Joseph Shraibman <jks(at)selectacast(dot)net> writes:
> OK this may seem like a stupid question, but isn't index scan always
> better except for the pathalogical simple case where the work to be done
> is trivial anyway?
No. If it were, the optimizer would be a whole lot simpler ;-)
In practice an indexscan only wins if it will visit a relatively
small percentage of the tuples in the table. The $64 questions
are how small is small enough, and how can the optimizer guess
how many tuples will be hit in advance of doing the query...
regards, tom lane