I'm pretty sure it didn't analyze in between - autovac is turned off
and I ran the test multiple times before posting.
But since I can't reproduce it anymore, I can't be 100% sure. And it
certainly doesn't make sense that the estimate for the index scan would
change based on an unrelated join condition.
If I ever get it to happen again, I'll be more careful and repost if it
is a real issue. Thanks for pointing me in the right direction!
On Wed, 10 Jan 2007 13:38:15 -0500, "Tom Lane" <tgl@sss.pgh.pa.us> said:
> "Jeremy Haile" <jhaile@fastmail.fm> writes:
> > I still don't understand why the inner join would be so different from
> > the left join prior to the analyze.
>
> Are you sure you hadn't analyzed in between? Or maybe autovac did it
> for you? The reason for the plan change is the change from estimating
> 1 row matching the transaction_date range constraint, to estimating lots
> of them, and the join type away up at the top would surely not have
> affected that.
>
> regards, tom lane