Hi,
On 2019-05-20 13:46:55 -0400, Tom Lane wrote:
> Andres Freund <andres@anarazel.de> writes:
> > On 2019-05-21 01:43:35 +0900, Fujii Masao wrote:
> >> Thanks for the commit! And sorry, make check-world took very long time
> >> and has just finished right now in my env... I'd like to confirm that
> >> make check-world reports no error just in case.
>
> > make -j16 -s && make -Otarget -j10 -s check-world && echo success || echo fail
>
> > makes that much more bearable, by running the slow TAP tests in
> > parallel. Doesn't work < 10 though.
Kinda wonder whether we should backpatch the necessary changes to make
-j .. check work back to 9.4.
> Yeah, parallelization makes all the difference in the world, assuming
> you have a multi-core machine. You can parallelize within TAP suites
> too. I use
>
> make -s check-world -j4 PROVE_FLAGS='-j4 --quiet --nocolor --nocount'
>
> on an 8-core machine, and get typical runtimes under 2m30s.
FWIW, I found that parallelizing on the top-level to get better
performance than within prove. The system load is more predicatable.
> If there were a way to get prove to be absolutely quiet (its --quiet is
> a joke unfortunately), I'd lobby for a switch to make pg_regress quiet
> too. The amount of useless noise this recipe generates is annoying,
> and tends to make it hard to find where the actual error is when
> there is one.
Indeed. Couldn't we just invent that on our end? I.e. in our prove
invocation, just redirect the output?
The most annoying noise imo is the pg_upgrade test. The set -x and the
fact that it resets MAKEFLAGS makes that pretty annoying.
Greetings,
Andres Freund
