Em 24/08/2018 23:02, Charlie Gonzalez escreveu:
Lately, I have been updating a disabled.yml (
https://github.com/itcharlie/cpan-testing-configs/blob/master/disabled.yml
) file on distros that break while installing them on Strawberry Perl
in Windows ( I am not bothering to look at the reasons why the modules
are failing to install ).
At least one of the reasons is that doing IPC on Windows sucks big time.
On UNIX-like OS that is much more reliable (although sometimes I have
the impirical impression that on OpenBSD is slower than compared to
Linux), but on Windows it just fails from time to time without aparent
reason.
The only sane way to exchange data between two processes over there is
through sockets.
That being said, I wonder why we don't have something like a "make"
daemon, waiting to receive requests. The processe of doing smoke tests
involves creating a lot of processes that run ony for a short period of
time, at least that is what I can see checking OpenBSD with vmstat
during a smoke test (lots of context switch compared to other uses of CPU).
I'm not saying that this is something easy to fix, far from it. But
maybe it would worth the effort, I guess.
After upgrading the server instance with 2 virtual cpus and 5.5 gigs
of ram , google cloud monitoring has stop yelling at me with
notifications to upgrade the server specs and CPAN::Reporter::Smoker
seems to be running a bit quicker now.
5.5GB looks a lot to me, but I'm not playing with Windows and Perl for a
while nowadays. Are you running multiple smokers at the same time?
You could look for using RAM disks on Windows (not very useful 5 years
back) to store the build_dir content. That should help a bit.
Since then I have set up 2 more servers to run CPAN tester reports for
Perl 5.26 and 5.24 ( I am using this opportunity to learn how its
done and hash out the steps needed to setup a Smoker on Windows )
If you could, it would be great to have your notes on the CPAN Testers
wiki (http://wiki.cpantesters.org/). :-)
