Ben <bench(at)silentmedia(dot)com> writes:
> Yes, it is horribly wrong.
> select count(*) FROM jennyann where target like '/music/%'
> gives me 93686 rows.
Well, that misestimation is the source of the problem, then, not any
misestimation of the cost of a sort.
7.0 didn't have very good estimation rules for LIKE clauses, at least
not by default. Have you tried the new LIKE estimator (see
contrib/likeplanning/README in the source tree)? I'm not sure it will
do any better, given that your data appears to be mightily nonuniform
;-), but it's worth a try.
regards, tom lane