Re: [Netconf] Draft Charter Proposal for NETCONF WG

[Robert Wilton <[e-mail]>]

Hi Tom,


On 09/03/2017 12:12, t.petch wrote:
> ----- Original Message -----
> From: "Robert Wilton" <[e-mail]>
> Sent: Friday, March 03, 2017 5:41 PM
>
>> On 03/03/2017 17:18, t.petch wrote:
>>> ----- Original Message -----
>>> From: "Mehmet Ersue" <[e-mail]>
>>> Sent: Friday, March 03, 2017 2:33 PM
>>>
>>>>> Back to your question, it seems obvious to me that YANG and the
> XML
>>>> encoding rules naturally belong to NETMOD, the 'NETCONF protocol
>>> details
>>>> that NETCONF
>>>>> did not define' naturally belong to NETCONF.
>>>> Basically it is our aim to make the YANG language specification
>>> generally
>>>> applicable to all protocols and to put protocol-specific details
> into
>>> the
>>>> protocol specifications.
>>> See my response to Juergen; I agree with you but I define XML as not
>>> being a protocol and so XML would remain; and I think that YANG will
>>> have to say something about operations on the data it defines, just
> that
>>> they are defined as an abstract 'create', 'delete' etc and not as
> the
>>> set that NETCONF currently offers.
>> FWIW, this is the block
>> "      Common protocol abstraction
>> (that all YANG protocols should conform to). "
>>
>> That I was referring to in the diagram that I gave previously,
> although
>> I was suggesting that should belong in NETCONF WG rather than in YANG.
> Robert
>
> It has taken me a while to work out what you mean but now I have, I
> disagree!
>
> You seem to place data(stores) at the heart of things, the root from
> which all else flows.  I think that this can work with application
> software in a stable, secure, delay-less environment where nothing ever
> goes wrong (a mobile phone app perhaps!).
Yes, I definitely want to place accurate and meaningful data at the 
heart of it.

As I see it:
- YANG is the schema for that data.
- Datastores are really just views on data, bound to the schema and the 
data life cycle
- Protocols are mechanism to access and modify that data, aided by the 
schema & datastores.

I really hope that the solution that we are constructing will work well 
for systems that have real delays, unreliable communications, and 
potentially buggy software.  Certainly, that is my goal ... Some of this 
will need protocol assistance.

>
> Network management is different;  the failing network is both the
> subject under consideration and an integral part of the solution.  The
> operator has to use the failing network to find out what is failing and
> what might be done about it and then use the failing network to convey
> changes to the failing component of the network.  SNMP recognised this
> but I am not sure the NETCONF/YANG do - after all, their focus is on
> configuration, before things start going wrong.
I don't know the history, but my perception is that NETCONF/YANG was 
focused on config because that is the part of SNMP that failed to gain 
traction in the industry.  As NETCONF/YANG gains traction, it seems 
reasonable to want to fix the operational state aspect of it that seems 
somewhat incomplete today.

>
> I see revised-datastores as an attempt to fix this but one that will
> fail, in the sense that it cannot go far enough; what may be needed is a
> paradigm shift in Computer Science so a server can say that the model it
> has been given cannot reflect reality but here is a better one freshly
> created for the client to use!

So, I think that that issue that you are raising here is that a device 
might not be able to accurately populate the schema being used for the 
operational state datastore.  The latest (unpublished) datastores draft 
states that even all values (including defaults) are returned in the 
operational state datastore.  I.e. everything is explicit, meaning that 
if a device cannot return the correct value for a node then it has the 
choice of returning no value at all.

In terms of dynamic schema, devices can already define their own custom 
schema and augmentations that can carry any extra vendor/device specific 
data that cannot be readily mapped back into the standard schema.  The 
problem here is that these schema are non standard (between vendors 
and/or devices) and hence much harder for automated clients to use.  I 
think that there is also a scope question of these additional vendor 
schema, given that a lot of the data is likely to be verbose, possibly 
expensive to obtain, and perhaps more diagnostics orientated.

>
> I don't see that happening just yet so revised-datastores will have to
> do but I think it wrong to make that central - it will not be close
> enough to reality.
It will be central in the sense that YANG models will either be built 
assuming that it exists, or that it doesn't.  I don't think that you can 
really have well constructed, fully useful, YANG models where the 
operational state datastore is optional.

I don't think that the datastores draft is going to be a silver bullet 
that solves all problems, but if the solution gains traction then I do 
think that it will give a step improvement to making it easier to manage 
network devices in an automated and robust way.

Regards,
Rob


>
> Tom Petch
>
>> Rob
>>
>>> Tom Petch
>>>
>>>> Mehmet
>>>>
>>>>> -----Original Message-----
>>>>> From: Juergen Schoenwaelder [mailto:j.schoenwaelder@jacobs-
>>>>> university.de]
>>>>> Sent: Friday, March 3, 2017 2:35 PM
>>> <snip>
>>>
>>> _______________________________________________
>>> Netconf mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/netconf
>>> .
>>>
> .
>