Add modelling for "Errata" content

[dralley]

closes #3201
https://pulp.plan.io/issues/3201

[dralley]

With UpdateCollections and UpdateReferences, much like with "depends" we can either go the route of JSON representation, or make actual models with a foreign key that provides a related_name of "collections".

Do we have a preference? Is there value in modelling it vs. shoving it into JSON?

[dralley]

Pulp 2 did model "pkglist" / collections separately

[goosemania]

I suggest to go with the simplest solution at this point aka JSON. We can move to models later if we need to.
I hope we won't take the approach similar to pulp 2 for errata :)

[bmbouter]

+1 to JSON for simplicity. Are we using the custom json field or just storing it as a StringField and serializing/deserailizing when we read/write?

[dralley]

TextField for now... messing with JSONField directly is fraught with enough pitfalls generally speaking that until further notice I wouldn't want to rely on it. It works fine for reading (how we use it with Tasks) but writing is really hard to make work properly

[bmbouter]

Great. +1 to using TextField for now.

[dralley]

@goosemania Updated. I will squash the commits before we merge.

[dralley]

The "id" field on the UpdateRecord model conflicts with the "id" field on the Content model, we need to rename one of them

[bmbouter]

@dralley I think we need to change the "id" on the "Content model itself to pulp_id. This is the second content type I've heard that can't use the expected field name because Pulp is already asserting it.

[dralley]

@bmbouter Should we take it to #pulp-dev then?

[bmbouter]

@dralley yes: https://www.redhat.com/archives/pulp-dev/2018-May/msg00126.html

[dralley]

Have not tested this, it probably doesn't work. It's just a conceptual thing

[daviddavis]

s/Errata/Erratum/ ?

[dralley]

Nope, referring to the actual model name

https://github.com/pulp/pulp_rpm/blob/2-master/plugins/pulp_rpm/plugins/db/models.py#L1092

[daviddavis]

Ah, can you maybe clarify this comment then by saying something like 'Formerly "Errata" in Pulp 2'?

[daviddavis]

For some models in pulp_rpm, we're using Content in the name (RpmContent, SrpmContent) but these records lack the Content suffix. What about renaming RpmContent and SrpmContent to Rpm and Srpm to be consistent?

[dralley]

Is that a problem? In the Python plugin, the model is named PythonPackageContent but we don't use the content suffix either.

[daviddavis]

I think it's inconsistent. It should either be UpdateRecordContent, RpmContent, etc or UpdateRecord, Rpm, etc.

[dralley]

So, I think this is an interesting question. One point: ATM the Errata-related models don't subclass Content because I thought that it might be inappropriate, given that unlike packages, errata do not have have a 1-to-1 corresponding artifact. I think they're all stored together in updateinfo.xml, afaik.

@bmbouter Do you know if that is correct and/or have input? I should have brought this up as a discussion topic at some point.

[daviddavis]

Doesn't UpdateRecord subclass Content though?

[daviddavis]

Also, I'm not arguing for or against subclassing Content. Just pointing out that the class names are inconsistent.

[daviddavis]

So, I think this is an interesting question. One point: ATM the Errata-related models don't subclass Content because I thought that it might be inappropriate, given that unlike packages, errata do not have have a 1-to-1 corresponding artifact.

I think the way you have it is correct. The litmus test is probably whether a content unit can be added to a repository via RepositoryContent and have endpoints under places like /pulp/api/v3/content/.... Not whether the content unit has artifacts. So UpdateRecord should be the only erratum-related class to extend Content.

Another example of an artifact-less content unit is package group.