Re: order by slowing down a query by 80 times

Rajesh Kumar Mallah <mallah(dot)rajesh(at)gmail(dot)com> writes:
> co_name_vec is actually the auxiliary tsvector column that is mantained via
> a
> an update trigger. and the index that you suggested is there .

Well, in that case it's just a costing/statistics issue. The planner is
probably estimating there are more tsvector matches than there really
are, which causes it to think the in-order indexscan will terminate
earlier than it really will, so it goes for that instead of a full scan
and sort. If this is 8.4 then increasing the statistics target for the
co_name_vec column should help that. In previous versions I'm not sure
how much you can do about it other than raise random_page_cost, which is
likely to be a net loss overall.

regards, tom lane