to duplicate DAG runs. Where to get it? The alpha snapshot is available at: https://dist.apache.org/repos/dist/dev/airflow/3.0.0a1/ - apache-airflow-3.0.0a1-bin.tar.gz: Binary Python "sdist" snapshot. - apache_airflow-3.0.0a1-py3-none-any.whl: Binary Python wheel snapshot. This snapshot has not been published to PyPI. Also present are alpha releases for apache-airflow-task-sdk, apache-airflow-providers-standard, apache-airflow-providers-fab, and apache-airflow-providers-celery, which you will also need for the 3.0.0a1 to work. These have also not been published to PyPI. Public Keys & Verification Public keys for verification are available at: https://www.apache.org/dist/airflow/KEYS Instructions: You can build a virtualenv that installs this alpha, and other required packages (e.g. task sdk), like this: ``` uv venv uv pip install --find-links https://dist.apache.org/repos/dist/dev/airflow/3.0.0a1/ apache-airflow==3.0.0a1 apache-airflow-task-sdk==1.0.0a1 apache-airflow-providers-standard==0.1.0a1 apache-airflow-providers-fab==2.0.0a1 apache-airflow-providers-celery==3.11.0a1 ``` Below are some of the changes youโll need to consider in order to run this Alpha release: - The standalone DAG processor is now required, and can be started with `airflow dag-processor`. - The new UI and public API is started by running `airflow fastapi-api`, and the UI is available on port `9091` and path `/webapp`. - Depending on your deployment setup, you may need to set the `[workers] execution_api_server_url` config option. This defaults to ` http://localhost:9091/execution/`. Get Involved We encourage the community to test this release and report any issues or feedback. Your contributions help us ensure a stable and reliable Airflow 3.0.0 release. A huge thank you to all the contributors who have worked on this milestone release! Best, Utkarsh and Jed
