The incremental JSON dumps operate using a mirror database, but also require foreign keys in the database (which are normally not created on mirrors) in order for the FollowForeignKeys role to work. This is hardly documented, and also needs to be maintained by hand after each schema change. Furthermore, we cannot guarantee there to be no replication issues with such a setup.

It would be better if we could simply dump foreign key information from another (standalone) database, and read it from a file.