Add ULN vs normal URL type validation when updating or creating a remote

[quba42]

Version
Affects all versions.

Describe the bug
pulp_rpm knows two types of remote. ULN and RPM. The url field for a ULN remote must always have a value starting with uln://. Conversely a RPM remote with a url starting with uln:// never makes sense, and cannot possibly work. We should add sanity validation to both remotes to prevent users from accidentally creating a remote of the wrong type when the intention is to create a ULN/RPM remote. For RPM remotes we should validate the URL does not start with uln://, for ULN we should validate that it does.

Without this validation users will only hit an error once they try to sync using the remote, and the error won't be very clear about what went wrong. We saw this situation with Katello users who decided to change URL type on a Katello repo, but Katello would update the existing remote, rather than destroying it and re-creating one of the other type without an error. We have fixed this on the Katello side, but it still seems like Pulp should throw a meaningful error at remote creation time, rather than a cryptic one at sync time.

To Reproduce
Create or update a ULN or RPM remote with a url of the wrong type. => Successfully creates an unusable remote.

Expected behavior
Throw an error stating clearly what was wrong about the user input.

Additional context
The related Katello issue: https://projects.theforeman.org/issues/37279

[quba42]

A relevant place in the sources: https://github.com/pulp/pulp_rpm/blob/main/pulp_rpm/app/serializers/repository.py#L295

[dralley]

This is reasonable. In hindsight though we maybe should have avoided having a separate ULNRemote type at all. We already have divergent behavior if you pass a file:// path to a normal remote, it could have been handled similarly with uln://

[quba42]

This is reasonable. In hindsight though we maybe should have avoided having a separate ULNRemote type at all. We already have divergent behavior if you pass a file:// path to a normal remote, it could have been handled similarly with uln://

That would have made some things easier. On the other hand it also has advantages not to overload a single remote type with many different special meanings and fields that are only meaningful to a small subset of users. ULN is a pretty singular use case that in addition to the special upstream url values also adds the uln_server_base_url. This is perhaps similar to the SLES specific fields already on the normal remote. I am torn on which approach is best. Which probably means simply sticking with the choices we made in the past.

[pedro-psb]

To tight this up, can we validate that RpmRemote acceptable urls are only http, https and file?