On Wed, 2007-04-04 at 16:26 -0400, Tom Lane wrote: > Bruce Momjian <[EMAIL PROTECTED]> writes: > > Simon Riggs wrote: > >> Having both default GUC and individual table-level WITH parameters seems > >> like the best way to me. > > > Agreed. > > There's an extremely good reason not to have a GUC variable, which is > that changes in it would fail to reflect into decisions about whether to > create TOAST tables. When I first brought up the point I didn't see a > way around it, but now that I do, I don't think we should expose a > failure mode just to have a GUC.
It depends how it works. If the GUC was a default that was applied only at CREATE TABLE time, then it would be safe. Changing default_with_oids didn't cause all tables to stop/start using oids. Why would it? > > OK, but we need to throw a clear message when the TOAST table needs to > > be created by the administrator. > > No, we just need to not have a GUC for this. There's no GUC for default > fill factor; have you seen anyone complain about that? I'd rather set it once than many times, thats all. I certainly care more about temp_tablespaces than I do about this GUC... that is something I'll be moaning about if that gets deferred. > > The big question is whether this is for 8.3 or 8.4. > > What I would definitely like to see for 8.3 is some performance testing > done to determine whether we ought to change the current defaults. > (Both TOAST_TUPLES_PER_PAGE and EXTERN_TUPLES_PER_PAGE ought to be looked > at.) > > Whether it's possible to get the storage parameter in there depends on > how soon someone produces a patch. Given that we understand this area > fairly well, I personally would be willing to give it a pass on the > "feature freeze" rule, as long as we have the patch by say mid-April. I meant to say a clear "yes" to that, but I've other business stuff for two weeks in mid-April so I'll need to rely on colleagues to take up the challenge. -- Simon Riggs EnterpriseDB http://www.enterprisedb.com ---------------------------(end of broadcast)--------------------------- TIP 9: In versions below 8.0, the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match