On Wed, Apr 8, 2020 at 11:17 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Amit Langote <amitlangote09@gmail.com> writes:
> > On Tue, Apr 7, 2020 at 2:41 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >> I *think* that it might be all right, because although what we
> >> are matching to is a user-written COALESCE() not an actual
> >> FULL JOIN USING column, it has to behave in somewhat the same
> >> way. In particular, by construction it must be a coalesce of
> >> some representation of the matching partition columns of the
> >> full join's inputs. So, even though it might go to null in
> >> different cases than an actual USING variable would do, it
> >> does not break the ability to partition the join.
>
> > Seems fine to me too. Maybe users should avoid writing it by hand if
> > possible anyway, because even slight variation in the way it's written
> > will affect this:
>
> I'm not particularly concerned about users intentionally trying to trigger
> this behavior. I just want to be sure that if someone accidentally does
> so, we don't produce a wrong plan.
>
> I waited till after the "advanced partitionwise join" patch went
> in because that seemed more important (plus I wondered a bit if
> that would subsume this). But this patch seems to still work,
> and the other thing doesn't fix the problem, so pushed.
Thank you for your time on this.
--
Amit Langote
EnterpriseDB: http://www.enterprisedb.com
