On Tue, Jun 27, 2006 at 08:02:33PM -0700, Jeff Frost wrote: > On Tue, 27 Jun 2006, Jim Nasby wrote: > > >>What the heck? Why would it have shrunk down so much? The vacuum full > >>isn't scheduled until tonight. :-/ > > > >Take a look at the source for autovacuum... I'm guessing you're right and > >that it doesn't make any special considerations for toast tables, which it > >probably should (since AFAIK you have to first vacuum the base table, then > >commit, then vacuum the toast table). > > > >BTW, tracking pg_class.relpages or actual file size over time would > >probably be helpful. > > Actually, it says this: > > /* > * Scan pg_class and determine which tables to vacuum. > * > * The stats subsystem collects stats for toast tables independently of > * the stats for their parent tables. We need to check those stats since > * in cases with short, wide tables there might be proportionally much > * more activity in the toast table than in its parent. > * > * Since we can only issue VACUUM against the parent table, we need to > * transpose a decision to vacuum a toast table into a decision to vacuum > * its parent. There's no point in considering ANALYZE on a toast table, > * either. To support this, we keep a list of OIDs of toast tables that > * need vacuuming alongside the list of regular tables. Regular tables > * will be entered into the table list even if they appear not to need > * vacuuming; we go back and re-mark them after finding all the vacuumable > * toast tables. > */ > > So I guess it does take toast tables into account. Only if it issues 2 vacuums on the base tables... -- Jim C. Nasby, Sr. Engineering Consultant jnasby@xxxxxxxxxxxxx Pervasive Software http://pervasive.com work: 512-231-6117 vcard: http://jim.nasby.net/pervasive.vcf cell: 512-569-9461
