MicrosoftLiwei I came here to find the answer for 2.If there are multi-column stats, does it make different?
But it looks like the answer is cut off!
I'm trying to give the engine enough data to make the right decisions in a scenario where a excessive memory grant is causing performance issues. I add multi column user statistics (sql 2019 w/new CE) but in my tests it appears the engine ignores them.
It seems counter-intuitive that user created multi column statistics that are an exact match to the supplied predicate are ignored in favor of an algorithmic guess. Is there a reason that we would prefer the algorithmic approach over the sampled data?
