Re: WAL-logging facility for pgstats kinds

On Thu, Jan 02, 2025 at 08:08:42PM -0500, Andres Freund wrote:
> I can't think of a real case where we would want to WAL log the stats
> themselves, rather than re-emitting stats during replay based on the WAL
> record of the "underlying object".
>
> Do you have counter-examples?

I'm not sure if the rebuild based on the WAL records is simpler than
logging a snapshot of them that the startup process could digest.

Anyway, I've also wanted to be able to replicate stats for historical
tracking for stats logged through hooks, and a second case was
injection point stats. All these would require registering a custom
rmgr on top of a custom stats kind, and centralizing the logic eases a
lot some of the sanity checks they'd require as the redo callback of a
stats kind can just be attached to its PgStat_KindInfo.

I know that Lucas has been playing a bit with the area, and perhaps
he has more cases in mind where the replication of stats data could be
relevant. So I am adding him in CC. Perhaps I could be wrong, of
course.
--
Michael