> I beg your pardon?
Get it - I didn't know about enable_seqscan -:)
> Unfortunately neither of these plans is likely to be especially speedy
> on ~3 million rows. The index scan will just thrash the disk, unless
> the table has been clustered recently --- and given the
> deficiencies of our CLUSTER implementation, I'd hesitate to recommend
> using it.
Easy to try - why don't do.
> I have a personal TODO item to see about implementing group +
> aggregate with a hash table of active aggregate values, per a
> suggestion recently from devik@cdi.cz. That would allow this
> query to be done with a sequential scan and no sort, which is
> probably what Oracle is doing. Won't happen for 7.1 though ...
Well, definitely good approach. But for the moment increasing
sort memory is only hope and it should help.
Vadim