Re: index / sequential scan problem

At 11:17 17.07.2003, Shridhar Daithankar wrote:
>On 17 Jul 2003 at 11:01, Fabian Kreitner wrote:
> > psql (PostgreSQL) 7.2.2
> >
> > perg_1097=# VACUUM ANALYZE ;
> > VACUUM
> > perg_1097=# EXPLAIN ANALYZE    select  notiz_id, obj_id, obj_typ
> > perg_1097-#     from    notiz_objekt a
> > perg_1097-#     where not exists
> > perg_1097-#     (
> > perg_1097(#       select  1
> > perg_1097(#       from    notiz_gelesen b
> > perg_1097(#       where   ma_id  = 2001
> > perg_1097(#       and     ma_pid = 1097
> > perg_1097(#       and     a.notiz_id = b.notiz_id
> > perg_1097(#     )
> > perg_1097-# ;
>
>For 31K records, seq. scan does not sound like a bad plan to me but anyway..

Im not generally worried that it uses a seq scan but that the second
example (where an index on the sub select is used on a table with only 45
entries) executes more than 4 times faster. Its not a cache thing either,
since i can enable seqscan again and it will run with 2300ms again.

>How about
>
>  where   ma_id  = 2001::integer
>and     ma_pid = 1097::integer
>
>in above query?

I dont really understand in what way this will help the planner but ill try.

Thanks,
   Fabian