Reinhard Max <max(at)suse(dot)de> writes:
> I've just found a case where forcing indexscans results in much higher
> speed.
> -> Index Scan using foo_pkey on foo
> (cost=0.00..25153.18 rows=352072 width=4)
> (actual time=0.03..157.57 rows=38432 loops=1)
The major estimation error is evidently in this indexscan. What
statistics does pg_stats show for this table?
regards, tom lane