[netmod] Re: comments on system-config-08 draft

[Andy Bierman <[e-mail]>]

On Thu, Aug 22, 2024 at 8:09 PM maqiufang (A) <[e-mail]> wrote:

> Hi, Andy,
>
>
>
>
>
> Changing the running config so it is split into 2 datastores makes
> operations more complicated.
>
> It doesn't actually work since YANG is hierarchical and has
> cross-references.
>
>
>
> IMO the only improvement needed is to add metadata to <running> so a
> client can
>
> better understand the system config and make edit requests that will not
> fail.
>
>
>
> Most deployment (90%?) is non-NMDA and it will probably stay that way.
>
> The developer focus is on data model deployment, not redoing the
> foundation. IMO people want YANG to
>
> be simpler and faster. I don't see how splitting config up into 2
> datastores helps.
>
>
>
> I am not sure I agree this draft is “changing the running config so it is
> split into 2 datastores”.
>
>
>
> I think it is already the case that NMDA never puts system configuration
> into <running>, it’s only in <operational>. So has NMDA already made NBC
> changes to the behavior of non-NMDA servers? If you know any existing
> standards that say system configuration is defined in <running>, I would
> appreciate that if you could help navigate, as I’ve always thought putting
> system configuration into <running> is just one of the existing various
> proprietary implementations, our implementation used to be this way, but
> not now, after some customers complained that they don’t want to see some
> system config magically appeared when they write some config into <running>
> and then read back, and there are other reasons like performance.
>
>
>

There is no such thing as "system config" defined in YANG.
It was never mentioned as an operator requirement in RFC 3535 (but separate
config and state is a requirement).

The "split config" is a result of suppressing system config (whatever that
is) when <running> is retrieved.

IMO this is no different than "with-defaults".  A client can retrieve
<running> with the defaults or without them.
A similar filter for "with-system" or "with-unused-system" would be simpler
than adding a new datastore to NMDA.
The complex interactions between datastores and operations on datastores
(like copy-config) are non-trivial.



>
> Best Regards,
>
> Qiufang
>
>
>
>
>

Andy