On Thu, Oct 12, 2023 at 2:58 PM Nikita Malakhov <hukutoc@gmail.com> wrote:
Why pg_upgrade cannot be used?
We document both a pg_dump/pg_restore migration and a pg_upgrade one (not to mention that logical backup and restore would cause the oids to change). It seems odd to have a feature that requires pg_upgrade to be the chosen one. pg_upgrade is an option, not a requirement. Same goes for pg_basebackup.
pg_upgrade itself warns that should the on-disk file format change then it would be unusable - though I suspect that we'd end up with some kind of hybrid approach in that case.
OID preservation logic is already implemented
for several OIDs in catalog tables, like pg_class, type, relfilenode, enum...
We are allowed to preserve oids if we wish but that doesn't mean we must, nor does doing so constitute a declaration that such oids are part of the public API. And I don't see us making OIDs part of the public API unless we modify pg_dump to include them in its output.
Actually, I've asked here because there are several references to PG_PROC oids
from other tables in the system catalog
Of course there are, e.g., views depending on functions would result is those. But pg_upgrade et al. recomputes the views so the changing of oids isn't a problem.
Long text fields are common in databases; and if there are concerns with parsing/interpretation we can add functions to make doing that simpler.