"Greg Sikorski" <gte@atomicrevs.demon.co.uk> writes:
> cmaster=# explain analyze SELECT user_id,channel_id FROM levels WHERE
> suspend_expires <= 1017589362 AND suspend_expires <> 0;
> NOTICE: QUERY PLAN:
> Seq Scan on levels (cost=0.00..13709.09 rows=2609 width=8) (actual
> time=208.98..1521.08 rows=17 loops=1)
> Total runtime: 1521.29 msec
First question is *can* the thing use an index? (Try "set enable_seqscan
to off" then explain again.) If not, it's probably a datatype
compatibility issue --- you'll need to quote or explicitly cast the
constant 1017589362 to match the type of suspend_expires.
If it could use an index but chooses not to, at least part of the
problem is the factor-of-100 overestimate in the number of matching
rows. That might be alleviated by increasing the statistics target
for the suspend_expires column. (ALTER TABLE SET STATISTICS, then
ANALYZE or VACUUM ANALYZE.)
You might also find that reducing random_page_cost produces better
indexscan cost estimates for your environment.
regards, tom lane