Re: disfavoring unparameterized nested loops

On Fri, Sep 30, 2022 at 3:19 PM Benjamin Coutu <ben(dot)coutu(at)zeyos(dot)com> wrote:
> > For all I know you might be onto something. But it really seems
> > independent to me.
>
> Yeah, I‘m sorry if I highjacked this thread for something related but technically different.

I certainly wouldn't say that you hijacked the thread. I'm glad that
you revived the discussion, in fact.

The way that I've framed the problem is probably at least somewhat
controversial. In fact I'm almost certain that at least one or two
people will flat out disagree with me. But even if everybody else
already thought about unparameterized nested loop joins in the same
terms, it might still be useful to make the arguments that you've
made.

What I'm saying is that the probability of "getting it right" is
virtually irrelevant in the case of these unparameterized nested loop
join plans specifically. Any probability that's less than 1.0 is
already unacceptable, more or less. A probability of 1.0 is never
unattainable in the real world, no matter what, so why should the true
probability (whatever that means) matter at all? The appropriate
course of action will still be "just don't do that, ever".

To me this dynamic seems qualitatively different to other cases, where
we might want to give some weight to uncertainty. Understanding where
the boundaries lie between those trickier cases and this simpler case
seems important and relevant to me.

--
Peter Geoghegan