Re: Slow performance with Group By

Erik Norvelle <signups(at)norvelle(dot)org> writes:
>>> it=> explain select codelemm, sectref, count(codelemm) from indethom
> group by codelemm, sectref;
>>> QUERY PLAN
>>> -----------------------------------------------------------------------
> ---------
>>> GroupAggregate (cost=2339900.60..2444149.44 rows=1790528 width=13)
>>> -> Sort (cost=2339900.60..2364843.73 rows=9977252 width=13)
>>> Sort Key: codelemm, sectref
>>> -> Seq Scan on indethom (cost=0.00..455264.52 rows=9977252
> width=13)

Actually the painful part of that is the sort. If you bump up sort_mem
enough it will eventually switch over to a HashAggregate with no sort,
which may be a better plan if there's not too many groups (is the
estimate of 1.79 million on the mark at all??)

regards, tom lane