Hi,
On 2022-03-29 12:17:27 -0700, Andres Freund wrote:
> Separate from the minutia in [1] I'd like to discuss a few questions of more
> general interest. I'll post another question or two later.
5) What does track_counts = off mean?
I just was trying to go through the shared memory stats patch to ensure
pgstat_track_counts has the same effect as before. It's kinda hard to discern
what exactly it supposed to be doing because it's quite inconsistently
applied.
- all "global" stats ignore it (archiver, bgwriter, checkpointer, slru wal)
- subscription, replication slot stats ignore it
- *some* database level stats pay heed
- pgstat_report_autovac, pgstat_report_connect don't
- pgstat_report_recovery_conflict, pgstat_report_deadlock,
pgstat_report_checksum_failures_in_db do respect it
- function stats have their own setting
Unless we conclude something else I'll make sure the patch is "bug
compatible".
Greetings,
Andres Freund
