On Thu, Aug 02, 2007 at 10:40:24PM -0400, Joseph S wrote: > Tom Lane wrote: > >Heavy use of temp tables would expand pg_class, pg_type, and especially > >pg_attribute, but as long as you have a decent vacuuming regimen (do you > >use autovac?) they shouldn't get out of hand. > > I do use autovac. Like I said they don't get really out of hand, only > up to 20 megs or so before I noticed that it was weird. The large > indexes are what tipped me off that something strange was going on.
Unexpected bloat in pg_shdepend led me to discover a problem with statistics for shared tables a couple of months ago: http://archives.postgresql.org/pgsql-hackers/2007-06/msg00190.php http://archives.postgresql.org/pgsql-hackers/2007-06/msg00245.php > I only noticed this because I was making an effort to monitor index > bloat on my regular tables. It could be there are a lot of people out > there who are experiencing this but don't notice because 20 megs here > and there don't cause any noticeable problems. Anybody making heavy use of temporary tables and relying on autovacuum is probably suffering bloat in pg_shdepend because no released version of PostgreSQL has the fix for the statistics bug (it has been fixed in CVS, however). As I mention in the second message above, vacuuming pg_shdepend resulted in an immediate performance improvement in an application I was investigating. -- Michael Fuhr ---------------------------(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
