Hi, Xufeng Liu <[email protected]> wrote: > Hi Martin, > > I know that this is a previously discussed solution, but I still have > some issues with it: > > 1) In the case of the interfaces model, for each referenced > interface-state, the operator needs to configure an interface object > with the same interface name. However, in the case of the topology > model, if we need to reference an underlay link-state, we will need to > configure a topology (called network), a source node, a destination > node, a source termination-point, a destination termination-point, > which are five objects without including other consequent mandatory > objects.
You don't have to configure a "dummy" object if you are not using leafrefs to config. And the number of nodes that you need for unique identification should be totally independent. > 2) This approach stretches the definition of "system generated" > non-configurable objects. The system generated objects mentioned in 1) > are designed to be not configurable. Configuring them may result > un-desirable consequences. See above. > 3) In general, this approach will not work if the referenced schema > leaf is marked as "config false". In such a case, we cannot configure > such a referenced leaf since it is not configurable. I don't understand this comment. BTW, maybe further discussion about a new solution should be on the i2rs ML only? /martin > > Thanks, > > - Xufeng > > > -----Original Message----- > > From: Martin Bjorklund [mailto:[email protected]] > > Sent: Thursday, February 2, 2017 2:33 AM > > To: [email protected] > > Cc: [email protected]; [email protected]; draft-ietf-i2rs-yang-l3- > > [email protected]; [email protected]; [email protected] > > Subject: Re: [i2rs] Kathleen Moriarty's No Objection on > > draft-ietf-i2rs-yang-l3- > > topology-08: (with COMMENT) > > > > Alexander Clemm <[email protected]> wrote: > > > We are working this separately and will articulate the different > > > options and their respective issues. > > > > > > The fundamental issue is still the fact that you may have dependencies > > > in overlay topologies on underlay topologies that are discovered and > > > represent “state”, and that in fact your underlay may be either. > > > > > > RFC 7223, as far as I can tell, sidesteps this issue. It does define > > > a type “interface-ref” with a path to reference a configured > > > interface, and it does define a type “interface-state-ref” to > > > reference an operationally present interface. However, > > > interface-state-ref is used only in read-only objects, whereas (to put > > > the analogy) it is needed for configurable objects as well. Likewise, > > > there are two types; really we need a union which would allow either > > > (or a leafref with alternate paths, which is not supported). While > > > there are some analogies with a preprovisioning scenario, there are > > > also differences. > > > > When people refer to the "pre-provisioning approach" in RFC 7223, it > > is not the > > "interface-ref" or "interface-state-ref" they refer to. > > > > The pre-provisioning mechanism in RFC 7223 says that when the device > > initializes a detected interface, it will check the configuration to > > see if there is > > config available for an interface with the same name as the newly > > detected one. > > If so, that config is used. > > > > I think the idea was to use something similar here. E.g., allow a > > configured > > overlay to refer to a discovered underlay by name. In YANG, this can > > be done > > with a node with the same type; or possibly with a leafref to the > > state data with > > "require-instance false". > > > > This design allows an overlay to be configured for an existing > > detected underlay. > > Let's say the device reboots and starts to rebuild its topologies. > > During some > > period of time, the configured overlay still exist in the config, but > > not in the state, > > since the underlay is not yet available. Once it becomes instantiated > > in the state, > > the overlay is also instantiated in the state. (This assumes that the > > system- > > generated topology names do not change). > > > > > > /martin > > > > > > > > > > > > Anyway, Xufeng, Kent, Pavan and I are having offline discussions and > > > will come back with further elaboration on this. > > > > > > --- Alex > > > > > > From: i2rs [mailto:[email protected]] On Behalf Of Alia Atlas > > > Sent: Wednesday, February 01, 2017 1:14 PM > > > To: Lou Berger <[email protected]> > > > Cc: [email protected]; [email protected]; > > > [email protected] > > > Subject: Re: [i2rs] Kathleen Moriarty's No Objection on > > > draft-ietf-i2rs-yang-l3-topology-08: (with COMMENT) > > > > > > On Wed, Feb 1, 2017 at 3:56 PM, Lou Berger > > > <[email protected]<mailto:[email protected]>> wrote: > > > > > > > > > On 2/1/2017 2:32 PM, Juergen Schoenwaelder wrote: > > > > On Wed, Feb 01, 2017 at 01:52:25PM -0500, Lou Berger wrote: > > > >> Juergen, > > > >> > > > >> What precludes treating such dependencies in the same way > > > >> per-provisioning is handled by RFC7223? > > > >> > > > > This is fine. But having direct dependencies, e.g., leafrefs from > > > > config true leafs to config false leafs, is not. > > > > > > > > /js > > > > > > > > > > Okay, then we're on the same page -- I think some may have missed the > > > possibility of handling references to dynamic topology information in > > > config using a 'pre-provisioning' approach. > > > > > > I would be happy to see Alex, Xufeng, Kent & Pavan articulate what > > > this would look like and how it would work for the base topology > > > model, so that the WG can consider all potentially viable options. > > > I'm not certain how it would function for the recursive nature and it > > > does presume the separate /config and /oper-state trees in the > > > data-model that were a concern (though certainly the current > > > recommended approach for YANG models). > > > > > > Regards, > > > Alia _______________________________________________ i2rs mailing list [email protected] https://www.ietf.org/mailman/listinfo/i2rs
