Subject: Re: Simultaneous builds
To: Martin Weber <Ephaeton@gmx.net>
From: Manuel Bouyer <firstname.lastname@example.org>
Date: 01/11/2002 18:55:59
On Fri, Jan 11, 2002 at 06:39:49PM +0100, Martin Weber wrote:
> On Fri, Jan 11, 2002 at 06:34:12PM +0100, Manuel Bouyer wrote:
> > [...]
> > I think this is because of broken linux NFS client, for which file
> > creation isn't atomic (it can delay file creation on the server,
> > instead of checking immediatly if the file already exists), making
> > open(O_CREAT|O_EXCL) broken.
> > On NetBSD shlock should be fine.
> If that is the case our shlock manpage needs to be rewritten a bit,
> Does not work on NFS or other network filesystem on different systems be-
> cause the disparate systems have disjoint PID spaces.
Ha yes, but this is because shlock tries to ping the process and
can't do it if it's on another host. It's not because a NFS issue.
What we could do is have shlock record the name of the host in
addition to the pid, and skip the pid check if the host name is different
Manuel Bouyer, LIP6, Universite Paris VI. Manuel.Bouyer@lip6.fr