Constraints in 3.0.3 have incompatible combination of pandas & sqlalchemy

[potiuk]

Discussed in #53392

^{Originally posted by aa-matthias July 15, 2025}

Apache Airflow version

3.0.3

If "Other Airflow 2 version" selected, which one?

No response

What happened?

constraints for 3.0.3 have pandas 2.2.3 and SQLAlchemy 1.4.54 causing this issue now:

pandas-dev/pandas#57049

What you think should happen instead?

No response

How to reproduce

Upgrade to 3.0.3 using constraints for 3.0.3 python3.12

Operating System

Ubuntu 24.04

Versions of Apache Airflow Providers

No response

Deployment

Virtualenv installation

Deployment details

No response

Anything else?

No response

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[potiuk]

Converted it back from discussion for posterity as it turned out to be a real issue. So posting here with details of reasoning and workaround for anyone who looks for a solution.

The real issue was that our postgtres and common.sql providers should have postgres and sql-other extra of pandas added in their dependencies. This has been added in #53535 and will take effect on main and 3.0.4 once the next wave of providers gets released (next week).

I do not feel there is a need to update constraints for 3.0.3 and rebuild images because it seem to affect small number of users and there is an easy workaround:

1. you can copy airflow constraints for 3.0.3 and change pandas version to 2.1.4 and numpy 1.26.4 and use that constraints file instead of the one from github
2. when you are using airflow image you can build your own image and downgrade pandas:

FROM apache/airflow:3.0.3
RUN pip install pandas==2.1.4

This issue will be fixed permanently in Airflow 3.0.4