Upgrading Airflow 2 to Airflow 3 - A Checklist for 2026

In a detailed guide to upgrading Airflow deployments from version 2 to 3, the text emphasizes the strategic importance of making this transition due to the upcoming end of life for Airflow 2 in April 2026, which poses security risks from unpatched vulnerabilities. The new version introduces long-awaited features such as automatic DAG versioning, event-driven scheduling, and an enhanced React-based UI, all designed to boost data engineers' productivity and enable new use cases. The guide advises careful planning for the upgrade, recommending an initial upgrade to Airflow 2.11 to capitalize on deprecation warnings before moving to Airflow 3.1, which offers improved stability over its predecessor. It outlines the need to clean metadata databases, pin Python packages, and ensure compatibility with Python 3.11+, alongside updating configuration files and employing tools such as ruff to automate DAG code updates. A significant architectural change in Airflow 3 is the removal of direct access to the metadata database from within tasks, requiring developers to adapt their code to use the Airflow API or Python client instead. The guide stresses testing all changes in a local environment before deploying to production to avoid disruptions, especially during high-volume periods.