Fixes a Pulp2Content unique-constraint that wasn't very unique.

[ggainey]

In Postgres, all NULLs are unique - so if a nullable field is included
in a UiniqueConstraint, it allows rows that differ only by the NULL.

Build two constraints, one when the 'optional' field is NULL, and
one where it has content.

fixes #8329
[nocoverage]

[ggainey]

This is in draft for two reasons:

First - Making the UQ 'correct' means we probably do NOT need to make sure errata store repo_id in subid. After discussion around this approach, we may remove the Pulp2Content(type=erratum).update() from the migration, and the code that copies repo-id into subid.

Second, and more painful - with the corrected UQ in place, the reproducer from https://pulp.plan.io/issues/8329#note-3 fails with the stacktrace noted in comment-10. Discussion needed on how we address the problem.

[ggainey]

Second, and more painful - with the corrected UQ in place, the reproducer from https://pulp.plan.io/issues/8329#note-3 fails with the stacktrace noted in comment-10. Discussion needed on how we address the problem.

bulk_create(pulp2content, ignore_conflicts=True) causes more trouble than it's worth :(

[ggainey]

This is in draft for two reasons:

First - Making the UQ 'correct' means we probably do NOT need to make sure errata store repo_id in subid. After discussion around this approach, we may remove the Pulp2Content(type=erratum).update() from the migration, and the code that copies repo-id into subid.

Second, and more painful - with the corrected UQ in place, the reproducer from https://pulp.plan.io/issues/8329#note-3 fails with the stacktrace noted in comment-10. Discussion needed on how we address the problem.

Both of these issues should now be resolved.

[goosemania]

Thank you!