Re: Huge memory consumption on partitioned table with FKs

On Fri, Dec 4, 2020 at 2:48 PM Kyotaro Horiguchi
<horikyota(dot)ntt(at)gmail(dot)com> wrote:
> At Fri, 4 Dec 2020 12:00:09 +0900, Keisuke Kuroda <keisuke(dot)kuroda(dot)3862(at)gmail(dot)com> wrote in
> > Hi Amit,
> >
> > > I have attached a patch in which I've tried to merge the ideas from
> > > both my patch and Kuroda-san's. I liked that his patch added
> > > conparentid to RI_ConstraintInfo because that saves a needless
> > > syscache lookup for constraints that don't have a parent. I've kept
> > > my idea to compute the root constraint id only once in
> > > ri_LoadConstraint(), not on every invocation of ri_BuildQueryKey().
> > > Kuroda-san, anything you'd like to add to that?
> >
> > Thank you for the merge! It looks good to me.
> > I think a fix for InvalidateConstraintCacheCallBack() is also good.
> >
> > I also confirmed that the patch passed the make check-world.
>
> It's fine that constraint_rood_id overrides constraint_id, but how
> about that constraint_root_id stores constraint_id if it is not a
> partition? That change makes the patch a bit simpler.

My patch was like that before posting to this thread, but keeping
constraint_id and constraint_root_id separate looked better for
documenting the partitioning case as working differently from the
regular table case. I guess a comment in ri_BuildQueryKey is enough
for that though and it's not like we're using constraint_root_id in
any other place to make matters confusing, so I changed it as you
suggest. Updated patch attached.

--
Amit Langote
EDB: http://www.enterprisedb.com