Thanks, Adrian, but i was looking for something that goes into more depth. For example there is one case described there where we have a Nested Loop with rows=33, and its 2 child nodes have each rows=10. But first of all this is a very exotic join condition (t1.hundred < t2.hundred) and second of all i cannot find any explanation how this number 33 is derived. They literally spend less than 2 sentences on this case.
In my case i as well have Nested Loop's rows value different than the product of its 2 children's rows values, but with a normal join condition on a foreign key. My guesses (again) are the same way it keeps some statistics (very curious what exactly) about what is the probability 2 random rows from each table satisfy that condition (t1.hundred < t2.hundred), it also keeps statistics what is the probability 2 random rows from each table satisfy the regular normal join condition like the one i have in my case (basically t1.t2_id=t2.id). And in both cases it just applies that probability to the product of the rows values of the 2 child nodes, to calculate the expected rows value of the result of the Nested Loop. But i am just guessing. >-------- Оригинално писмо -------- >От: Adrian Klaver adrian.kla...@aklaver.com >Относно: Re: Question about Expected rows value in EXPLAIN output for Nested Loop node >До: bb ddd <nnickoloff1...@abv.bg> >Изпратено на: 10.05.2019 19:02 > On 5/10/19 8:56 AM, bb ddd wrote: > > Thanks for the reply. Here it is: https://explain.depesz.com/s/LQCS > > > > My main trouble is not with this exact case, but i am looking for a general > > description of the algorithm how those numbers are calculated. In the > > meanwhile i also constructed a couple of artificial tables to experiment > > with and see how these numbers change, and am pretty sure my guesses there > > in the original question are quite close to what is going on, but would be > > so much easier to read explain plans, if there was some documentation about > > how the expected rows are calculated for different nodes, in different > > contexts (like when loops=1 vs loops>1) maybe what statistics they use, how > > they depend on the numbers of their child nodes etc. > > > > > > How about?: > https://www.postgresql.org/docs/11/using-explain.html > > > -- > Adrian Klaver > adrian.kla...@aklaver.com
