Re: [mpls] [netmod] [Technical Errata Reported] RFC8349 (6251)

[tom petch <[e-mail]>]

From: Acee Lindem (acee) <[e-mail]>
Sent: 14 August 2020 16:27

Hi Tom,
The problem is that RFC 8349 defines an AF-agnostic model for a RIB. The draft augments the base model with optional MPLS attributes that would ostensibly be applicable to all AFs. What most people miss, is that it also hijacks the base model for the MPLS address-family and subverts the AF-agnostic model. This shouldn't and can't be fixed with an Errata for RFC 8349.

Acee

We agree that this Erratum should be rejected.  When I first read the I-D, I queried that brief statement in s.2.1 about a new address family and eventually got pointed to RFC8349 s.3.  I did then and do now wonder about the meaning of address family and found its use scattered across the Routing Area with only BGP4 making any coherent, to me, use of the term; I found the use in RFC8349 a little surprising.  I tagged that sentence in the I-D as one I did not understand and still do not although Tarek's comment about routing with no IP gives me a clue.  So I see IP routing, an IP address family, as something that could have an MPLS label augmented to it; while an MPLS address family is IP free and something I am not familiar with.

I noted the lack of conditionals in the early versions of this I-D and I think it was my comments that got them added but I noted that they were not based on address family, as the IPv4 unicast augmentations of RFC8349 are, but on mpls being enabled on the box.  So I have always seen this I-D as different as adding MPLS to all the existing address families, the existing RIBs.

So I agree, as it stands the I-D augments everything but I had assumed that that is exactly what it intended to do and so does.  I did not see that as subverting the model but what it comes down to is that when we have IP with MPLS labels, or MPLS with no IP, then how many RIBs do we want and what to we want the RIBS to look like.

I would love to hear Lada's view on this as I am aware of how much research he did before the production of RFC8022 and wonder if he came across an implementation which illuminates this issue.

Tom Petch.

Thanks,
Acee
On 8/14/20, 7:45 AM, "tom petch" <[e-mail]> wrote:

    From: Acee Lindem (acee) <[e-mail]>
    Sent: 11 August 2020 14:08

    Hi Tom, Draft Authors,

    The draft could easily be fixed. You just need to:

      1. Expand on the single sentence in section 2.1 on the need for non-IP MPLS routes. Given that the draft wasn't modeled correctly, this wasn't apparent to most of the reviewers.
      2. Add an MPLS AF only augmentation (enforced via a when statement) to each route for the MPLS AF specific destination-prefix or destination-address.
      3. Limit the current local-label augmentation to non-MPLS AFs.
      4. Limit the active-route augmentation to AF MPLS and change the input to destination-address.

    <tp>
    'easily'  mmm... I have been getting my mind around this and see what Acee means and agree.  The I-D specifies an MPLS address family but then does not create the RIB entries for it in the way that e.g. IPv4 unicast does in  RFC8349.  Easy may be but quite a change  Another option, again a significant change, would be to cater only for IP address family with MPLS and drop the MPLS address family.  Either way, whatever the authors decide, I cannot personally see this falling within the scope or an Erratum or for the IETF Last Call to complete in a week's time.

    I have dropped the RFC Editor and netconf-chairs pro tem

    Tom Petch

    Thanks,
    Acee

    On 8/11/20, 6:10 AM, "tom petch" <[e-mail]> wrote:

        From: Acee Lindem (acee) <[e-mail]>
        Sent: 11 August 2020 10:47

        Hi Tom,
        I fully understood your original comment. There are other problems with this model. See inline.

        On 8/11/20, 4:59 AM, "tom petch" <[e-mail]> wrote:

            Tarek

            Picking up on an earlier point,

            ________________________________________
            From: Tarek Saad <[e-mail]>
            Sent: 10 August 2020 21:23

            Hi Acee,

            The existing RPC is used to query (defined AFIs=IPv4/IPv6) RIB to return the matching active route identified by a "destination address".
            The MPLS module is trying to reuse this RPC so to query the MPLS RIB to return the matching active route identified by a "local label".
            The RPC defined in RFC 8349 readily accepts MPLS AFI in it (/rt:routing/rt:ribs/rt:rib/rt:active-route) - unless we augment and suppress it with a "when check".
            IMO, it is reusable as-is but the text below is limiting the leaf name that identifies an entry in RIB to "destination-address" only - in MPLS RIB the entry leaf name that identifies an entry is "local-label".

        It is not reusable as is since the augmentation RPC augmentation must have a when statement restricting it to AF MPLS. Also, local-label is a leaf which is applicable to all address families. It cannot be the AF MPLS destination-prefix. This augmentation is missing.

        <tp>
        I am probably getting out of my depth here,  On 1may20 I raised the issue of why the 'MUST' in the description in RFC8349 was not enforced in the YANG and 5may20 Martin explained that there is a rule in the YANG ABNF of input-stmt that makes the obvious impossible:-(  You are raising more profound issues about the RIB that I had not perceived when I reviewed mpls-base-yang for which I, and I hope everyone else, will be grateful.

        If this mpls I-D is to proceed in the immediate future, it looks like the action may have to be deferred for future study.

        More generally, I think that the interaction of forward by address and forward by label is challenging.  When first I looked at the MPLS I-D I was surprised at the way RFC8349 was augmented.  I had not seen MPLS as an alternative to IPv4 or IPv6 or ... in the way in which the RFC is used although the RFC does state that it can be; rather, to me, labels are a different animal, but I assumed that everyone knew what they were doing.

        Tom Petch


        Thanks,
        Acee


            <tp>
            There should be a 'when' check to enforce the 'MUST' but the rules of YANG do not allow it in this structure.  I raised this on the NETMOD list at the time of WGLC and Martin pointed me to a rule in the ABNF which prohibits such a check.  He also said that the rule was not needed and would be a candidate to remove when YANG is revised.

            Hence I have always thought of this MUST in the documentation as a constraint that must be enforced in the YANG

            Tom Petch
                    >            action active-route {
                    >              description
                    >                "Return the active RIB route that is used for the
                    >                 destination address.
                    >
                    >                 Address-family-specific modules MUST augment input
                    >                 parameters with a leaf named 'destination-address'.";

            Regards,
            Tarek

            On 8/10/20, 3:27 PM, "Acee Lindem (acee)" <[e-mail]> wrote:

                [External Email. Be cautious of content]


                All (Speaking as an author of RFC 8349),
                I just looked at this in more detail and I don't think the ietf-mpls.yang model should be augmenting the /rt:routing/rt:ribs/rt:rib/rt:active-route RPC. The intent of the RPC is to return the address-family specific active-route corresponding to the destination-address. This model attempts to overload this RPC with a different action all together - returning a route that has the local-label as an optional attribute. I'd reject the Errata and believe the augmentation should be removed from ietf-mpl.yang. Whether it is replaced with a different one is up to the co-authors of ietf-mpls.yang.
                Thanks,
                Acee

                On 8/10/20, 2:29 PM, "Rob Wilton (rwilton)" <[e-mail]> wrote:

                    [Resend to hopefully pass recipient limit filter]

                    Hi Tom,

                    I would be interested to hear from the original authors.

                    My impression is that this is a technically reasonable change, but I don't think that an erratum can create a new revision of a YANG module.

                    If this erratum was processed as "Hold for document update" then would that be sufficient to do the right thing in the MPLS YANG module?

                    Regards,
                    Rob


                    > -----Original Message-----
                    > From: tom petch <[e-mail]>
                    > Sent: 10 August 2020 17:32
                    > To: RFC Errata System <[e-mail]>; [e-mail]; Acee
                    > Lindem (acee) <[e-mail]>; [e-mail]; [e-mail];
                    > Rob Wilton (rwilton) <[e-mail]>; [e-mail];
                    > [e-mail]; [e-mail]
                    > Cc: [e-mail]; [e-mail]
                    > Subject: Re: [netmod] [Technical Errata Reported] RFC8349 (6251)
                    >
                    > From: netmod <[e-mail]> on behalf of RFC Errata System
                    > <[e-mail]>
                    > Sent: 07 August 2020 16:45
                    >
                    > <tp>
                    > This is the erratum of whose arrival I speculated on this list on June
                    > 16th.
                    >
                    > There is a degree of urgency about it.  The I-D in question is mpls-base-
                    > yang, currently in IETF Last Call, which is a Normative dependency of bfd-
                    > yang which is a Normative dependency for a small mountain of I-D which
                    > have been waiting a year or so (e.g.  ospf-yang).
                    >
                    > I suspect that the technically perfect solution would involve a YANG
                    > union, choice or some such structure but as I said in my Last Call comment
                    > I can live with a label that contains such as 'address' encompassing such
                    > as 'label' in the context of forwarding.  I take labels to mean what
                    > labels mean rather than what I might find in a work of reference.
                    >
                    > Tom Petch
                    >
                    > The following errata report has been submitted for RFC8349,
                    > "A YANG Data Model for Routing Management (NMDA Version)".
                    >
                    > --------------------------------------
                    > You may review the report below and at:
                    > https://urldefense.com/v3/__https://www.rfc-editor.org/errata/eid6251__;!!NEt6yMaO-gk!URK5WVsqD5g7WpzCU1VuzKJA0AUiawXBFLB_gENlsYMrpiMqDtyFoxw8DnSr2A$
                    >
                    > --------------------------------------
                    > Type: Technical
                    > Reported by: Tarek Saad <[e-mail]>
                    >
                    > Section: 7
                    >
                    > Original Text
                    > -------------
                    > The RPC "active-route" is used to retrieve the active route in a RIB.
                    > RFC8349 defined two AFIs (v4/v6).
                    >
                    > draft-ietf-mpls-base-yang is defining a new RIB AFI for MPLS as per
                    > section 3 in RFC8349.
                    >
                    > The RPC has a "MUST" statement that all RIBs must augment input
                    > parameters with a leaf named 'destination-address'.
                    >
                    > For MPLS RIB, it makes sense to augment with leaf named 'local-label'
                    > since MPLS routes are identified by MPLS label.
                    >
                    > We ask to make the following change:
                    >
                    > OLD:
                    >            action active-route {
                    >              description
                    >                "Return the active RIB route that is used for the
                    >                 destination address.
                    >
                    >                 Address-family-specific modules MUST augment input
                    >                 parameters with a leaf named 'destination-address'.";
                    >
                    >
                    > Corrected Text
                    > --------------
                    > NEW:
                    >            action active-route {
                    >              description
                    >                "Return the active RIB route that is used for the
                    >                 destination address.
                    >
                    >                 Address-family-specific modules MUST augment input
                    >                 parameters with a suitable leaf that identifies the
                    > route.";
                    >
                    >
                    > Notes
                    > -----
                    >
                    >
                    > Instructions:
                    > -------------
                    > This erratum is currently posted as "Reported". If necessary, please
                    > use "Reply All" to discuss whether it should be verified or
                    > rejected. When a decision is reached, the verifying party
                    > can log in to change the status and edit the report, if necessary.
                    >
                    > --------------------------------------
                    > RFC8349 (draft-ietf-netmod-rfc8022bis-11)
                    > --------------------------------------
                    > Title               : A YANG Data Model for Routing Management (NMDA
                    > Version)
                    > Publication Date    : March 2018
                    > Author(s)           : L. Lhotka, A. Lindem, Y. Qu
                    > Category            : PROPOSED STANDARD
                    > Source              : Network Modeling
                    > Area                : Operations and Management
                    > Stream              : IETF
                    > Verifying Party     : IESG
                    >
                    > _______________________________________________
                    > netmod mailing list
                    > [e-mail]
                    > https://urldefense.com/v3/__https://www.ietf.org/mailman/listinfo/netmod__;!!NEt6yMaO-gk!URK5WVsqD5g7WpzCU1VuzKJA0AUiawXBFLB_gENlsYMrpiMqDtyFoxxyc2_LZA$



            Juniper Business Use Only