misc/tinderbox (fwd)

[Dmitry Morozovsky]

Dear colleagues,

it seems -jobs options now produce harm instead of intended purpose and should 
be disabled.

Sincerely,
D.Marck                                     [DM5020, MCK-RIPE, DM3-RIPN]
------------------------------------------------------------------------
*** Dmitry Morozovsky --- D.Marck --- Wild Woozle --- marck at rinet.ru ***
------------------------------------------------------------------------

---------- Forwarded message ----------
Date: Tue, 2 Jan 2007 13:48:31 +0300 (MSK)
From: Dmitry Morozovsky <marck at rinet.ru>
To: Ion-Mihai IOnut Tetcu <itetcu at FreeBSD.org>
Subject: Re: misc/tinderbox

On Tue, 2 Jan 2007, Ion-Mihai IOnut Tetcu wrote:

IIT> > playing with your baby port, I found that parallel jobs in
IIT> > tinderbuild simply do not work: logs are not serialized, and after
IIT> > finishing building of one port parallell builds got killed.
IIT> 
IIT> Are you trying to do parallel builds in the same build ? This
IIT> shouldn't work at all AFAIK.
IIT> 
IIT> I don't have any problems doing them in different builds.
IIT> 
IIT> Could you tell me exactly how you do them ?

Cite from the distrib:

root at tiger:/ar/tinderbox/scripts# ./tinderbuild
usage: ./tinderbuild -b <build name> [-init] [-cleanpackages] [-updateports] 
[-skipmake] [-noclean] [-noduds] [-plistcheck] [-nullfs] [-cleandistfiles] 
[-fetch-original] [-nolog] [-trybroken] [-jobs] [-onceonly] [portdir/portname 
[...]]
root at tiger:/ar/tinderbox/scripts# grep -3 jobs README 
 -nolog          : disables log analysis code
 -trybroken      : builds ports marked as BROKEN (this does NOT require
                   -noduds)
 -jobs           : starts <n> number of parallel port builds (NOTE: the default
                   is 1, and for best results should not exceed the number
                   of physical CPUs in the Tinderbox host)
 -onceonly       : only performs one build pass (i.e. tinderbuild Phase 1)
root at tiger:/ar/tinderbox/scripts# 

I suppose -jobs option should be simply disabled until proper serialization has 
invented.

As for different builds: in my situation, there is mostly one build (RELENG_6 + 
fixed ports subset); moreover, on contemporary (not too hi-end) hardware build 
process is usually io-bound, not cpu-bound, hence on rather cheap machine with, 
say, 2G of RAM and a pair os SATA drives (would cost below $500-600) 3 jobs per 
CPU core seems the most reasonable.

Sincerely,
D.Marck                                     [DM5020, MCK-RIPE, DM3-RIPN]
------------------------------------------------------------------------
*** Dmitry Morozovsky --- D.Marck --- Wild Woozle --- marck at rinet.ru ***
------------------------------------------------------------------------