Allow keeping PostgresHook SQLAlchemy engines on psycopg2

[sehajsandhu123]

Since postgres provider 7.0.0, PostgresHook builds its SQLAlchemy engines on psycopg (v3) whenever SQLAlchemy 2.x is installed, and the 7.0.0 changelog notes there is no connection or configuration option to keep hooks on psycopg2.

This silently changes semantics for Dag code that uses hook.get_uri() or hook.get_sqlalchemy_engine(). The psycopg (v3) dialect renders typed bind casts, so string parameters that PostgreSQL used to coerce implicitly now fail server-side. The most common casualty is pandas.DataFrame.to_sql into a table with a uuid column, which worked for years on psycopg2 and now fails with:

psycopg.errors.DatatypeMismatch: column "id" is of type uuid but expression is of type character varying
LINE 1: INSERT INTO orders (id, item) VALUES ($1::VARCHAR, $2...

This PR adds the missing opt-out by honoring the sqlalchemy_scheme connection extra (and an equivalent hook parameter), the same convention OdbcHook and DbApiHook.dialect_name already follow. PostgresHook even lists sqlalchemy_scheme in ignored_extra_options, so it is already excluded from libpq connect args — it just wasn't used when building the URL. With this change, a connection that needs the old behaviour can set:

{"sqlalchemy_scheme": "postgresql+psycopg2"}

The value is validated to be postgresql or postgresql+<driver> with no : or /, so a connection extra can't smuggle in a different URL. Nothing changes for connections that don't set it.

Tested against PostgreSQL 17 with pandas 3.0.5 / SQLAlchemy 2.0.51: df.to_sql into a uuid column reproduces the error above on the default psycopg3 engine and succeeds with the extra set. Unit tests cover the override, parameter precedence, get_uri propagation, and rejection of invalid schemes.

related: #71977

---

Was generative AI tooling used to co-author this PR?

• Yes — Claude Code (Fable 5); all changes reviewed and validated by the author

Generated-by: Claude Code (Fable 5) following the guidelines

[boring-cyborg]

Congratulations on your first Pull Request and welcome to the Apache Airflow community! If you have any issues or are unsure about any anything please check our Contributors' Guide
Here are some useful points:

• Pay attention to the quality of your code (ruff, mypy and type annotations). Our prek-hooks will help you with that.
• In case of a new feature add useful documentation (in docstrings or in docs/ directory). Adding a new operator? Check this short guide Consider adding an example Dag that shows how users should use it.
• Consider using Breeze environment for testing locally, it's a heavy docker but it ships with a working Airflow and a lot of integrations.
• Be patient and persistent. It might take some time to get a review or get the final approval from Committers.
• Please follow ASF Code of Conduct for all communication including (but not limited to) comments on Pull Requests, Mailing list and Slack.
• Be sure to read the Airflow Coding style.
• Always keep your Pull Requests rebased, otherwise your build might fail due to changes not related to your commits.
  Apache Airflow is a community-driven project and together we are making it better 🚀.
  In case of doubts contact the developers at:
  Mailing List: dev@airflow.apache.org
  Slack: https://s.apache.org/airflow-slack

[Dev-iL]

The implementation looks correct and the tests are helpful.

I'm conflicted about whether we want to support this functionality. On the one hand, we should be careful about adding workarounds for the behavior of other libraries. On the other hand, because this an explicit, non-default escape hatch for users migrating to ppg3, it might be justified.

My main request is related to regression tests: please ensure CI reproduces the pandas DataFrame.to_sql() -> uuid failure with ppg3 (via an xfail test) but passes on ppg2.

Please include links to any upstream pandas/SQLAlchemy discussions or issues related to this topic so we can track whether this remains necessary long-term. Examples:

• pandas-dev/pandas#63511
• apache/arrow#50325

[sehajsandhu123]

@Dev-iL Thanks for the review, I've added the requested regression test and upstream discussion links.

I understand the caution, but this is an opt-in param, off by default. And it isn't really a new one — the convention already exists in the codebase: OdbcHook and MsSqlHook both support sqlalchemy_scheme for per-connection driver choice (pymssql vs pyodbc), and MySqlHook does the same through its client extra.

Also, the 7.0.0 notes already give the metadata DB an explicit psycopg2 opt-out via sql_alchemy_conn — hook connections were the only part of this migration without one. Since it's off by default and the strict xfail will tell us when upstream makes it unnecessary, it should be easy to retire too.

[boring-cyborg]

Awesome work, congrats on your first merged pull request! You are invited to check our Issue Tracker for additional contributions.