>> Playing around with seq_page_cost (1) and random_page_cost (1), I can get >> the correct index selected. Applying those same settings to our production >> server does not produce the optimal plan, though. > > I doubt setting seq_page_cost and random_page_cost to the same value is > reasonable - random access is almost always more expensive than sequential > access.
If the data figures to be read from the OS cache, it's very reasonable, and the right value is somewhere in the 0.05 - 0.10 range.
For the most part, it will indeed be cached. Thanks for the tip on the values.