Shiar <shiar@shiar.org> writes:
>> EXPLAIN ANALYZE SELECT id FROM userinfo WHERE icq <> '0';
> Seq Scan on userinfo (cost=0.00..47355.90 rows=849244 width=4) (actual time=0.563..1222.963 rows=48797 loops=1)
> Filter: (icq <> 0::bigint)
> Total runtime: 1258.703 ms
The rows estimate is way off, which might or might not have much to do
with the issue, but it's surely suspicious.
> We're running PostgreSQL 7.4.1 on a Debian/Linux 2.4 system with 4GB RAM and a
> fast SCSI RAID array, with settings:
Update to 7.4.2 and follow the procedure in the release notes about
fixing pg_statistic; that may make things better. int8 columns are
vulnerable to the statistic misalignment bug.
regards, tom lane