On Mon, Apr 11, 2005 at 03:41:12PM -0500, Nathan Tallent wrote:
> >Can you explain further why storing the current session dir inside
> >daemonrc is a strait jacket? Specifically, what behaviour would break?
> I was referring primarily to having fixed locations for the
> configuration file and sample directories. This thread has been
> discussing the fixed location of the sample directory and in the
> context, I had complained about the fixed location of the configuration
> file. When these locations are fixed, it becomes a problem on
> Clustermatic systems. Clearly samples cannot be stored in a RAM disk
> (small size and no persistence) and locating the config file at /root is
> not optimal because of persistance and the fact that the RAM disks
> contain no /root without special configuration.
My worries about virtualizing the daemonrc location is that daemonrc is
very much an implementation detail: it can change, or disappear at will.
I don't want to encourage people to believe the file is directly
editable, because it isn't.
Also, it's worth noting that this config file is user-specific. It's not
at all a stretch to image ~toor/.oprofile/daemonrc etc.
You must have some method of enabling oprofile on the nodes in the first
place: in that case, isn't it equally possible that that method also
includes the opcontrol invocations you need to configure? I suppose
you'd still need a /root/.oprofile/ in your RAM disk for this situation
but I don't see that as a big burden.
If I'm misunderstanding something here, let me know - if it proves to be
really needed, then we could consider an environment variable locating
the daemonrc. But I'd rather not do that right now.