Re: [Teas] Yangdoctors last call review of draft-ietf-teas-yang-te-topo-08

[Benoit Claise <bclaise@xxxxxxxxx>]

Hi Xufeng,

Hi Benoit,

-----Original Message-----
From: Benoit Claise [mailto:bclaise@xxxxxxxxx]
Sent: Thursday, June 22, 2017 6:12 AM
To: Robert Wilton <rwilton@xxxxxxxxx>; Xufeng Liu <Xufeng_Liu@xxxxxxxxx>;
Mahesh Jethanandani <mjethanandani@xxxxxxxxx>; yang-doctors@xxxxxxxx
Cc: ietf@xxxxxxxx; teas@xxxxxxxx; draft-ietf-teas-yang-te-topo.all@xxxxxxxx
Subject: Re: [Teas] Yangdoctors last call review of draft-ietf-teas-yang-te-topo-
08

Dear draft-ietf-teas-yang-te-topo authors,

Hi Xufeng,

OK, by tooling, I don't mean the pyang plugins that I have been
working on to convert between different types of models.  As you
aware, the TE YANG models can easily be converted to NMDA style since
I have already done it
(https://github.com/rgwilton/ietf-models-to-combined).

My comment actually relates to the fact the structure used by TE YANG
modules don't match any other YANG modules - they are using their own
unique style of structure.

This is an important issue to resolve.

Today, there are three common styles of modules:
(1) IETF style split config/state trees (e.g. ietf-interfaces).
(2) IETF NMDA style combined config/state trees (i.e. where all IETF
modules are heading to).
(3) OpenConfig style modules with the config/state containers
immediately above the config leaves.

Tooling is likely to be optimized to work with these model structures,
but the TE modules do not fit into any of the three styles above.
They are a yet another OpenConfig-like style, but that is different
enough that tooling that is designed to work with OpenConfig style
YANG modules would likely not work with the TE YANG modules.

Specifically, for clients that expecting to work with an OpenConfig
style YANG module, then if it knows the path to config leaf X, then
they would expect the applied config value to be available at the path
"../state/X".  But, this doesn't hold for the structure being used in
the TE YANG models.

I believe strongly that the models being produced by organizations
should be structures in a consistent way, hence why I think that the
published standard version of the TE YANG modules should immediately
align to the NMDA style.

Agreed.
Here is the OPS and RTG AD message:
https://www.ietf.org/mail-archive/web/netmod/current/msg18252.html

I understand that the I2RS topology YANG modules will be improved to the
NMDA style.

[Xufeng] The I2RS topology model is already NMDA style, but the NMDA models are not immediately implementable without NETCONF/RESTCONF protocol update. We will need the "-state" module for the I2RS topology model (and for all the top-level models) for now. 

Adding -state to the Appendix is a possibility.
"In all cases, the NMDA and non-NMDA modules SHOULD be published in the same document, with NMDA modules in the document main body and the non-NMDA modules in an Appendix."
See https://www.ietf.org/mail-archive/web/netmod/current/msg18252.html

While the NMDA is nice, the NMDA discussions have already significantly delayed publishing new models, and doing NMDA without staging will further delay the process.

This is not solving the important issue brought up by Rob Witlon:

My comment actually relates to the fact the structure used by TE YANG modules don't match any other YANG modules - they are using their own unique style of structure.

Regards, B.

Thanks,  Xufeng

Regards, Benoit

Thanks,
Rob


On 22/06/2017 04:16, Xufeng Liu wrote:

Hi Rob,

While the tooling is very nice to have, especially for writing new
models or converting models, we do not have to use it for every
model. It seems not necessary to use the tooling on this model for
now. We already know that we can manually convert it to NMDA style if
needed.

Thanks,
- Xufeng

-----Original Message-----
From: Robert Wilton [mailto:rwilton@xxxxxxxxx]
Sent: Wednesday, June 21, 2017 5:34 AM
To: Xufeng Liu <Xufeng_Liu@xxxxxxxxx>; Mahesh Jethanandani
<mjethanandani@xxxxxxxxx>; yang-doctors@xxxxxxxx
Cc: ietf@xxxxxxxx; teas@xxxxxxxx;
draft-ietf-teas-yang-te-topo.all@xxxxxxxx
Subject: Re: [Teas] Yangdoctors last call review of
draft-ietf-teas-yang-te-topo-
08

Hi Xufeng,

On 12/06/2017 22:28, Xufeng Liu wrote:

Hi Mahesh,

Thank you much for the review. We have submitted an updated draft

(https://tools.ietf.org/html/draft-ietf-teas-yang-te-topo-09) to
address these issues. More detailed explanations are put below
inline.

If the responses and updates are satisfactory, we are ready for the
last call.

Best regards,
- Xufeng

-----Original Message-----
From: Mahesh Jethanandani [mailto:mjethanandani@xxxxxxxxx]
Sent: Wednesday, May 24, 2017 11:44 AM
To: yang-doctors@xxxxxxxx
Cc: ietf@xxxxxxxx; teas@xxxxxxxx;
draft-ietf-teas-yang-te-topo.all@xxxxxxxx
Subject: Yangdoctors last call review of
draft-ietf-teas-yang-te-topo-08

Reviewer: Mahesh Jethanandani
Review result: Ready with Issues

Document reviewed: draft-ietf-teas-yang-te-topo-08

Status: Ready with Issues

I am not an expert in Traffic Engineering. This review is looking
at the draft from a YANG perspective. With that said, I have
marked it as “Ready

with Issues”

because of some of the points discussed below.

Summary:

This document defines a YANG data model for representing,
retrieving and manipulating TE Topologies. The model serves as a
base model that other technology specific TE Topology models can

augment.

Comments:

Almost all the containers in the model are presence containers. Is
there a reason why they have to be presence containers? Note, that
presence containers are containers whose existence itself
represents configuration data. What particular configuration data
is each container

representing in itself?

[Xufeng] Containers that use “presence” are:
    - Container “underlay”
      o  We have changed 13 occurrences of such containers to be
not

presence container.

    - Container “te” under augmentation
      o  To indicate that “TE” is enabled (configuration data)
      o  Also used to do augmentation. The “presence” statement can

prevent the mandatory child from affecting augmented base model.

    - /nw:networks/nw:network/nw:network-types/te-topology!
      o  A mechanism required by I2RS topology model to specify the

topology type.

It is difficult to co-relate the diagram with the model itself
because of different terms being used to define different parts of
the model.
There is “TE Topology Model” and then there is “Generic TE
Topology

Model”.

Are these one and the same models? If so, a common term for both
of them would be helpful.

[Xufeng] Yes. These two terms are the same. Figure 12, Figure 13,
and relevant

descriptions have been updated to make the document consistent.

There is extensive use of groupings in the document. However, not
all instances of groupings are used multiple number of times.
Where they are not being repeated, it would be better to move the
grouping directly where the uses statement resides. Case in point
the grouping

connectivity-label-restriction-list.

[Xufeng] We have removed the following groupings
      te-link-augment
      te-node-augment
      te-termination-point-augment
      te-topologies-augment
      te-topology-augment
      te-link-state-underlay-attributes
      te-node-state-derived-notification
      te-topology-type

The remaining groupings have been kept so that we can:
    - Share the groupings in this model
    - Prepare to be shared by a model augmenting this model
    - Prevent a grouping or configuration section from being too long
    - Improve readability

The split between config and state containers does not seem to
follow any particular pattern.

[Xufeng] The pattern is clear:
For each manageable entity (object), there is a config container
and state

container. The configurable properties go into the config container
and state properties go into the state container. Such objects are
identified by a list item or a presence container so that the
“create”, “delete”, and “modify”
operations
can be performed on them. The non-presence containers do not
represent configuration data so they do not introduce such objects.

It is neither a top level split, as is the case with existing IETF
models,

[Xufeng] We could not do top level split because the base I2RS
network

topology model
(https://tools.ietf.org/html/draft-ietf-i2rs-yang-network-topo-
12) that we augment does not have the top-level split (for its own
reasons).

nor do they follow the OpenConfig style of splitting config and
state under each relevant leaf,

[Xufeng] The pattern is consistent with this style in principle,
with some

adjustments to fit to our multiple levels of hierarchy.
This is effectively a new forth style of YANG models that is not
consistent with any of the three existing styles, i.e.:
   - Current IETF config/state split model style
   - NMDA combined config/state tree
   - OpenConfig split config/state containers immediately above the
config true leaves.

Tooling that it designed to work with OpenConfig models will need
customization to work with these TE models because the config/state
containers will not be where the tooling expects them to be.

Thanks,
Rob

.