[qmtest] partial order on prereqs independent of outcome?

[Allyn Dimock]

On Thu, 6 Apr 2006, Stefan Seefeld wrote:

> [...] QMTest simply doesn't have
> any way to express can-run-in-parallel and can-not-run-in-parallel 
> properties.

I understand that there should be other ways of dealing with parallel vs 
not in parallel.  I had thought that there might have been some call 
before for multiple allowable outcomes on a pre-req.  For instance "ERROR 
or FAIL".  If that feature existed, I could build on it.

I can't decrease the parallelism from 4 to 1 since that would increase the 
test time from 6 to 24 hours per cycle.

The obvious solution is to throw independent hardware at the problem.
I can't see how to use ssh from a batch process to get access to multiple 
systems without exposing my passphrase or my password.  Using a background 
interactive process where I run ssh-agent is vulnerable to the occasional 
system reboot.

One of these days, when not fighting fires, I will put a 
PBS back end on to solve this problem.  PBS's idea of distributed batch 
processing is a poor match for QMTest's idea of being in control of the
order of processing, but if I run QMTest with more threads than the number 
of simultaneous processes that PBS will allow me to run, then I should be 
able to jury-rig something.  It's just that I have been trying for the 
last several months to find a bit of free time to do this and there has 
always been a higher-priority task that needs doing -- I'm sure that you 
know the situation.

-- Allyn