Dave Cramer <pg(at)fastcrypt(dot)com> writes:
> This query is a couple orders of magnitude slower the first result is
> 9.2.1, the second 9.1
Thanks for sending me the test case off-list. I found the reason why
I'd not been able to reproduce the problem: the index you're hoping it
will use is declared
"note_sets_parent_id_idx" btree (parent_id) WHERE parent_id IS NOT NULL
Apparently 9.2 is less bright than 9.1 about when it can use a partial
index. I'm not sure where I broke that, but will look.
regards, tom lane