Re: INDEX suggestion needed

On Fri, 13 Dec 2002 16:41:38 +0100, Thomas Beutin
<tyrone@laokoon.IN-Berlin.DE> wrote:
>> >(visit >= '2002-10-01' AND visit <= '2002-10-31')    index scan, but long (>5sec)
>>
>> Ca. 3%; interesting that it takes so long; I'd like to see EXPLAIN

Oops!  Should be 10% according to the last two histogram bounds:
>> >           "2002-09-29 09:09:31+02"
>> >           "2002-10-29 23:25:13+01"

>> ANALYZE output for enable_seqscan on and off.
>There is no difference in cost.

Oops again!  If it's already using an index scan, switching seqscan
off won't change anything.  I should have meant "for enable_indexscan
on and off".


>> The negative correlation looks strange.  How did you insert your data?
>It is a dump from the production system, and the production system gets
>the data once a day from webserver logs line by line.

Did you insert in reverse order (newest first)?

| most_common_freqs = {0.000666667,...,0.000666667}

These values occurred exactly twice in the analyzer's sample of 3000
values, so for the accuracy we need here it's ok to ignore them.

Servus
 Manfred