I'd like opinions and answers on the following items:
- Has the idea of abstracting NF out of VNF and PNF been considered in the TOSCA node types proposed for ETSI NFV? I think this is of interest to the PNF team and the VES team. See the tosca.nodes.nfv.NF code block and then following discussion for questions.
- I know that use of TOSCA semantics was avoided in the SOL001 spec to not require a consumer of the SOL001 spec to need to use TOSCA orchestration. That means to me that SOL001 is a logical data model (LDM) that just happens to use TOSCA vs. an implementation-specific data model. ONAP is an implementation therefore it should be possible for different components to prune and refactor the model to something most useful. If a component or set of components wants to use TOSCA semantics for orchestration those component(s) are thus free to change the model. It means we could change the TOSCA proposed in SOL001 into TOSCA that DOES leverage TOSCA semantics. In the tosca.nodes.nfv.NF code block, you can see that capabilities are used to capture some of the information that was documented in the LDM that was captured in a node.
tosca.nodes.nfv.NF
# There is no need to derive VNF or PNF node types at the *data model* level # Use composition to cover the differences tosca.nodes.nfv.NF: derived_from: tosca.nodes.Root properties: descriptor_id: type: string # GUID required: true descriptor_version: type: string required: true provider: type: string required: true product_name: type: string required: true software_version: type: string required: true product_info_name: type: string required: false product_info_description: type: string required: false localization_languages: type: list entry_schema: type: string required: false default_localization_language: type: string required: false # configurable_properties: # type: tosca.datatypes.nfv.VnfConfigurableProperties <== remove Vnf prefix? # required: true # modifiable_attributes: # type: tosca.datatypes.nfv.VnfInfoModifiableAttributes <== remove Vnf prefix? # required: true # lcm_operations_configuration: # what use is made of this and by whom? # type: tosca.datatypes.nfv.NfLcmOperationsConfiguration # description: Describes the configuration parameters for the NF LCM operations # required: true # monitoring_parameters: # type: list # entry_schema: # type: tosca.datatypes.nfv.NfMonitoringParameter # description: Describes monitoring parameters applicable to the NF. # required: false flavour_id: type: string required: true # flavour_description: # should be in the NF's DF definition and not need to be provided here (join it with the id) # type: string # required: false # vnfm_info: # I wouldn't recommend putting this in any static model. Maybe a reference to a dynamic property? # type: list # entry_schema: # type: string # required: true #capabilities: # monitoring_parameter: # modelled as ad hoc capabilities in the VNF node template requirements: - virtual_link: capability: tosca.capabilities.nfv.VirtualLinkable relationship: tosca.relationships.nfv.VirtualLinksTo node: tosca.nodes.nfv.VirtualLink # Change to remove vnf occurrences: [ 0, UNBOUNDED ] # Work with OASIS on normative representations of compute, storage, architecture, etc. # I'll just lump them into ExecutionEnvironmentAspects for this example tosca.capabilities.nfv.Moveable: derived_from: tosca.capabilities.Root # Only PNFs could have this capability tosca.capabilities.nfv.PhysicallyMoveable: derived_from: tosca.capabilities.nfv.Moveable # Only VNFs could have this capability tosca.capabilities.nfv.VirtuallyMoveable: derived_from: tosca.capabilities.nfv.Moveable # Only PNFs could have this capability tosca.capabilities.nfv.PurposeBuiltHostingPlatform: derived_from: tosca.capabilities.Root properties: provider: type: string required: true serial_number: type: string required: false attributes: serial_number: type: string required: true valid_source_types: []
I need to understand what all the subtyping and indirection is doing in the "modifiable properties example" code block. There is a data type defined which extends VnfInfoModifiableAttributes. It defines its own derived datatype for the extensions property that introduces the http_proxy and https_proxy properties. If I had simply put http_proxy and https_proxy as properties on my derived VNF, I would have had equal access to them in the script; I would have referenced them in a shorter way (i.e., \[vnf, http_proxy\]). I am missing the beauty of all the indirection. Can someone help?