Comment 0 for bug 1655034

Problem
=======

AFAIK, our design looks like this:

                    compute-node
    +-----------------------------------------+
    | |
    | +----------------+ +----------------+ |
    | | n-cpu | | n-cpu | |
    | | cpc_subset=foo | | cpc_subset=bar | |
    | +------+---------+ +--------+-------+ |
    | | | |
    +-----------------------------------------+
             | |
             | |
             | |
             | cpc |
    +-----------------------------------------+
    | | | |
    | +---v----+ +---v----+ |
    | | subset | | subset | |
    | | foo | | bar | |
    | +--------+ +--------+ |
    | |
    +-----------------------------------------+

A system z CPC is divided into multiple subsets. Each subset is managed
by exactly one nova-compute service. Each nova-compute service manages
exactly one subset.
AFAIK, this setup -having multiple nova-compute services on one operating
system- is (AFAIK) not tested upstream. That means we have a testing gap.

Definition of Done
==================
As this setup cannot be properly tested in a unit test setup, we should
have a functional test [1] for this. Examples for that can be seen in the
Nova repo [2]. The "tests" directory [3] should have a sub directory for
that. The functional test case for that:
* uses a faked zhmcclient
* simulates two cpc-subsets
* starts two nova-compute services
* starts (at least) two instances which should be scheduled to different

References
==========
[1] http://docs.openstack.org/developer/nova/test_strategy.html#functional-tests
[2] https://github.com/openstack/nova/tree/master/nova/tests/functional
[3] https://github.com/openstack/nova-dpm/tree/master/nova_dpm/tests