Tom, just did a pg_dump -s from 9.3 and there are no warnings in the output file. It goes smoothly and fast (few seconds). Database is around 50GB, schema 9MB. Has postgis and a few more extensions – ogr_fdw (from postigs), file_fdw, postgres_fdw, table_func.
I also did a pg_dumpall -s and also no warnings.
I do a pg_dump full backup every week with no incidents.
Versions are:
Windows x64
9.3.22
9.5.15
Thanks for taking a look.
Duarte
De: Duarte Carreira
Enviada: 24 de janeiro de 2019 21:52
Para: Tom Lane <tgl@sss.pgh.pa.us>
Cc: pgsql-general@postgresql.org
Assunto: Re: duplicate OID issue when using pg_upgrade to move from 9.3 to 9.5
I'll try pgdump tomorrow.
Duarte Carreira <DCarreira@edia.pt> writes:
> I've trying to upgrade a 9.3 instance to 9.5 using pg_upgrade and facing this issue...
9.5.what?
Perusing the commit logs, I note that 9.5.3 included a fix for a
pg_upgrade issue that could possibly lead to this symptom, see
https://git.postgresql.org/gitweb/?p=postgresql.git&a=commitdiff&h=196870f2f
If you are on 9.5.recent, it'd be worth looking closer, because this
is certainly pretty odd. One wouldn't expect a CREATE FUNCTION to
result in assignment of a type OID, at least not in pg_dump/pg_upgrade
scripts --- they should always put out a shell CREATE TYPE first.
By any chance, if you attempt a "pg_dump -s" from the problematic database,
does it emit any warnings (about dependency loops, perhaps)?
regards, tom lane