Re: GiST indexing problems...

So, I've migrated my code to do the TOAST'ing thing required of 7.1
clients, and I've updated my operator to use the areaselectors:
CREATE OPERATOR = (
leftarg = graph,
rightarg = graphrange,
procedure = graph_inrange,
commutator = '=',
restrict = areasel,
join = areajoinsel
);

But I still get the issue that my queries don't seem to trigger the
GiST indexes to be used. Perhaps the problem is that the system
just thinks that the query doesn't need an index to increase
performance, i've only got about a dozen elements in the database
right now for testing purposes.

-David

Tom Lane writes:
> David McWherter <udmcwher(at)mcs(dot)drexel(dot)edu> writes:
> > I've tried adding the parameters 'restrict = eqsel' and 'join = eqjoinsel'
> > to the datatype operators, but that doesn't seem to change anything.
>
> You might have better luck if you use area-related selectivity
> estimators. Your problem seems to be that the optimizer doesn't
> think the index is worth using, and the cause almost certainly is
> overly pessimistic selectivity estimates for the indexable operators.
> areasel and friends are completely bogus, but at least they deliver
> small enough numbers to encourage use of the index ;-)
>
> As Oleg says, the GiST support in 7.0.* is in pretty poor shape
> (it had been suffering from neglect for a long time). Try 7.1.
>
> regards, tom lane

----------------------[=========]------------------------
David T. McWherter udmcwher(at)mcs(dot)drexel(dot)edu

The truth is rarely pure, and never simple.
-- Oscar Wilde