Steven Rosenstein <srosenst(at)us(dot)ibm(dot)com> writes:
> I used plain old VACUUM. Do you think VACUUM FULL might be faster or more
> effective?
No. I think you probably want to do a dump and reload, but first you
have to get past the anti-wraparound check.
One possibility I hadn't thought of before is to use a standalone
backend to increment the pg_database.datfrozenxid values by a few
thousand transactions. This would be a bad idea if you intended
to keep using the DB, but if you're just trying to get to a state
where you can run pg_dump, it seems acceptable.
regards, tom lane