Mike Wilson <mfwilson@gmail.com> writes:
> Just for my interest I also created a new PG 842 db (initdb) and put
> some sample data it it and successfully did a pg_upgrade from 842 ->
> 920b2. Whatever the issue is it seems to be related to my actual
> database as a sample db conversion works.
Yeah, I was suspecting that, because there's no obvious problem in the
part of pg_upgrade that your stack trace fingers.
What we need to do next is get something that one of the PG developers
can put under a microscope. What I would suggest first is doing a
"pg_dumpall -s" (ie no data) of your database, and seeing whether
loading that into 8.4 produces a database on which pg_upgrade fails.
If so, then you could send us that dump (perhaps after some sanitizing
of names) and we could have at it.
If you aren't able to create a self-contained reproducible case in this
way, the only other way forward is for you to debug it yourself or let
one of the PG developers have access to your system to poke at it.
regards, tom lane