testbed nova instance flavor selection

[Vincent Ladeuil]

Selecting the testbed nova instance flavor broke when we moved from canonistack to hpcloud because those clouds use different names.

In the general case, tests don't consume a lot of resources so we can default to m1.small for canonistack and standard.small for hpcloud (fix in progress for bug #1287955).

Points to consider:
- we want a sane default that works for supported clouds (see below),
- we need a way to override that default for specific packages/tickets while providing some way to avoid abuses (how do we prevent someone from picking an xxl image so their tests complete faster than everyone elses)

ev proposed:
from novaclient.v1_1 import client
import os
import sys

def find_flavour(client, min_vcpu, min_mem):
    '''Find the most minimal flavour that meets the vCPU and memory
    constraints.'''

    flavors = nova_client.flavors.list()
    filtered = [x for x in flavors if x.vcpus >= min_vcpu and x.ram >= min_mem]
    if len(filtered) < 1:
        msg = 'Could not find any flavors that matched the constraints'
        msg = '%s (%d vCPU, %d mem)' % (msg, min_vcpu, min_mem)
        raise KeyError, msg

    by_cpu = sorted(filtered, cmp=lambda x, y: cmp(x.vcpus, y.vcpus))
    by_mem = sorted(by_cpu, cmp=lambda x, y: cmp(x.ram, y.ram))

    return by_mem[0]

if __name__ == '__main__':
    args = [
            os.environ['OS_USERNAME'],
            os.environ['OS_PASSWORD'],
            os.environ['OS_TENANT_NAME'],
            os.environ['OS_AUTH_URL'],
    ]
    nova_client = client.Client(*args)
    print find_flavour(client, int(sys.argv[1]), int(sys.argv[2]))

Which would give us a cloud agnostic default (well, based on min_vcpu and min_mem but that's better than using cloud specific names).