(POC) (8) Create inception-based testing for libvirt functionality

[herlo]

Because linch-pin now has a plugin for libvirt provisioning, additional libraries must be installed. It seems it may be unreasonable to expect a jenkins (or other CI) to be willing to install libvirt on a node which they control. To that end, testing the libvirt plugin in linch-pin may require a bit of inception-based testing.

The concept is as follows:

1. Provision a single node from a CI slave using another plugin (duffy, ec2, openstack). Alternatively, a container could be provisioned, instead?
2. Use an ansible playbook to connect to the single node to provision via libvirt and perform tests
3. ????
4. Profit

[arilivigni]

Yes this is pretty much we discussed here in the office:

1. Provision duffy node
2. Checkout PR of linch-pin on slave (same as today)
3. Run playbook on duffy node that is in linch-pin/tests/playbook does the following:
   • Pass sha1 to playbook
   • Build and execute tests in a venv
   • Copy xunit results to slave workspace
4. Archive test results (same as today)
5. publish test results (same as today)

Tasks:
1, 2 and 3 need to be created and JJB updated

[herlo]

After conversations here at PyCon, I discovered an easy way to implement this by adjusting the host value. After provisioning the first node, modify linchpin/provision/hosts to represent localhost as a remote node. From there, run linchpin again to configure and provision libvirt machines on the remote node. The useful part is to abuse the hosts file feature which allows us to first run on localhost to provision a node, then run on localhost to provision a vm on the remote node.

[herlo]

@p3ck has created this with the Dockerfile tests.