Running multiple tinderd instances

[Boris Samorodov]

On Thu, 12 Mar 2009 13:48:03 +0000 Tim Bishop wrote:
> On Wed, Mar 11, 2009 at 11:48:21PM +0300, Boris Samorodov wrote:
> > On Wed, 11 Mar 2009 14:25:06 +0100 Beat Gätzi wrote:
> > > A patch for this was
> > > posted on the tinderbox list last December:
> > > http://www.marcuscom.com/pipermail/tinderbox-list/2008-December/001390.html
> > 
> > At one tindy (it's fast amd64) I'd had some strange artefacts:
> > . sometimes one port was build in parallel ${tinderd_instances}
> >   times;
> > . sometimes it took long time to start parallel buildings;
> > . etc.
> > 
> > It seemed to be a race somewhere. And while looking at the above
> > mentioned patch I noticed one common directory /tmp/tinderdlock
> > which was used by all tinderd daemons. Since I've done:
> > -----
> > % sed -e 's:/tmp/tinderdlock:/tmp/tinderdlock.$$:g' tinderd
> > -----
> > ... I've never seen those artefacts.

> Isn't the point of that lock to ensure the multiple tinderd processes
> don't try and do that section of code at the same time? By making it
> unique you're defeating the point of it.

Hm, you are definitely right. I need to re-analyse my results though...


WBR
-- 
Boris Samorodov (bsam)
Research Engineer, http://www.ipt.ru Telephone & Internet SP
FreeBSD Committer, http://www.FreeBSD.org The Power To Serve