Bump, and curious if anyone on hackers has any ideas here: of particular
interest is why the (pk, created_at) index can possibly be more valuable
than the (created_at, pk) variant since the former effectively implies
having to scan the entire index.
On Fri, Sep 7, 2018 at 12:17 PM James Coleman
I have the following tables:
- m(pk bigserial primary key, status text): with a single row
- s(pk bigserial primary key, status text, action_at date, m_fk bigint):
* 80% of the data has action_at between the current date and 1 year ago
and status of E or C
* 20% of the data has action_at b
