On Sun, 2005-04-17 at 09:17, Todd Underwood wrote:
> howdy,
>
> i'm by no means a postgres expert, so i'm probably making a dumb
> mistake, but this is v. troubling.
>
> we have a script that bulk-uploads about 45K rows. we do this via
> perl DBI using the COPY functionality.
>
> the problem is that we do this about once per 5 minutes and the time
> it takes to do the operation steadily increases (going from about 5s
> to over 5m over the course of 24 hrs). the table has one primary key
> (so hence one index), that we drop and recreate during the bulk load.
I'm guessing that either you aren't vacuuming enough, or your fsm
settings are too low.
What does vacuum verbose say? What are your fsm settings? Does 'vacuum
full <tablename>' fix the problem?
Do NOT TURN OFF FSYNC.