Re: Proposal: OProfile & Clustermatic

[John L. <le...@mo...>]

On Mon, Apr 11, 2005 at 12:53:32PM -0500, Nathan Tallent wrote:

> >I don't know about that. --session-dir should be sticky (that is, live
> >in daemonrc). So operations should operate on the current session dir.
> 
> This won't "just work" if your file sytem is not "sticky" (i.e. a reboot 
> clears the RAM disk and daemonrc).

But that case is really not relevant. If you want something to work,
then naturally you need a daemonrc with the location of the session.
Just because the session isn't persistent across a reboot is no good
reason to make it non-persistent across multiple opcontrol invocations.

I'm trying to make sure these new facilities fit in with the traditional
behaviour as well as what you require.

> >Your way leads to absurdities such as having to specify --session-dir to
> >--reset.
> 
> No... there's no need for --reset with DCPI-like behavior because you 
> specify a directory each time you restart the daemon.

We can support such a feature, but it will not be "the default". I want
this option to make sense for other uses, namely a different location,
but not necessarily a different directory each time.

> I agree that "just working" is nice (so defaults are good) but they 
> aren't so nice when they become a straight jacket.

Can you explain further why storing the current session dir inside
daemonrc is a strait jacket? Specifically, what behaviour would break?

regards,
john