Re: relfilenode statistics

On Sat, Jul 11, 2026 at 04:16:13AM +0000, Bertrand Drouvot wrote:
> Maybe resume this thread [1] instead of creating a new one? It's in this thread
> that it has been suggested to start with relfilenode stats first [2] but that
> was due to the design of the patch at that time that may not apply to yours (did
> not look at it yet).
>
> [1]: https://postgr.es/m/f572abe7-a1bb-e13b-48c7-2ca150546822%40gmail.com
> [2]: https://postgr.es/m/20230105002733.ealhzubjaiqis6ua%40awork3.anarazel.de

Honestly, I don't know which way is the best. This was 3 years ago so
starting afresh may be better at the end. Please note that I have
spent a couple of hours eyeing at your previous patch, where I'd like
to think that I have found an approach with a better balance,
finishing by being less invasive. The two main points are 1) the
handling of the xact data for a split for the fields we care about and
2) a union structure felt less intrusive, but these two points really
come down to where you want to put a separation line. Another side thing
is the renaming of the SQL functions for non-index relkinds, your
patch using *_tab_*.
--
Michael