Heikki Linnakangas <heikki.linnakangas@enterprisedb.com> writes:
> On second thought, if XID wraparound is close enough that the DBA has to
> log in to do manual vacuums to avoid it, relfrozenxid of the
> trouble-making tables are surely older than default
> vacuum_freeze_table_age, so plain VACUUM is enough to scan the whole table.
OK, good. I think we're creating real hazards if anything but the
plainest form of VACUUM is required in this scenario. It seems like
we're safe at the moment, but maybe these considerations should be
documented in the code somewhere, so we don't break the case
accidentally in future.
regards, tom lane