On Mon, Mar 16, 2026 at 5:27 PM Fujii Masao <[email protected]> wrote:
>
> On Fri, Mar 13, 2026 at 11:52 AM Sami Imseih <[email protected]> wrote:
> >
> > > it would be better to verify that the stats of the specified sequence are
> > > actually reset by using pg_stat_reset_single_table_counters().
> > >
> > > Also, the documentation for pg_stat_reset_single_table_counters() should
> > > mention that it can reset sequence statistics as well.
> >
> > good points.
> >
> > For the  documentation of pg_stat_reset_single_table_counters(), I also
> > mentioned materialized views for completeness.
>
> Thanks for updating the patch!
>
> I made a few cosmetic changes and pushed the patch. Thanks!

Hmm... buildfarm member crake reported a pg_upgradeCheck failure, and it seems
the test I added last time is causing it :(

--- /home/andrew/bf/root/HEAD/pgsql/src/test/regress/expected/stats.out
2026-03-16 04:27:05.805748763 -0400
+++ 
/home/andrew/bf/root/HEAD/pgsql.build/testrun/pg_upgrade/002_pg_upgrade/data/results/stats.out
2026-03-16 05:17:15.460202339 -0400
@@ -1196,7 +1196,7 @@
   FROM pg_statio_all_sequences WHERE relname ='test_seq1';
  ?column? | ?column?
 ----------+----------
- t        |        0
+ t        |        1
 (1 row)

The test assumed that blks_read and blks_hit in pg_statio_all_sequences would
be zero immediately after calling pg_stat_reset_single_table_counters().
However, on crake, either of them became 1 even right after the reset.
This might happen if another process (e.g., autovacuum) accesses
the sequence in the short window between the reset and the query of
pg_statio_all_sequences??? More investigation would be necessary.

Since checking the blks_read and blks_hit counters in pg_statio_all_sequences
can make the test unstable, I'm thinking of removing that part of the test.

Regards,

-- 
Fujii Masao


Reply via email to