Re: [netmod] [Netconf] Alternative YANG library structure for 7895bis

[Vladimir Vassilev <[e-mail]>]

On 11/15/2017 01:20 PM, Martin Bjorklund wrote:
> Vladimir Vassilev <[e-mail]> wrote:
>
> [...]
>
>> It is clear that datastores with non-identical models can not be
>> supported with yang-library:1.0. However for the many usecases that do
>> not require the complexity of having different datastore models
>> (variation of the set of modules and the relevant deviations e.g. more
>> complex datastore centric conformance model) one can implement NMDA
>> with yang-library:1.0.
>>
>> My initial proposal was a change to draft-ietf-netconf-nmda-netconf-01
>> sec. 2.4. 'YANG Library Capability' to allow that usecase:
>>
>> OLD:
>> Support for NMDA requires the server to implement at least revision
>> 201X-XX-XX of the "ietf-yang-library"
>> ...
>> NEW:
>> Support for NMDA with datastores with non-identical models requires
>> the server to implement at least revision 201X-XX-XX of the
>> "ietf-yang-library"
> But that would imply that we keep two versions of ietf-yang-library
> around, and "current".
>
> Also note that the new proposed version is very similar in structure
> to the old yang library.  And if the server implements "identicial
> models", then the "not-implemented-in" will be empty, and so the
> contents of yang-library 1.1 is more or less the same as yang-library
> 1.0.
I aggree with what you say. My point is that the upgrade to 
yang-library:1.1 can come later then the upgrade to NMDA ietf-interfaces 
<get-data> etc. I will implement yang-library:1.1 as soon as I find a 
single usecase that justifies the work needed. At the moment I only have 
motivation to implement <get-data> and migrate to the NMDA modules and 
the only issue with that plan is the text in 
draft-ietf-netconf-nmda-netconf-01 sec. 2.4. 'YANG Library Capability'.

Following my gut feeling I would rather wait to see if yang-library:1.2 
is not coming up before I am done with the process of migrating to NMDA 
with identical datastore models. A reasonable concern taking into 
account yang-library:1.0 is about to be obsoleted so soon after the 
publication of rfc7895 because of a feature (introduction of 
datastore-centric conformance model) that has no value for me at the 
moment unlike the feature (unification of configuration and state trees) 
introduced with NMDA.

That said there might be a counter argument that the confusion of 
allowing NMDA implementations with both yang-library:1.0 and 
yang-library:1.1 is worse then forcing all NMDA applications to 
implement only yang-library:1.1. If this is so that should be stated 
clearly.

Vladimir

> /martin