Comment 9 for bug 1837995

I've actually just been able to replicate this in an environment. The was due to the fact that instance_extras actually had 2 records for the same instance_uuid, one which was fully empty and the second which was populated.

When checking the database log, it seems that the system was actually failing to create instance_info_caches because of a conflict in the database, complaining that a record with the same instance_uuid already existed (...somehow?)

I assume that on the first run, the code creates instance_extra that's empty (but then fails to create instance_info_cache for ????) and on the second run it actually creates instance_extra once again (and the latter row starts getting populated).

We would fail harder on this if we had a constraint on a unique instance_uuid in instance_extra which we don't seem to have.