Re: generalizing the planner knobs

Neil Conway <neilc(at)samurai(dot)com> writes:
> On Thu, 2005-12-01 at 21:01 -0500, Gregory Maxwell wrote:
>> If we'd really like to avoid people using the knobs to rig queries,
>> how about making them only work with explain analyze, useful for
>> debugging but not so useful for actual queries.

> That seems a pretty arbitrary limitation. I agree that it's not ideal to
> have users adjust planner behavior via this means, but until we have
> something better, I think applying that limitation would only make the
> status quo worse.

Yeah, I agree. Adding code to prevent people from using a facility
doesn't seem very reasonable, even if it's our policy that using the
facility for production purposes is not a good idea. In fact, we just
today had a counterexample --- see this thread:
http://archives.postgresql.org/pgsql-performance/2005-12/msg00015.php
Being able to use enable_nestloop got Markus out of a short-term bind,
which to me is exactly what you want to be able to do with this sort
of thing.

I don't have any problem with expending small amounts of work to make
it easier to hack the planner in small ways. The real problem I have
with a "planner hints" facility (in the form that I think most people
who ask for it have in mind) is that it would be a *very large* amount
of work to do it reasonably well, and I think that amount of effort
would be better spent in other ways.

regards, tom lane