This is also a work-in-progress documentation capturing current models for XOS and CORD. It's currently exploratory and contains free-flowing commentary regarding observations from reviewing the originating reference code repository as well as any deviations and recommendations on expressing them as YANG models.
The initial effort is centered around capturing the CORD Subscriber model and its dependent models. We'll be reguarly updating this document as we capture any additional models from XOS/CORD repository in the coming days.
This YANG module is the primary module that will house all XOS related data models going forward. The models for these came from xos/core/models
directory in the XOS repository. It will eventually house the Service
class, the Tenant
class, etc. One notable convention here is the existence of the /api/tenant
and /api/service
configuration tree inside this module. In a sense, we're considering this YANG module to be the master module that all other modules derive from and augments into this module. You can see this augment behavior in the cord-subscriber.yang schema.
As we capture more XOS data models, we will likely organize the additional models as separate YANG modules, such as xos-service
, xos-tenant
, xos-slice
, etc. which will be imported by this module.
For now, we've captured the TenantRoot
and Subscriber
classes.
This module contains basic placeholder configuration data tree and passed into yang-express
.
This YANG module is the primary module that will house all CORD related data models going forward. It currently captures the cord-subscriber
configuration tree and imports the subscriber schema from the cord-subscriber
YANG module. From a pure data modeling perspective, the current originating reference implementation has yet to achieve a clean functional separation between XOS and CORD. This will be a key area of focus as we attempt to define a clear degree of abstraction between the XOS models and CORD models.
This module contains subscriber-related configuration data tree and passed into yang-express
.
There are two main entry-points on the subscriber
instances. I've defined a list subscriber
construct directly in the module cord-core
which basically uses the grouping subscriber-controller
data model. This means that the cord-core
YANG module itself will be the authorative holder of all subscriber instances. Subsequently, I've augmented the xos
module at the /api/tenant
configuration tree to have a new cord
tenant container along with a node:link
to the subscriber list within the cord-core
YANG module. This convention makes it possible to access the subscriber
instances by directly accessing the cord-core
module, such as /cord-core:subscriber
or via the xos
module, such as /xos-core:api/tenant/cord/subscriber
.
This YANG module is based on the CordDevice
class found inside the subscriber.py
(API) module implementation. I think this particular model is rather under-developed and currently not placed in the right place (shouldn't be inside subscriber.py
which should really just be the controller definitions). This module has a potential to be leveraged more effectively if the goal is for this to become one the the core CORD models from which other devices inherit from (which I'm guessing it will be).
We will need to review its association with the cord-subscriber
model and better understand its role in relation with other device oriented data models.
This module provides export definitions and does not contain any configuration data tree.
This module contains the heart of the initial modeling exercise. It captures the CORD Subscriber data model (which extends XOS Subscriber model, which extends XOS TenantRoot model). There are two primary models: grouping subscriber
and grouping subscriber-controller
.
This module provides export definitions and does not contain any configuration data tree.
The 'subscriber' model extends the xos:subscriber
(from the xos-core.yang) and mirrors as closely as possible the CordSubscriberRoot
class. Some deviations were largely around the variable name convention where I've replaced all underscore with a dash (upload_speed is now upload-speed). This is largely to comply with YANG convention where the schema is used to model XML element structure and underscores are not really used in XML based representations (not even sure if it is valid...).
The other key deviation is in the organization of the various attributes. Instead of having a simple flat list of properties, I've grouped them into related 'services' (pseudo-JSON below):
services: { cdn: { enabled: true }, firewall: { enabled: true, rules: [] }, url-filter: { enabled: true, level: 'PG', rules: [] } uverse: { enabled: true } }
Eventually, I think these four hard-coded services will be moved out of the cord-subscriber
data model altogether. I'm not sure what on-boarded services actually implement these features but it should be augmented by those individual service's YANG models into the cord-subscriber data model.
I've internally debated creating this controller model because I thought that the necessary attributes were rather effectively captured in the prior grouping subscriber
schema definition. But the presence of the related
object container in the controller (that shouldn't be in the underlying model) convinced me to model it according to the CordSubscriberNew
class found inside api/tenant/cord/subscriber.py
. This subscriber-controller
model extends the subscriber
model (inheriting all its attributes) and introduces the additional containers for features
, identity
, and related
. Since the main function of the subscriber-controller
model is to essentially layer a view-like overlay on top of the underlying cord-subscriber model, I've introduced a new custom extension construct called node:link
. This is to capture the fact that the various attributes being expressed are simply a reference link to the exact same data that is located at a different place within the same object.
One note on the related
object, it is currently a placeholder container and I expect it will remain that way as part of the model. The reason is, the attributes that are currently mapped inside all come from the VOLT
service (which then has other attributes from VSG
service, etc.). When we get to the step of modeling the various Service
entities, we'll capture the necessary augment behavior in those separate YANG modules (i.e. cord-service-volt.yang, cord-service-vsg.yang, etc.).