I am all for it. As we saw already and we see it more in the future -
moving code of out of Airflow core to provider and having separate
provider's release cycle and lifecycle is generally beneficial:

* dependencies can be more decoupled - even if we pin FAB with a particular
version of provider, our users can freely change provider version
without updating airflow and the other way round - update airflow without
changing provider version (thus without changing FAB).
* we will be able to remove FAB in the future from active maintenance -
similarly to discussed dask executor we can stop maintaining FAB provider
in the future (no plans now of course - but it might happen - especially if
we will implement a way to migrate current USER/ROLE of FAB to something
else

I think we've learned already a lot on how to manage lifecycle of providers
and have a number of tools and processes, so I think it will not be too
problematic.

J.



On Tue, Nov 14, 2023 at 9:46 PM Beck, Vincent <[email protected]>
wrote:

> Hello,
>
> I am sending this email to gather feedbacks/concerns on what is going to
> happen regarding AIP-56 (
> https://cwiki.apache.org/confluence/display/AIRFLOW/AIP-56+Extensible+user+management).
> The majority of the work related to AIP-56 is now completed. As a summary,
> all the code related to user authentication and user authorization has been
> packaged in a new component called "FAB auth manager" (
> https://github.com/apache/airflow/blob/main/airflow/auth/managers/fab/fab_auth_manager.py).
> Now, all user authentication and user authorization operations are done in
> this FAB auth manager. The purpose of having this new component is to be
> able to plug (and/or create) another "auth manager" if desired.
>
> For simplification reasons, this FAB auth manager is still in core Airflow
> under "airflow.auth.managers.fab" (
> https://github.com/apache/airflow/tree/main/airflow/auth/managers/fab)
> but the ultimate plan is to move the module "airflow.auth.managers.fab" to
> a new provider. Of course, this new provider would still be installed by
> default in Airflow and will have no impact for the users.
>
> An issue had been created for that work and a discussion has started but I
> wanted to increase the audience and potentially get more feedbacks/concerns
> before actually doing so: https://github.com/apache/airflow/issues/32210.
> In this issue you will also find the motivations behind moving this code to
> a new provider.
>
> Cheers,
> Vincent
>

Reply via email to