Re: after vacuum analyze, explain still wrong

Actually most of the estimates seem pretty good. There are some that
are a ways off, but the real nasties seem to be these. I'm a bit
confused because it looks like two of your joins don't have Join
Filters -- and one of those is a left join for which I thought that
was impossible.

Are you sure this query is doing what you expect? You have
add_missing_from enabled which will happily bring in additional joins
if you reference a table which isn't already included in the join and
do a full cartesian-product join.

-> Nested Loop
Left Join (cost=145.04..2381.33 rows=1 width=74) (actual
time=38.955..9678.057 rows=3754 loops=1)
Join
Filter: ((c.partid = e.partid) AND leadcompcheck_ab(a.leadfree,
c.leadstateid))
-> Nested
Loop (cost=145.04..2366.26 rows=1 width=62) (actual time=0.791..83.482
rows=3754 loops=1)
->
Nested Loop Left Join (cost=145.04..2365.59 rows=1 width=28) (actual
time=0.721..52.840 rows=3754 loops=1)

Filter: (f.commited IS NOT TRUE)

--
greg
http://mit.edu/~gsstark/resume.pdf