[dev] FRR North bound API

Tue Jan 15 07:47:25 EST 2019

Hi Mohammed,

Thanks for your interest in the interaction between ODL and FRR proposal.
I showed the two drawings, to try to identify the gap between that
design and the current FRR.

About #2585, this is an unrelated feature request. This was for
substituting the ZAPI interface with gRPC.
This is different in our case; In our case, we look for having a gRPC
northbound API. Concretely, instead of vty, this would be gRPC.

The fact is that at first sight, I was thinking that the first drawing
was the only solution to be accepted.
Actually, the more I look with solution, the more I think there is yet a gap.
I can expose you the gaps that I identified with solution 1.

o There is first an issue in the drawing, that mixes both FIB and RIB.

Currently, the remote dataplane plugin is able to configure
remote-dataplane as FIB.
So, one could imagine having a dataplane plugin extension as
Southbound interface equipped with gRPC.

The reverse is not possible; this is not ZEBRA FIB is configurable,
but ZEBRA RIB.
That implies that an other plugin extension as Northbound interface
equipped with gRPC should be made.
This would either be located in Zebra, or in a separate daemon.

o The second issue is that ODL, when configuring the BGP layer,
handles some pure BGP concepts.
For instance, Route Distinguisher is a pure BGP concept. Injecting
entries in a non BGP instance would
mean that we hardcode the relationship between VRF and RD.
This would mean that ODL, when injecting route entries, would inject
per-VRF route entries, and not per-RD route entries.

As other example, End-of-RIB is something handled after having
injected a whole series of prefixes. This is a BGP marker
that informs the protocol about the termination of the route entries.
This is useful on graceful restart scenarios.
Those concepts are internal to BGP, as far as I know. Using those
marker outside of BGP will require some work.

o The third issue is related to the Zebra layer.
Having a remote dataplane plugin is not enough. Some BGP information
is kept on BGP only, and never goes to ZEBRA layer.
Because ODL acts as the real dataplane configurator, ODL today has not
really the need to rely on ZEBRA layer.
However, having remote dataplane will require some extensions to
ZEBRA. For instance, some EVPN entries are not exported to
the correct VRF.

Actually, like the first proposal, the second drawing takes reuses
gRPC and yang based approach, thus facilitating
the interoperability work between ODL and FRR.
Adding to this, the second drawing does not have the 3 above issues.
The configuration of BGP route entries would be
done by using BGP. Also the update of BGP route entries would be done
by registering to yang nodes.

Mohammed,
Now that I exposed you the main problems of the first drawing, what is
your opinion about the concrete solutions that should be applied ?

Thanks,
Philippe

On Mon, Jan 14, 2019 at 3:57 PM Mohammed Anzal K via dev
<dev at lists.frrouting.org> wrote:
>
>
>
>
> ---------- Forwarded message ----------
> From: Mohammed Anzal K <mohammeda at altencalsoftlabs.com>
> To: dev at lists.frrouting.org
> Cc:
> Bcc:
> Date: Mon, 14 Jan 2019 20:26:04 +0530
> Subject: Re: [dev] FRR North bound API
> Hi Philippe,
>
> Thanks for your inputs.
>
> I am also interested in the use case of L3VPN/L2VPN for ODL-FRR interaction.
>
> I would root for the approach mentioned by you in the first slide, of interfacing the zebra daemon for prefix handling, to ODL over gRPC.
>
> Meanwhile, when are we planning to start with the bgpd specific north bound API related work ?
>
> Is there any planned timeline for this feature request         #2585 - gRPC support for Zebra
>
> Thanks
>
> Anzal
>
> On 19-12-2018 08:34 PM, Philippe Guibert wrote:
>>
>> Hi Anzal,
>>
>> About your question related to ODL-FRR interaction, may I know if you
>> have a specific ODL service in mind ?
>>
>> I have come up with design proposals on ODL/FRR interaction for L3VPN
>> and L2VPN in ODL service.
>> Main points are:
>> - rely on the new yang NB API ( using GRPc technology).
>> - optional : extend L3VPN/L2VPN to Zebra, via remote dataplane
>>
>> You can find specific proposal in [0]
>>
>> Anzal, or everyone else in the FRR community,
>> your comments are welcome,
>>
>> Thanks,
>>
>> Philippe
>>
>> [0] https://docs.google.com/presentation/d/18LkCGc8HAXto-c5IRT0NUBLmKRBYM6Oq6D5QNk_W-_k/edit#slide=id.g49ea94a8e7_0_0
>>
>> On 11/21/18, Jeff Tantsura <jefftant at gmail.com> wrote:
>>>
>>> Hi,
>>>
>>> Netconf is a configuration protocol and is not particularly suitable for
>>> streaming data, one could use
>>> YANG-push(https://tools.ietf.org/html/draft-ietf-netconf-yang-push-20), has
>>> been implemented in ODL’ Baron release
>>> (https://docs.opendaylight.org/en/stable-boron/user-guide/yang-push.html)
>>> however - general performance implications apply,
>>> gRPC on another hand is de-facto industry standard for streaming telemetry
>>> OpenConfig has defined number of frameworks.
>>> Specifically for prefix injection, gRIBI
>>> (https://github.com/openconfig/gribi/blob/master/doc/motivation.md) could be
>>> considered, this what we (IETF) envisioned with I2RS effort but, reality is
>>> different.
>>>
>>> Cheers,
>>> Jeff
>>>
>>>
>>>> On Nov 17, 2018, at 12:26 AM, Philippe Guibert
>>>> <philippe.guibert at 6wind.com> wrote:
>>>>
>>>> Hi Anzal,
>>>>
>>>> I agree with your proposal, with some concerns however.
>>>>
>>>> I presented a few weeks ago at FRR weekly meetings, some slides about
>>>> what would be the mechanisms involved for an SDN controller to be able
>>>> to use FRR as BGP speaker. ODL can be the target, obviously. See in
>>>> [0]
>>>>
>>>> The yang solution applied to BGP northbound API has been proposed.
>>>>
>>>> However, there were two remarks:
>>>> - on the data exchanged between SDN and FRR.
>>>> What was ok was for configuration ( BGP, VPN, BGP peers, ...), since
>>>> it fits well with BGP NB API.
>>>>
>>>> However, at one moment, SDN had to populate the IPs information coming
>>>> from its VMs. The best solution was to use an other API plugged into
>>>> ZEBRA ( and not BGP), so that this IP information could be seen as a
>>>> remote dataplane. Here, the API is not well defined yet. Should this
>>>> API be equipped with Yang too?
>>>>
>>>> - on the constraints between SDN and FRR. Some API require performance.
>>>> For instance, will yang fit well when SDN retrieves thousands of
>>>> prefix entries ?
>>>>
>>>> Remember that if you use a netconf server like netopeer, the data flow
>>>> will look like the below:
>>>> odl client <---> netopeer server <---> sysrepo server <---> bgp/zebra
>>>> server
>>>>
>>>> which may not be the optimised path when you want to retrieve a lot of
>>>> prefix entries. to be confirmed.
>>>>
>>>> There is an alternative. From document [1], figure 2 depicts a new NB
>>>> API that bypasses sysrepo, and relies on scripts (it looks like a data
>>>> format ( gRPC with protoBuf) carried close to sysrepo bgp data
>>>> exchanges ( using protobuf) could be used to exchange with routing
>>>> daemons.
>>>> Because that path would bypass netopeer and sysrepo, while keeping a
>>>> data format close to what is done with netconf ( though not all would
>>>> be possible), retrieving prefix entries ( and why not configuration
>>>> too) could be optimised. Here too, that path has to be confirmed.
>>>> Which technology could suit well with FRR (also from licensing point
>>>> of view..).
>>>>
>>>> Philippe
>>>>
>>>> [0]
>>>> https://docs.google.com/presentation/d/194vUFaWxmhfTpHyANHsP66Ffw0BQuJ9sI0ir2jrXyXo/edit#slide=id.g4538793aa3_0_0
>>>> [1] https://github.com/opensourcerouting/frr/wiki/Architecture
>>>> On Thu, Nov 15, 2018 at 1:59 PM anzal <mohammedanz at gmail.com> wrote:
>>>>>
>>>>> Hi
>>>>>
>>>>> @Renato Westphal
>>>>> "https://github.com/opensourcerouting/frr/wiki/Architecture"
>>>>>
>>>>> Can this not be used to manage, configure and get updates from the FRR
>>>>> routing daemons by an external NETCONF device, say an SDN controller ?
>>>>>
>>>>> Like ODL has a south bound Netconf -
>>>>> https://docs.opendaylight.org/en/stable-oxygen/user-guide/netconf-user-guide.html
>>>>>
>>>>> So by using this I think ODL would be able to configure and get prefix
>>>>> updates from the signalling daemons in FRR. Correct me if I am wrong
>>>>> here.
>>>>>
>>>>> Thanks
>>>>>
>>>>> _______________________________________________
>>>>> dev mailing list
>>>>> dev at lists.frrouting.org
>>>>> https://lists.frrouting.org/listinfo/dev
>>>>
>>>> _______________________________________________
>>>> dev mailing list
>>>> dev at lists.frrouting.org
>>>> https://lists.frrouting.org/listinfo/dev
>>
>> _______________________________________________
>> dev mailing list
>> dev at lists.frrouting.org
>> https://lists.frrouting.org/listinfo/dev
>
>
>
>
>
> ---------- Forwarded message ----------
> From: Mohammed Anzal K via dev <dev at lists.frrouting.org>
> To: dev at lists.frrouting.org
> Cc:
> Bcc:
> Date: Mon, 14 Jan 2019 06:57:46 -0800 (PST)
> Subject: Re: [dev] FRR North bound API
> _______________________________________________
> dev mailing list
> dev at lists.frrouting.org
> https://lists.frrouting.org/listinfo/dev