On Tue, Nov 13, 2007 at 02:36:14PM -0500, Greg Smith wrote:
> Sure, but in this case the reasoning seems sound enough.
Yes. But. . .
> I see this as similar to the old optimizer hint argument, where there
> certainly exist some edge cases where people know something the optimizer
> doesn't which changes the optimal behavior.
. . .the abuse of such hints in applications I have seen is so rampant as to
make me doubt the utility of adding them anyway. It's true that by adding
hints, you give a facility to a good, competent designer who has a really
peculiar case that no general purpose system is likely to solve well. In
practice, however, it also seems to mean that every slack-jawed fool with
access to the manual thinks that he or she is going to "fix" the "broken"
query plan by forcing index scans where they're useless (has a week yet gone
by where someone doesn't post to -performance with that problem?). So I'm
divided on whether actually providing the facility is a good idea, even
though I can think of a handful of cases where I doubt even the smartest
planner will get it right. (By analogy, pinning in memory, and I'm
similarly divided.)
A
--
Andrew Sullivan
Old sigs will return after re-constitution of blue smoke