Re: [netmod] Proposal for minimalist full NMDA support in schema mount

[Robert Wilton <[e-mail]>]

Hi Chris,

The future issues that I'm predicting is not tied to per datastore 
schema, but because the SM YANG module is tied to a version of YL that 
is being deprecated and replaced by a similar but different structure in 
YLbis.

If all mounted modules are also available at top level modules, and 
hence will be reported in YLbis regardless, then I agree that it 
probably doesn't matter so much.

But if there are YANG modules that are only available at mounted 
locations (and not at the top level) then they would not be listed in 
the top level YANG library and I think that you will hit the issues that 
I describe below.  E.g. you wouldn't be able to retrieve the SemVer 
information for those mounted modules, etc.

Thanks,
Rob


On 26/02/2018 11:52, Christian Hopps wrote:
>
> Hi Rob,
>
> You do realize that no-one trying to actually deploy and run networks 
> cares about live-discovery of different schema per datastore for the 
> same mount point right? Like 99.999% of the clients know where things 
> are supposed to reside and expect them to be there. At most (although 
> still not common) they may want to know what modules are supported 
> under a mount point. What your talking about is a severe edge case 
> that apparently has achieved extreme importance in a very small group 
> of people's views.
>
> Thanks,
> Chris.
>
> Robert Wilton <[e-mail]> writes:
>
>> Hi Chris,
>>
>> I've got no desire or intent to try and slow down the NI and LNE 
>> drafts, or any
>> that depend on them. I actually agree that this is critically 
>> important that
>> IETF gets modules standardized/finished so that everyone can use them.
>>
>> However, ...
>>
>> YLbis has quite a different structure to YL. The main part of this 
>> change was
>> to support NMDA, the other part of this change was to better support 
>> things like
>> SM, or YANG packages.
>>
>> I don't think that there is a clean, backwards compatible, way to go 
>> from the
>> YANG module in SM -08 to one that is going to work well with YLbis 
>> and other
>> YLbis extensions/augmentations that seem to be coming down the line. 
>> Given what
>> we know now, I believe that the correct medium/long term structure 
>> for the SM
>> YANG module, taking YLbis into account, is the one proposed in pre09, 
>> because it
>> directly augments the YLbis structure, and hence any future 
>> augmentations to
>> YLbis should automatically extend to SM mounted schema as well.
>>
>> I think that the likely future technical issues with the -08 module 
>> will be:
>> - supporting NMDA in a clean consistent way
>> - adding in support for SemVer
>> - additional capability reporting as an augmentation to YANG library
>>
>> So, if -08 proceeds as is, then it seems to me like one of three 
>> things will
>> need to happen:
>> 1) Their will need to be a non backwards compatible update to the SM 
>> model that
>> is the same/similar to pre09.
>> 2) YLbis and SM diverge, stuff that augments YLbis doesn't work for 
>> explicitly
>> mounted schema.
>> 3) We accrue technical debt, implementations need to support two YL 
>> module
>> structures, the one in SM and the one in YLbis, and future extensions 
>> need to
>> augment both the SM structure and YLbis structure.
>>
>> I don't like the idea of (2) or (3), but I don't know if others will 
>> find (1)
>> acceptable.
>>
>> But I do agree that we are just going round in circles on this:
>> - Using the pre09 structure is not acceptable to some folks
>> - Publishing a draft with both -08 and pre09 structure is liked by 
>> even less
>> folks.
>>
>> Perhaps publishing -08 is the only option. My hope is that the WG 
>> will support
>> somebody subsequently doing solution (1), otherwise it seems like a 
>> missed
>> opportunity to get this right.
>>
>> Thanks,
>> Rob
>>
>>
>> On 24/02/2018 13:54, Christian Hopps wrote:
>>> My position,
>>>
>>> It may be the case that there's even a better cleaner solution; 
>>> however, it's
>>> simply too late for major modifications to this work that don't 
>>> actually
>>> address functional failures. The draft as proposed works for the 
>>> people who
>>> need to get work done.
>>>
>>> We have multiple pending RFCs - MISREF on this document. These RFCs 
>>> would have
>>> to be pulled from the RFC EDITOR queue, and reworked to be compliant 
>>> again,
>>> and this very well could lead to discovering issues with your new 
>>> proposal.
>>> Any new issues discovered in either the pending RFCs *or* in the new 
>>> solution
>>> would then need to be worked out and fixed. Please recall that this 
>>> actually
>>> occurred on the first round (i.e., doing the examples led to 
>>> discovering
>>> problems with the drafts), so it's not unreasonable at all to assume 
>>> this
>>> would happen again.
>>>
>>> Look this just isn't a simple change your proposing. It involves a 
>>> large
>>> upheaval, killing the pending RFC status on multiple documents that the
>>> industry is waiting on. Please see this.
>>>
>>> Thanks,
>>> Chris.
>>>
>>>
>>> Martin Bjorklund <[e-mail]> writes:
>>>
>>>> Hi,
>>>>
>>>> Robert Wilton <[e-mail]> wrote:
>>>>> Hi Lou,
>>>>>
>>>>> I think that this solution is inferior to the model presented in
>>>>> pre-09.
>>>>
>>>> I agree. Servers that are NMDA-compliant, or implements YANG Library
>>>> bis will have to present schemas in two different structures,
>>>> depending on where the schema is used, and clients will have to code
>>>> for both. With the solution in pre-09, there is just one structure.
>>>> A single structure also has other benefits (apart from being simpler),
>>>> e.g., if we augment it with the meta data that has been discussed
>>>> recently, we can augment a single structure.
>>>
>>>> /martin
>>>>
>>>>
>>>>
>>>>> I would prefer that we publish pre09 instead, potentially including
>>>>> the -08 model in the appendix if that helps progress the document 
>>>>> in a
>>>>> more expedient fashion.
>>>>>
>>>>> Thanks,
>>>>> Rob
>>>>>
>>>>>
>>>>> On 22/02/2018 16:18, Lou Berger wrote:
>>>>> > Hi,
>>>>> >
>>>>> > (I have a bunch of different roles WRT this work. This mail is 
>>>>> being
>>>>> > sent as an individual - as chair, I fully support the previous 
>>>>> chair
>>>>> > statements on this draft.)
>>>>> >
>>>>> > Chris and I have come up with a proposal on how to provide full 
>>>>> NMDA
>>>>> > as part the existing schema-mount module. Our motivation was to
>>>>> > enable full NMDA support with *minimal* change to the model and
>>>>> > disruption to the LC'ed work. The key NMDA limitation, with -08, 
>>>>> that
>>>>> > we are aiming to address is the ability to support different 
>>>>> mounted
>>>>> > schema in different datastores for non-inline mount points. (See 
>>>>> more
>>>>> > detailed description below if interested full nuances of 
>>>>> limitations
>>>>> > of -08)
>>>>> >
>>>>> > What we came up with was to simply add a (leaf)list to identify in
>>>>> > which datastores a
>>>>> > schema-mount schema is valid/present. This is somewhat similar to
>>>>> > YL-bis schema/module-set. Specifically we're proposing (see 
>>>>> below for
>>>>> > full tree below):
>>>>> >
>>>>> > +--ro schema* [name]
>>>>> > +--ro name string
>>>>> > ADD +--ro datastore* ds:datastore-ref {revised-datastores}
>>>>> >
>>>>> > This approach has the advantages of supporting different mounted
>>>>> > schema in different DSes, working with both NMDA and non-NMDA
>>>>> > implementations, supporting all of the extensively discussed 
>>>>> features
>>>>> > of schema mount (including recursive mounts), and having 
>>>>> minor/scoped
>>>>> > impact on all dependent work. The main downside is that it isn't 
>>>>> the
>>>>> > most optimal/compact solution possible if we were to base this 
>>>>> work on
>>>>> > YL-bis/pre09 draft. Of course -08 isn't necessarily optimal from 
>>>>> all
>>>>> > perspectives, but it is what was agreed to as sufficient by 
>>>>> those who
>>>>> > contribute to the WG discussion.
>>>>> >
>>>>> > In short, we see this as a solution to addresses the raised last 
>>>>> call
>>>>> > issue with the minimal impact on -08 and dependent work -- which is
>>>>> > what is appropriate given where we are in the process.
>>>>> >
>>>>> > So our/my question really is:
>>>>> >
>>>>> > Is this a solution that you/all can live with?
>>>>> >
>>>>> > Note: optimization, design preference and perfect alignment with 
>>>>> use
>>>>> > or YL-bis are not part of our question as we both don't think 
>>>>> that is
>>>>> > the right question given where we are in the WG process.
>>>>> >
>>>>> > Lou (with ideas developed with Chris, and chair hat off)
>>>>> >
>>>>> > ======
>>>>> > Details -- for those who want
>>>>> > ======
>>>>> > As background, my understanding/view is that the -08 version of the
>>>>> > both NMDA and non-NMDA supporting implementations, but there are
>>>>> > limitations in its NMDA applicability. Used with Yang Library,
>>>>> > [rfc7895], only non-NMDA implementations can be supported. When 
>>>>> used
>>>>> > with the revised Yang Library defined in
>>>>> > [I.D.ietf-netconf-rfc7895bis-], NMDA implementations can be
>>>>> > supported with certain limitations. Specifically, this document
>>>>> > requires use of the now deprecated module-list grouping, and the 
>>>>> same
>>>>> > schema represented in schema list of the Schema Mount module 
>>>>> MUST be
>>>>> > used in all datastores. Inline type mount points, which don't 
>>>>> use the
>>>>> > schema list, can support different schema in different data stores
>>>>> > not by instantiating the [I.D.ietf-netconf-rfc7895bis-] version of
>>>>> > YANG library under the inline mount point.
>>>>> >
>>>>> > module: ietf-yang-schema-mount
>>>>> > +--ro schema-mounts
>>>>> > +--ro namespace* [prefix]
>>>>> > | +--ro prefix yang:yang-identifier
>>>>> > | +--ro uri? inet:uri
>>>>> > +--ro mount-point* [module name]
>>>>> > | +--ro module yang:yang-identifier
>>>>> > | +--ro name yang:yang-identifier
>>>>> > | +--ro config? boolean
>>>>> > | +--ro (schema-ref)?
>>>>> > | +--:(inline)
>>>>> > | | +--ro inline? empty
>>>>> > | +--:(use-schema)
>>>>> > | +--ro use-schema* [name]
>>>>> > | +--ro name
>>>>> > | | -> /schema-mounts/schema/name
>>>>> > | +--ro parent-reference* yang:xpath1.0
>>>>> > +--ro schema* [name]
>>>>> > +--ro name string
>>>>> > ADD +--ro datastore* ds:datastore-ref {revised-datastores}
>>>>> > +--ro module* [name revision]
>>>>> > | +--ro name yang:yang-identifier
>>>>> > | +--ro revision union
>>>>> > | +--ro schema? inet:uri
>>>>> > | +--ro namespace inet:uri
>>>>> > | +--ro feature* yang:yang-identifier
>>>>> > | +--ro deviation* [name revision]
>>>>> > | | +--ro name yang:yang-identifier
>>>>> > | | +--ro revision union
>>>>> > | +--ro conformance-type enumeration
>>>>> > | +--ro submodule* [name revision]
>>>>> > | +--ro name yang:yang-identifier
>>>>> > | +--ro revision union
>>>>> > | +--ro schema? inet:uri
>>>>> > +--ro mount-point* [module name]
>>>>> > +--ro module yang:yang-identifier
>>>>> > +--ro name yang:yang-identifier
>>>>> > +--ro config? boolean
>>>>> > +--ro (schema-ref)?
>>>>> > +--:(inline)
>>>>> > | +--ro inline? empty
>>>>> > +--:(use-schema)
>>>>> > +--ro use-schema* [name]
>>>>> > +--ro name
>>>>> > | -> /schema-mounts/schema/name
>>>>> > +--ro parent-reference* yang:xpath1.0
>>>>> >
>>>>> > We would expect that the revised-datastores feature would be used
>>>>> > (perhaps required) for any implementation that supports
>>>>> > ietf-datastores
>>>>> > and yl-bis.
>>>>> >
>>>>> >
>>>>> >
>>>>> > _______________________________________________
>>>>> > netmod mailing list
>>>>> > [e-mail]
>>>>> > https://www.ietf.org/mailman/listinfo/netmod
>>>>>
>>>>> _______________________________________________
>>>>> netmod mailing list
>>>>> [e-mail]
>>>>> https://www.ietf.org/mailman/listinfo/netmod
>>>>
>>>> _______________________________________________
>>>> netmod mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/netmod
>>>
>>> .
>>>
>
> .
>