Do we add a fallback behavior to ensure other objects from the failing transaction are written nonetheless?

Sounds sensible to do so (in the mirror context).

Maybe we should evolve the hash collision exception to refeerence the content in error.
That'd make it easier to fix client side the behavior if we had the culprit immediately in sight.

Maybe we should evolve the hash collision exception to refeerence the content in error.

Right now we have HashCollision 'type' of hash collision as message (e.g. sha1)... [1]
That's not helpful enough.

[1] https://sentry.softwareheritage.org/share/issue/5a6f8c09fea1468997bb88ca3e19fc2d/

are these vars necessary?

Why not simply use (with proper exception handling stuff):

storage.skipped_content_add(
              (obj for obj in objects if obj.get('status') == 'absent'))
storage.content_add_metadata(
              (obj for obj in objects if obj.get('status') != 'absent'))

Depends if we want to act upon my question below or not i guess ;)

Well one thing at a time. Let's deal with this as is, then discuss how to improve cases of ingestion error...