Re: Suppressing noise in successful check-world runs

On Fri, May 24, 2019 at 12:31 PM Peter Geoghegan <pg(at)bowt(dot)ie> wrote:
>
> On Wed, May 22, 2019 at 3:57 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > I experimented with the attached quick-hack patch to make pg_regress
> > suppress notices from its various initial DROP/CREATE IF [NOT] EXISTS
> > commands. I'm not entirely convinced whether suppressing them is
> > a good idea though. Perhaps some hack with effects confined to
> > pg_upgrade's test would be better. I don't have a good idea what
> > that would look like, however.
> >
> > Or we could just say this isn't annoying enough to fix.
>
> I think it's worth fixing.

My development machine has 8 logical cores, and like you I only see
the NOTICE from pg_upgrade's tests with "-j10":

pg(at)bat:/code/postgresql/patch/build$ time make check-world -j10 >/dev/null
NOTICE: database "regression" does not exist, skipping
make check-world -j10 > /dev/null 86.40s user 34.10s system 140% cpu
1:25.94 total

However, I see something else with "-j16", even after a precautionary
clean + rebuild:

pg(at)bat:/code/postgresql/patch/build$ time make check-world -j16 >/dev/null
NOTICE: database "regression" does not exist, skipping
pg_regress: could not open file
"/code/postgresql/patch/build/src/test/regress/regression.diffs" for
reading: No such file or directory
make check-world -j16 > /dev/null 96.35s user 37.45s system 152% cpu
1:27.49 total

I suppose this might be because of a pg_regress/make file
"regression.diffs" race. This is also a problem for my current
workflow for running "make check-world" in parallel [1], though only
when there is definitely a regression.diffs file with actual
regressions -- there is no regression that I'm missing here, and as
far as I know this output about "regression.diffs" is just more noise.
I had intended to figure out a way of keeping "regression.diffs" with
my existing workflow, since losing the details of a test failure is a
real annoyance. Especially when there is a test that doesn't fail
reliably.

[1] https://wiki.postgresql.org/wiki/Committing_checklist#Basic_checks
--
Peter Geoghegan