Mark Dexter wrote:
> We use a development environment that works with Postgres via ODBC and
> uses cursors to insert and update rows in Postgres tables. I'm using
> Postgres version 7.4.5.
> A. If I TRUNCATE or DELETE all of the rows in the table and then run
> VACUUM or ANALYZE on the empty table, the test program takes over 15
> minutes to complete (i.e., 15X performance drop).
> If we routinely run VACUUM or VACUUM ANALYZE (e.g., nightly), these work
> tables will normally be empty when the VACUUM is run. So it would
> appear from the testing above that they will experience performance
> problems when inserting large numbers of rows through our application.
Yep - it's a known issue. The analyse is doing what you asked, it's just
not what you want.
> Is there some easy way around this problem? If there a way to force
> VACUUM or ANALYZE to optimize for a set number of rows even if the table
> is empty when it is run? Thanks for your help. Mark
There are only two options I know of:
1. Vaccum analyse each table separately (tedious, I know)
2. Try pg_autovacuum in the contrib/ directory
The autovacuum utility monitors activity for you and targets tables when
they've seen a certain amount of activity. Even if it hasn't got the
tunability you need, it should be a simple patch to add a list of
"excluded" tables.
--
Richard Huxton
Archonet Ltd