Re: Need theory/comprehension help on Multi-Column indexes

Tom, Merlin,

> It's not fundamentally different from single-column indexes. The only
> aspect of a btree index that requires any knowledge about the content of
> index entries is the "compare two index entries for lesser, equal, or
> greater" operation. For that, we just compare the first columns, then
> compare the second columns if the first are equal, etc. Plain
> lexicographic sort semantics.

So the different columns of the index don't have seperate data pages? It's
just a partitioned index node?

Wow, no wonder I couldn't figure it out, I was looking for something more
complicated ...

BTW, while we're on the optimizer, what is random_page_cost supposed to
represent, exactly? I used to think it was the ratio of index page
retreivals to direct page retrievals, but I see that that's already being
calculated for. I'm wondering if it might be possible to calculate RPC and
eliminate it as a GUC.

--
--Josh

Josh Berkus
Aglio Database Solutions
San Francisco