Proposal for different fix to 2.10 regression

[dwreeves]

Description

Fixes Airflow 2.10 regression / compat

Related Issue(s)

This is a fix for #1161, and is designed to be a different proposal to the changes in #1162.

Ultimately the issue that popped up in 2.10 relates to how the MRO was impacted by the Mixin class.

The simplest, most robust fix to make the MRO work properly is to make each mixin a subclass of BaseOperator. This doesn't require the unintuitive hack of needing an __init__() call to shuffle the MRO around, and it makes things more easily maintainable going forward, I think.

A note: This PR is merging to test-af-2.10, not main.

Tested for 2.10.

I also removed an unnecessary __init__ for the DbtTestMixin while I was at it.

Checklist

• I have made corresponding changes to the documentation (if required)
• I have added tests that prove my fix is effective or that my feature works

[dwreeves]

One more thing, after all of this gets resolved, we should either do this for each minor version from 1.3+, or alternatively put an upper bound on each previously released version as patch releases. I believe based on the Slack that there are a lot of users on 1.3 still.

[pankajastro]

Looks good. thanks @dwreeves for looking into.

[pankajastro]

cc: @pankajkoti we should include this in 1.6.0

[pankajastro]

we should either do this for each minor version from 1.3+, or alternatively put an upper bound on each previously released version as patch releases.

I'm not sure about this at the moment, but I'm open to hearing what others think. The user who has reported this issue using Cosmos 1.5.1.
Update: We may not need to patch it

• This error happens only for Airflow 2.10 and I believe users should upgrade Cosmos too
• I think we do not follow the practice of backporting fixes for this project?

I believe based on the Slack that there are a lot of users on 1.3 still.

Which Slack are you referring to?

[pankajastro]

Also, the CI is not running on this PR, possibly because the base branch is not main.

[pankajastro]

I'm merging this to branch test-af-2.10 to validate the CI check

[pankajastro]

Look like CI is not happy

[dwreeves]

@pankajastro That's weird. 😕 I did run tests locally for 2.10 and it worked, but I see now that 2.4 tests break.

Let's rollback on the other PR and we can merge this if/when it is ready. There is a higher criticality to just releasing a fix than there is to fixing it in a "good way."

Sorry for making this so complicated.

[dwreeves]

RE: upgrading other versions-- Fair enough, we won't backport then.

My one concern was that Airflow issues like this can be annoying because if a user upgrades their Airflow, downgrading Airflow to rollback a change can be very annoying because of the Postgres db requiring a downgrade, so if they are stuck on an old Cosmos version for some reason and they upgrade Airflow, now they have 2 problems (2.10 broke things for them + they cannot easily downgrade, but also, Cosmos cannot be upgraded either).

Which Slack are you referring to?

Official Airflow slack, dbt channel.