nfv-tosca definition is inaccurate.

[xiaodong shang]

The definition of TOSCA_nfv_definition_1_0.yaml is not consistent with the specification of "tosca-nfv-v1.0", a example of VDU definitions are shown belown:
  In TOSCA_nfv_definition_1_0.yaml:
    tosca.nodes.nfv.VDU: # openstack community
        derived_from: tosca.nodes.Compute
        ...
  But in tosca-nfv-v1.0:
    tosca.nodes.nfv.VDU: # nfv spec
        derived_from: tosca.nodes.SoftwareComponent
        ...

  I think the demand of specification must be satisfied.

[Sahdev Zala (spzala)]

Hi Shang, thanks for opening this bug. Are you planning to work on it? OR we can have Bob Haddleton take a look at it? Thanks!

[Sahdev Zala (spzala)]

We had a good discussion on #openstack-heat-translator IRC meeting March 31, 2016 and can be found in IRC log. Per discussion, current spec of deriving from 'tosca.nodes.SoftwareComponent' is a poor choice for implementation and it's under discussion and to be changed in future version of spec. So no changes in code/definition required for now. We will continue monitoring it.

[Matt Rutkowski (mrutkows)]

Actually, the TOSCA NFV Profile WG has changed VDU to derive from a Compute (variant) and it has been approved by the TC and will be published ASAP as CSD03 of the NFV Profile. In addition, new NUMA architecture (CPU) capabilities have been defined to allow great control of Compute node allocation at a processor level using Nova.

[xiaodong shang (shangxdy)]

VUD is a example to denote the difference between TOSCA_nfv_definition_1_0.yaml and spec. Will the definitions of VL and CP also be changed in CSD03? I hope so.

[Matt Rutkowski (mrutkows)]

Agree, the change from SoftwareComponent to Compute (inheritance) is new. It is just being published in latest NFV Profile. I was assuming that someone from Tacker would update the definition in the tosca-parser/heat-translator for this release cycle. However, I would defer to John or Sridhar as my contacts into Tacker project on what they need.

[Matt Rutkowski (mrutkows)]

Just to be clear, we need to respond to any needs of the Tacker project in order to get them running for next release.

[xiaodong shang (shangxdy)]

Well, i am also very interested in tacker, and hope to have a little contribution to this project too.

[Sridhar Ramaswamy (srics-r)]

@shangxdy - I remember chatting with you in the last heat-translator meeting. Please do join / contribute to Tacker project. You can reach us at #tacker channel

Just to recap our previous discussion and this bug. tosca-nfv-v1.0 spec using SoftwareComponent for VDU is not implementable for Tacker. Hence the code is using "tosca.nodes.Compute" for VDU. The next version of TOSCA NFV spec (CSD03) will have changes towards making VDU node_type based on Compute. It will also have changes to VL and CP to be based off Network and Port in the TOSCA Simple profile.

In short, this is *NOT* a bug but an intentional change in the implementation which we have taken back to incorporate in the standard specification's next version.

[Matt Rutkowski (mrutkows)]

So it seems we want to leave it as compute as per the latest CSD03 coming out and likely close out this bug as "working according to latest spec."

[Sahdev Zala (spzala)]

That's the decision we made in the meeting last week (something I have commented in this bug as well after the meeting), we are just keeping the bug open for now to spec officially have changes and then close this bug.