Airflow 2 End of Life

Apache Airflow 2 reached end of life in April 2026. The project no longer ships fixes for the 2.x line, and provider packages keep moving forward against Airflow 3. The scheduler you run tonight still works; it is simply nobody's job to fix it any more.

What actually changes

• No more security fixes. A vulnerability found in 2.x code stays open. How much that matters depends on how exposed your webserver and workers are.
• Provider packages drift away. New provider releases target Airflow 3. Pinning old versions works until a cloud API you depend on changes underneath them.
• Managed platforms move on. Amazon MWAA and Google Cloud Composer offer Airflow 3 environments, and older versions follow their own deprecation schedules.
• Hiring and knowledge get harder. Fewer engineers want to learn a line that is no longer maintained.

Your options

1. Migrate to Airflow 3. The right answer for most active estates. It is real work: DAGs move to the Task SDK, task code that reads the metadata database has to change, and deprecated imports go. See our migration approach.
2. Freeze and shrink. For estates that are mostly legacy, lock the environment down, retire the DAGs nobody needs, and migrate the remainder on a schedule that suits you.
3. Retire DAGs rather than port them. Every estate has DAGs that feed reports nobody reads. Retiring them is the cheapest migration there is.
4. Move part of the estate elsewhere. A few teams genuinely belong on a different orchestrator. If that is you, we will say so.

Where to start

An estate assessment gives you the inventory and the per-DAG call. While you decide, we will keep what runs today running rather than hold the migration hostage. Talk to an engineer.