On Mon, 2008-08-18 at 23:43 -0400, Tom Lane wrote:
> Simon Riggs <simon@2ndQuadrant.com> writes:
> > Probably also using a separate Sequence to allocate numbers rather than
> > using up all the Oids on LOs would be a good plan.
>
> Well, assuming that your Large Objects are actually Large, you aren't
> going to need as many OIDs as all that ;-)
>
> However: I was chewing on this point a few days ago, and it seemed to me
> that essentially duplicating the functionality of the OID generator
> wasn't likely to be a win. What seems more practical is to extend the
> internal next-OID counter to 64 bits, and allow callers to get either
> the full 64 bits or just the lowest 32 bits depending on what they need.
> This change would actually be entirely transparent to 32-bit callers,
> and the extra cycles to manage a 64-bit counter would surely be lost in
> the noise compared to acquiring/releasing OidGenLock.
Sounds very cool.
-- Simon Riggs www.2ndQuadrant.comPostgreSQL Training, Services and Support