> "TJ O'Donnell" <tjo(at)acm(dot)org> writes:
>
> > -> Seq Scan on structure (cost=0.00..191619.56 rows=1765669 width=32) (actual
> > time=8050.437..42117.062 rows=1569 loops=1)
^^^^^^^^
And do you vacuum regularly? Have you done batch updates or deletes and then
never inserted enough records to reuse that dead space? You might have to
vacuum full (or alternatively use the cluster command) to reclaim this dead
space.
I don't recall what the original motivation to rewrite the analyze sampling
was, did having lots of dead tuples cause bad estimates in 7.4?
--
greg