Re: [netmod] some comments on revised-datastores-01

[Martin Bjorklund <[e-mail]>]

Andy Bierman <[e-mail]> wrote:
> Hi,
> 
> I like this draft -- even more than I like datastores.
> Looks like some thought went into the terminology section.
> 
> One minor concern:
> 
> sec. 4.1:
> 
>    On a traditional NETCONF implementation, <running> and <intended> are
>    always the same.
> 
> 
> The intended datastore is completely proprietary.
> Is that part of the architecture or are there any plans for the standards
> to have some purpose for the intended datastore?

The idea was to have a defined place for the "intended configuration"
in the common architecture, even if there are no current standard
mechanism that affects it.  If such mechanisms are added, they can be
defined in terms of the datastores in this architecture.

> I want to support 2 datastores that don't fit your descriptions very well:
> 
> factory:  read-only config representing the values that a factory reset
> would use.  Allows copy-config to support a factory config reload.
> 
> library-config: needed as a bootstrap datastore with the module/bundle
> configuration. Like the ietf-yang-library, but the config does not exactly
> match
> the structure of the YANG library.
> 
> Can vendors add datastores to a server?
> Or is this IANA-registered?  And/or hard-wired in RFCs?

The draft has a section about what is needed to define dynamic
datastores.  IMO this should be generalized to any datastore.  The
datastore names are identities, so vendors can define their own
datastores.  Currently there is no explicit mechanism for a server to
advertise which datastores is supports, other that the advertisment of
features in "ietf-datastore".  Maybe we should add an explicit list of
supported datastores (but this will be protocol-dependent, since some
protocols might not expose all datastores).


/martin