Robert Haas <robertmhaas@gmail.com> writes:
> Of course, the other obvious question is whether we really need a
> consistent snapshot, because that's bound to be pretty expensive even
> if you eliminate the I/O cost. Taking a consistent snapshot across
> all 100,000 tables in the database even if we're only ever going to
> access 5 of those tables doesn't seem like a good or scalable design.
Mumble. It's a property I'm pretty hesitant to give up, especially
since the stats views have worked like that since day one. It's
inevitable that weakening that guarantee would break peoples' queries,
probably subtly.
What we need is a way to have a consistent snapshot without implementing
it by copying 100,000 tables' worth of data for every query. Hmm, I heard
of a technique called MVCC once ...
regards, tom lane