Subject: Re: pkgsrc on SMP machines
To: None <segv@netctl.net>
From: Geert Hendrickx <ghen@telenet.be>
List: tech-pkg
Date: 12/22/2005 08:39:00
On Wed, Dec 21, 2005 at 10:49:15PM +0000, segv@netctl.net wrote:
> On Wed, 21 Dec 2005 23:23:45 +0100
> Lars Nordlund <lars.nordlund@hem.utfors.se> wrote:
>
> > On Tue, 20 Dec 2005 00:20:30 +0000
> > segv@netctl.net wrote:
> > > > Have you tried turning off WRAPPER_UPDATE_CACHE like I suggested
> > > > earlier?
> > >
> > > No, I didn't realise it would cause problems. I'll give it a try.
> > > Thanks for the pointers.
> >
> > Have you had time to test -j6 building on an SMP machin with the
> > wrapper cache turned off? Did it work?
> >
> >
> > Best regards,
> > Lars Nordlund
>
> Yes I did, it seems to have worked, i.e. I've been building packages and have
> not yet encountered those errors (which previously originated from wrapper
> scripts)
>
> Below is what I have in my mk.conf to get make fork multiple jobs:
>
> WRAPPER_UPDATE_CACHE=no
> MAKE_FLAGS+=-j4
>
> on a 4 CPU Sun machine, www/firefox build is using all CPUs most of the time.
>
> To make this more automatic, we need to define a new variable MAKE_JOBS, so
> that the users could put MAKE_JOBS=4 in mk.conf. Each package would need to
> define MAKE_JOBS_SAFE=yes if it has been tested to support parallel make.
>
> mk/bsd.pkg.mk would need to have something like:
>
> .if ${MAKE_JOBS_SAFE} == "yes"
> MAKE_FLAGS+=${MAKE_JOBS}
> .else
> MAKE_FLAGS+=-j1
> .endif
See http://mail-index.netbsd.org/tech-pkg/2005/12/16/0005.html for an
implementation of this.
Geert