On Thu, Oct 23, 2025 at 06:36:01PM +0900, Shinya Kato wrote:
> Okay, since I'm not strongly attached to this idea, I've removed the
> 0003 patch for now.
The fact that we cannot access this information without a pg_waldump
or a pg_walinspect, which may not be available, and can be expensive,
is a deal-breaker for me.. Or we may not have a direct access to the
WAL segments.
Without the changes in instrument.c from patch 0002, patch 0001 that
implements the basics would not work. So.. I have moved the changes
of instrument.c to 0001, reordered the fields to be more consistent,
did two bumps (catalog, stats file), simplified the docs, then applied
the result.
By the way, Kato-san, what do you think about the attached extra
simplification? With the FPIs counted in bytes, I don't see much a
point in passing around the number of FPIs generated from
XLogRecordAssemble() to XLogInsertRecord() .
--
Michael