Jon Lapham <lapham@jandr.org> writes:
> Tom Lane wrote:
>> I've thought about this before. One simple trick would be to get rid
>> of the current pg_class reltuples/relpages fields in favor of a
>> tuples-per-page estimate, which could be multiplied by
>> RelationGetNumberOfBlocks() during planning.
> My first reaction is to wonder if this would give performance exactly
> equal to running a true ANALYZE in every situation?
No, this is quite orthogonal to ANALYZE (and several orders of magnitude
cheaper...) It would fix a lot of the simpler cases, though, since the
first-order effects like table size would be right.
regards, tom lane