Hi,
On 2022-09-15 15:59:52 -0400, Tom Lane wrote:
> Andres Freund <andres@anarazel.de> writes:
> > pgstat: add/extend tests for resetting various kinds of stats.
>
> crake just failed [1] with what appears to be a race condition in
> a test case added by this commit:
>
> diff -U3 /home/andrew/bf/root/HEAD/pgsql/src/test/regress/expected/stats.out
/home/andrew/bf/root/HEAD/pgsql.build/src/test/recovery/tmp_check/results/stats.out
> --- /home/andrew/bf/root/HEAD/pgsql/src/test/regress/expected/stats.out 2022-07-18 13:32:25.155847949 -0400
> +++ /home/andrew/bf/root/HEAD/pgsql.build/src/test/recovery/tmp_check/results/stats.out 2022-09-15
12:44:10.022975164-0400
> @@ -563,7 +563,7 @@
> SELECT sessions > :db_stat_sessions FROM pg_stat_database WHERE datname = (SELECT current_database());
> ?column?
> ----------
> - t
> + f
> (1 row)
>
> -- Test pg_stat_bgwriter checkpointer-related stats, together with pg_stat_wal
>
> It seems likely to me that the n_sessions increment hasn't made it to shared
> memory because some background process happened to have a lock on the stats
> entry when we tried. Don't we need a pg_stat_force_next_flush() call before
> trying to inspect the sessions count?
Indeed. Pushed a fix. To see if there are other omissions, I made
pgstat_report_stat() never flush stats unless force is passed in. No other
failures.
> BTW, the header comment for pgstat_report_stat is badly in need of
> copy-editing.
Hm - you're talking about this sentence?
... Callers can use the returned time to set up
* a timeout after which to call pgstat_report_stat(true), but are not
* required to to do so.
Reads awkward. I'm struggling with the english language trying to improve it
:(.
"Callers can, but are not required to, use the returned time to set up a
timer, whose expiration should trigger a call to pgstat_report_stat(true)."?
Greetings,
Andres Freund