Re: How to get around LIKE inefficiencies?

Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Sorry to be getting in here late.  Have you tried CLUSTER?

Prolly won't help much.  I think what he's getting burnt by
is that the planner thinks that an indexscan based on the
LIKE 'http://www.postgresql.org/%' condition will be extremely
selective --- it has no idea that most of the URLs in his table
will match that prefix.  It's ye same olde nonuniform-distribution
problem; until we have better statistics, there's not much hope
for a non-kluge solution.
        regards, tom lane