At Fri, 24 Dec 2021 18:24:39 +0900, Fujii Masao <masao.fujii@oss.nttdata.com> wrote in
>
>
> On 2021/12/24 18:00, Kyotaro Horiguchi wrote:
> > I saw the test has been revertd.
>
> Yes, I reverted the added unstable tests not to prevent
> buildfarm from testing other patches while I'm doing
> the investigation.
>
> > https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=crake&dt=2021-12-24%2008%3A02%3A25
> >
> >> NOTICE: identifier
> >> "pg_regress/postgres_fdwbuildfarmcontrib_regression_postgres_fdw757365%"
> >> will be truncated to
> >> "pg_regress/postgres_fdwbuildfarmcontrib_regression_postgres_fdw"
> > It's 70 characters long..
> > application_name: pg_regress/postgres_fdw
> > user_name : buildfarm
> > database_name : contrib_regression_postgres_fdw
> > Source PID : 757365
> > Maybe we can distribute the placeholders into several sessions.
>
> Or probably we don't need to test all escape sequences. How about
> picking up one or two from them? But even if only one or two are
> picked up, application_name still can be larger than 63 characters. So
> probably we also should use substring() in the test query, for that
> case. For example,
>
> SELECT count(*) > 0 FROM pg_stat_activity
> WHERE application_name = substring(current_setting('application_name')
> ||
> CURRENT_USER || current_database() || pg_backend_pid() || '%', 1, 63);
I once thought the same, but that change causes buildfarms *always*
truncate the resulting application name. I believe the base
application_name and database_name never grows further longer. So I'd
like to split the check to %a%u%% and %d%p.
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center