tvahakan@fw.cluster.net writes:
> Using libpq++ simple UPDATE has gone from under
> one millisecond to over 20 milliseconds after a few days of
> uptime (mostly idle)? VACUUM ANALYZE didn't help, but
> restarting postmaster did.
This seems *very* odd; restarting the postmaster should not make a
difference. What exactly is the query you're using, what exactly
is the table and index schema, and what does EXPLAIN say is the
query plan (in both fast and slow states)?
regards, tom lane