Re: monitoring usage count distribution

On Wed, Apr 5, 2023 at 4:16 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Nathan Bossart <nathandbossart@gmail.com> writes:
> > On Wed, Apr 05, 2023 at 12:09:21PM -0700, Andres Freund wrote:
> >> I would not mind having a separate function returning 6 rows, if we really
> >> want that, but making pg_buffercache_summary() return 6 rows would imo make it
> >> less useful for getting a quick overview. At least I am not that quick summing
> >> up multple rows, just to get a quick overview over the number of dirty rows.
>
> > This is what v1-0001 does.  We could probably make pg_buffercache_summary a
> > view on pg_buffercache_usage_counts, too, but that doesn't strike me as
> > tremendously important.
>
> Having two functions doesn't seem unreasonable to me either.
> Robert spoke against it to start with, does he still want to
> advocate for that?

My position is that if we replace the average usage count with
something that gives a count for each usage count, that's a win. I
don't have a strong opinion on an array vs. a result set vs. some
other way of doing that. If we leave the average usage count in there
and add yet another function to give the detail, I tend to think
that's not a great plan, but I'll desist if everyone else thinks
otherwise.

--
Robert Haas
EDB: http://www.enterprisedb.com