Re: OID/XID allocation (was Re: is PG able to handle a >500 GB Database?)

> Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
> > What about pre-fetching of OID's. Does that still happen for every
> > backend?
>
> Only ones that actually allocate some OIDs, I think.
>
> > What about XID's?
>
> XIDs are wasted on a postmaster restart, but not per-backend, because
> they are cached in shared memory instead of locally. I've been thinking
> about changing the code so that OIDs are allocated in the same fashion.
> That would mean an extra spinlock grab per OID allocation, but so what?
> We grab several spinlocks per row creation already. And we could
> increase the number of OIDs allocated per pg_variable file update,
> which would save some time.
>
> Haven't got round to it yet though, and I'm not sure but what Vadim
> might be planning to throw out all that code anyway ...

Added to TODO:

* Move OID retrieval into shared memory to prevent lose of unused oids

Also, currently the oid can _not_ be used to determine the order rows
were inserted because one backend can grab its block of 50 and another
backend can start and insert a row first.

If we could change this with litle risk, it would be nice to have in
7.1, but I am sure someone will object. :-)

--
Bruce Momjian | http://candle.pha.pa.us
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 853-3000
+ If your life is a hard drive, | 830 Blythe Avenue
+ Christ can be your backup. | Drexel Hill, Pennsylvania 19026