Code review comment for lp:~usc-isi/nova/instance_type_extra_specs

The code looks rather solid, however I do have a few higher-level questions:

1) The concept of 'metadata' doesn't fit in very well here. The data you are providing is integral to the provisioning process. Our usage of metadata in other places represents user-defined key/value pairs that our services will never care about. cpu_arch would probably fit better as a core instance type attribute, while we could look at the others as something closer to "optional attributes". An instance could use one of the standard instance types with a don't-care attitude when provisioning towards something like xpu_arch, while you could provide more specific instance types with those optional attributes being explicitly defined.

2) I'm not completely sold on the idea that a deployer should provide another method for communication of instance types. I would love to integrate this information into the existing flavors resource in the openstack api.

3) Am I correct in assuming another merge prop will be coming to fill in the scheduler integration? Otherwise, there is no way to respect whatever values are actually indicated in the metadata.