"Andrew Matthews" <andrew.m@corp.dslextreme.com> writes:
> [ PG 7.3.4 much slower than 7.2.1 ]
>
> Both have same databases, Both have had vacume full ran on them.
You did ANALYZE too, right?
The bulk of the time is evidently going into the seqscan on users in
each case:
> -> Seq Scan on users u (cost=0.00..1938.51 rows=71283 width=4) (actual time=0.81..30119.58
rows=70809loops=1)
> -> Seq Scan on users u (cost=0.00..1888.85 rows=71548 width=4) (actual time=18.38..2277152.51
rows=71028loops=1)
> Filter: (get_pwd(username, '127.0.0.1'::character varying, '101'::character varying,
'MD5'::charactervarying) IS NOT NULL)
I have to suspect that the inefficiency is inside this get_pwd()
function, but you didn't tell us anything about that...
regards, tom lane