Re: Draft for basic NUMA observability - Mailing list pgsql-hackers

On Mon, Apr 7, 2025 at 11:53 AM Bertrand Drouvot
<bertranddrouvot.pg@gmail.com> wrote:
>
> Hi,
>
> On Mon, Apr 07, 2025 at 10:09:26AM +0200, Jakub Wartak wrote:
> > Bertrand noticed this first in
> > https://www.postgresql.org/message-id/Z/FhOOCmTxuB2h0b%40ip-10-97-1-34.eu-west-3.compute.internal
> > :
> >
> > -               startptr = (char *) BufferGetBlock(1);
> > +               startptr = (char *) TYPEALIGN_DOWN(os_page_size, (char
> > *) BufferGetBlock(1));
> >
> > With the above I'm also not getting wonky (-1) results anymore. The
> > rest of reply assumes we are using this.
>
> yeah, I can see that you added it in v25-0007. In the same message I mentioned
> to "use the actual buffer address when pg_numa_touch_mem_if_required()
> is called?"
>
> So, to be extra cautious we could do something like:
>
> @@ -474,7 +474,7 @@ pg_buffercache_numa_pages(PG_FUNCTION_ARGS)
>
>                                 /* Only need to touch memory once per backend process lifetime */
>                                 if (firstNumaTouch)
> -                                       pg_numa_touch_mem_if_required(touch, os_page_ptrs[idx]);
> +                                       pg_numa_touch_mem_if_required(touch, buffptr + (j * os_page_size));
>
>
> what do you think?

Yeah, I think we could include this too as it looks safer (sry I've
missed that one). Attached v25 as it was , with this little tweak.

-J.