> Sure, but the model isn't the problem here, really -- not to me. The
> problem is that the planner can in some cases choose a plan that is
> inherently unreasonable, at least in practical terms. You're talking
> about uncertainties. But I'm actually talking about the opposite thing
> -- certainty (albeit a limited kind of certainty that applies only to
> one narrow set of conditions).
I absolutely agree and support your proposal to simply not generate those paths at all unless necessary.
> 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 just wanted to expand on
yourproposal by taking into account the join depth and also not just talking about unparametrized nested loop joins.
Theresearch is very clear that the uncertainty is proportional to the join level, and that is the point I am trying to
focusthe discussion on.
I really encourage everyone to read the VLDB paper. BTW, the unnamed proprietary DBMSs in that paper are the 3 big ones
fromWashington, California and NY, in that order.