Re: [Lsr] Question about draft-ppsenak-lsr-igp-ureach-prefix-announce

[Peter Psenak <[e-mail]>]

Zhibo,

On 05/08/2022 05:49, Huzhibo wrote:
> Peter:
> 
>> -----Original Message-----
>> From: Peter Psenak [mailto:[e-mail]]
>> Sent: Friday, August 5, 2022 1:55 PM
>> To: Huzhibo <[e-mail]>
>> Cc: [e-mail]
>> Subject: Re: Question about draft-ppsenak-lsr-igp-ureach-prefix-announce
>>
>> Zhibo,
>>
>> On 03/08/2022 21:09, Huzhibo wrote:
>>> Hi Peter:
>>>        Please see inline.
>>>
>>>> -----Original Message-----
>>>> From: Peter Psenak [mailto:[e-mail]]
>>>> Sent: Wednesday, August 3, 2022 11:20 PM
>>>> To: Huzhibo <[e-mail]>
>>>> Cc: [e-mail]
>>>> Subject: Re: Question about
>>>> draft-ppsenak-lsr-igp-ureach-prefix-announce
>>>>
>>>> Hi Zhibo,
>>>>
>>>> On 29/07/2022 20:49, Huzhibo wrote:
>>>>> Hi Peter:
>>>>>
>>>>> Supplement to yesterday's online questions, If a node that does not
>>>>> support IP Flexalgo, which has a default route, should the node
>>>>> process the IP Flexalgo prefix as a UPA?
>>>>
>>>> - I assume you are talking about the algo 0 default route. Because IP
>>>> Flex-algo default route does not make much sense really.
>>>>
>>>> - If the node does not support IP flex-algo, than it would not use
>>>> any IP algo prefix as BGP service endpoint or for any other purpose.
>>>>
>>>
>>> Which IP Algo prefix as BGP service endpoint is not determined by the ingress
>> node, Such as VXLAN and SRv6 VPN.
>>> When the ingress node receives an BGP Service cayyied a IP algo prefix
>>> as endpoint and it has a algo 0 default route, it should be process this BGP
>> service. and this can not be affected by the IGP Flexalgo prefix.
>>
>> sorry, but above is completely wrong.
>>
>> When you want to use IP flex-algo forwarding, you must advertise the prefix as
>> algo prefix, relying on the algo 0 default would not give you algo forwarding.
>>
>> Advertising IP algo prefix at the egress and relying in algo 0 default at the
>> ingress is going to cause all sorts of problems. You CAN NOT mix/change algos
>> along the path through the network - if you do, you may end up in a permanent
>> loop.
>    
>    If the ingress node does not support Flexalgo, the ingress node does not cause a
> permanent loop because once the packet is forwarded to the Flexalgo node, it always
> follows Flexalgo forwarding. If the packet does not reach the Flexalgo node, it always follows
> Algo 0 forwarding.

well, flex-algo was design for end to end forwarding. Switching between 
algos as packet traverses the network is not guaranteed to be loop free. 
If you don't trust me, I let you figure that out yourself when you do 
such a thing and it breaks.

> 
>>
>>> Therefore,
>>> the IGP does only not generate the RIB/Fib for LSinfinity Metric prefix, but can
>> not trigger BGP Service Down.
>>> In addition, LSinfinity Metric may be applied to other scenarios in
>>> the future. We cannot guarantee that LSinfinity Metric will not conflict with
>> other purposes when being processed as a UPA.
>>
>> no, it can not, because the LSinfinity has a very strict definition - it means
>> unreachable, which is exactly what the UPA is about.
>>
> I believe you are confusing a concept. The LSInfinity metric defined in RFC 5308
> can be considered as zero route, but PUA/UPA actually defines a negative route.
> It's not consistent

I'm not confusing anything:

rfc2328:

LSInfinity
         The metric value indicating that the destination described by an
         LSA is unreachable.

regards,
Peter



> 
>> Peter
>>
>>>
>>>> - If such node receives the IP algo prefix and even if it treats it
>>>> as UPA, given that such IP algo prefix was never reachable before on
>>>> this node, the UPA would result in no action.
>>>>
>>>> thanks,
>>>> Peter
>>>>>
>>>>> Thanks
>>>>>
>>>>> Zhibo
>>>>>
>>>>
>>>
>>>
>>
> 
>