On Wed, 30 Sep 2020 at 04:42, Tom Lane wrote:
>
> David Rowley writes:
> > On Tue, 29 Sep 2020 at 12:08, Tom Lane wrote:
> >> The idea I'd had was to adjust make_subplan and cost_subplan to estimate
> >> EXIST cases on the basis of either 50% retrieval (same as ANY/ALL) or
> >> maybe full retrie
David Rowley writes:
> On Tue, 29 Sep 2020 at 12:08, Tom Lane wrote:
>> The idea I'd had was to adjust make_subplan and cost_subplan to estimate
>> EXIST cases on the basis of either 50% retrieval (same as ANY/ALL) or
>> maybe full retrieval if you want to be pessimistic. I've not had time
>> to
On Tue, 29 Sep 2020 at 12:08, Tom Lane wrote:
>
> David Rowley writes:
> > Any opinions on this?
>
> This patch scares the heck out of me. It's a pretty much unprincipled
> change in a fundamental selectivity estimator, which is going to affect
> all sorts of queries not only the particular case
Thanks for chipping in here.
On Tue, 29 Sep 2020 at 12:08, Tom Lane wrote:
>
> David Rowley writes:
> > Any opinions on this?
>
> This patch scares the heck out of me. It's a pretty much unprincipled
> change in a fundamental selectivity estimator, which is going to affect
> all sorts of querie
David Rowley writes:
> Any opinions on this?
This patch scares the heck out of me. It's a pretty much unprincipled
change in a fundamental selectivity estimator, which is going to affect
all sorts of queries not only the particular case you have in mind.
There's no reason to think that the outco
On Mon, 28 Sep 2020 at 13:22, David Rowley wrote:
> I'm more leaning towards the fact that the planner picked the seqscan
> path in the first place. Unfotunately I don't have any great ideas on
> how to fix without fudging the costs a bit. Maybe we can make
> var_eq_non_const() divide down the sel
This follows on from what I mentioned in [1]. I'll reiterate below:
Really the topic is much more broad than what I wrote above, but the
example case I have is around subplan costing which results in the
alternative subplan code making a bad choice.
So the example case is:
create table t (a int,
