Cluster Testing

[Adrien Guillon]

Hello,

I have recently started to use qmtest for testing my open source
project.  My project is designed to run on compute clusters, therefore
I'm looking at how to use qmtest to test my code on a cluster.  My
test cases will do nasty things, like terminate the process on one
node, to see how reliable my application is.  I am hoping that someone
has already tackled this problem, and can provide some code.
Alternatively, I would love to have some help to create an extension
that meets my needs, and contribute that back to the community.

I did see that qmtest has the ability to run commands or compile
applications on remote nodes via SSH or RSH.  I'm not sure what a
cluster test harness would look like, but I imagine that I would want
to transfer source code to a remote host, compile it, and run the
application.  The selection of nodes may play a role in the test as
well, and you might not always want to run the same program on each
node.  For example, I might select a node which is big-endian and a
little-endian node to check that communication between the two is
sane.  I might also test just the server and client communication part
of an application, so I might have two different applications rather
than just one to run.  I may also have multiple clusters to test with,
each with machines with different properties.

I would appreciate general advice on how to proceed with qmtest
development.  Should I develop a test framework independently, then
integrate it later with qmtest?

Thanks,

AJ