Bruce Momjian <pgman@candle.pha.pa.us> writes:
> I am confused. I thought we already did optimization for LIMIT that
> assumed you only wanted a few values. Is there something we are missing
> there?
Yeah, he was proposing an alternative implementation of sorting that
would win in a scenario like
SELECT ... ORDER BY foo LIMIT <something small>
If you have an index on foo then there's no problem, but if you're
forced to do an explicit sort then the system does a complete sort
before you can get any data out. If the limit is small enough you
can instead do a one-pass "select top N" scan.
Note that this is only workable in the non-cursor case, where you
know the limit for sure.
regards, tom lane