[Teas] Re: Comments abount rfc2205 Resource ReSerVation Protocol (RSVP)

[Tuấn Anh Vũ <[e-mail]>]

Hi,
Thanks for all your answers, please find my view below:
*I./ Hi Adrian:*

*There are two questions that arise…*

   1. *Why isn’t R2 able to notice? Presumably the link failure detection
   is relying on a lower layer (L2 or L1) failure indication, and that is not
   happening. The answer to this is to run some other link failure detection
   mechanism such as BFD.*

*Such a mechanism would allow R2 to declare the link down and possibly
re-route/repair the LSP via R5, or notify the head end (R1) to let it
re-route.*

*AnhVT:* We use both LACP (1sx3) and micro BFD(300msx3) for this link but
some how LACP timeout on R3 first (this link is using Transmission System,
so the physical link is not down) and that triggers micro BFD sends Admin
down notify to remote (R2). This notification brings down micro BFD on R2
but not LACP client. This is expected behavior noticed in BFD RFC. After
2s, the link is stable again. Because of that, R2 does not know LACP was
flapped on R3 side.

   2. *How could R3 let R2 know that the LSP has been torn down? The answer
   is “by sending a PathErr or ResvTear or Notification”. In general, those
   messages are sent hop by hop, and so they would fail to be routed on the
   failed link R3-R2, however, it is possible to IP-tunnel to direct-address
   RSVP packets so that they would be IP-routed to R2 (or direct to R1) via
   R5.*

          * AnhVT: *Could you give me the document relative to this
behavior? I read some RSVP RFC but I don't observe this behavior.
*II./ Hi Pavan:*

   - *Stale state cleanup based on soft state time out (RFC2205): Since the
   link is down, the reservation state isn’t getting refreshed. So, when the
   reservation state times out (in about 157.5 secs for a refresh-interval of
   30 secs), R2 is expected to clean-up the reservation state and signal a
   ResvTear to the ingress*

          *AnhVT: *It takes a long time to clean up the reservation state
based on timeout. More than 2 minutes of blackhole traffic. This is
unacceptable for ISP services.


   - Use of RSVP Hello Session based on the Node-ID (RFC4558) for detection
   of RSVP-TE signaling adjacency failure: If there was an RSVP Hello session
   maintained between R2 and R3, R2 would be able to couple the state of the
   LSP with the state of signaling adjacency. And when the signaling adjacency
   failure is detected (Hello State timed out -- for a 9 sec hello interval,
   the time out takes 31.5 secs), R2 would clean up the reservation state and
   signal a ResvTear to the ingress. This option can be used to clean up stale
   state when long refresh intervals are used.
   - *AnhVT: *The LACP interface on R3 just flaps for 2s, so the node-hello
   can not work in this case. And 31.5 s is a long time too. I think we need
   some way faster. R3 should send Reserve Tear to R2 through IGP link
   (R3->R5->R2).


*III./ Hi Tarek*
*RSVP PathErr can be used to propagate errors upstream – there’s
Path_State_Removed that RFC3473 introduced to also notify that state has
been removed. Does that address your need?*
*AnhVT: *Let me check R3 send PathErr to R2 or not. As I know, R2 will not
bring down LSP even if R2 receives PathErr from R3. Base on
https://datatracker.ietf.org/doc/html/rfc2209
[image: image.png]

Regards,
AnhVT



Vào Th 4, 23 thg 10, 2024 vào lúc 01:23 Vishnu Pavan Beeram <
[e-mail]> đã viết:

> AnhVT, Hi!
>
>
>
> Since you are referring to an LSP in an IP/MPLS network, I’m assuming that
> you are using in-band RSVP signaling. I’m also assuming that this is an LSP
> that does not have any form of local-protection enabled.
>
>
>
> When R3 detects an upstream link-down event, it cleans up the local path
> state and sends a PathTear downstream -- in this scenario, the onus is not
> on R3 to notify the ingress of this outage. The typical expected behavior
> on R2 is to detect the downstream link-down event and send a PathErr to the
> ingress (signaled hop-by-hop) of the LSP. R2 would also clean-up the
> reservation state and send a ResvTear to the ingress (again, signaled
> hop-by-hop). If R2 is not able to detect the link-down event for some
> reason (and no other link state detection mechanism like BFD is available),
> there are a couple of control-plane options that RSVP already provides to
> clean up state (in due course of time) and bring down the LSP:
>
>    - Stale state cleanup based on soft state time out (RFC2205): Since
>    the link is down, the reservation state isn’t getting refreshed. So, when
>    the reservation state times out (in about 157.5 secs for a refresh-interval
>    of 30 secs), R2 is expected to clean-up the reservation state and signal a
>    ResvTear to the ingress
>    - Use of RSVP Hello Session based on the Node-ID (RFC4558) for
>    detection of RSVP-TE signaling adjacency failure: If there was an RSVP
>    Hello session maintained between R2 and R3, R2 would be able to couple the
>    state of the LSP with the state of signaling adjacency. And when the
>    signaling adjacency failure is detected (Hello State timed out -- for a 9
>    sec hello interval, the time out takes 31.5 secs), R2 would clean up the
>    reservation state and signal a ResvTear to the ingress. This option can be
>    used to clean up stale state when long refresh intervals are used.
>
>
> Hope this helps.
>
>
>
> Regards,
>
> -Pavan
>
> On Tue, Oct 22, 2024 at 4:34 PM Adrian Farrel <[e-mail]> wrote:
>
>> Hi Anh,
>>
>>
>>
>> [Redirecting from MPLS to TEAS as suggested by Tony Li]
>>
>>
>>
>> I think that (given you mention LSPs) you re talking about RSVP-TE (RFC
>> 3209) not plain old RFC 2205 RSVP.
>>
>>
>>
>> In your example, the link R2-R3 has failed in a way that R3 is aware of
>> the failure, but R2 is not aware.
>>
>>
>>
>> There are two questions that arise…
>>
>>    1. Why isn’t R2 able to notice? Presumably the link failure detection
>>    is relying on a lower layer (L2 or L1) failure indication, and that is not
>>    happening. The answer to this is to run some other link failure detection
>>    mechanism such as BFD.
>>
>> Such a mechanism would allow R2 to declare the link down and possibly
>> re-route/repair the LSP via R5, or notify the head end (R1) to let it
>> re-route.
>>
>>    2. How could R3 let R2 know that the LSP has been torn down? The
>>    answer is “by sending a PathErr or ResvTear or Notification”. In general,
>>    those messages are sent hop by hop, and so they would fail to be routed on
>>    the failed link R3-R2, however, it is possible to IP-tunnel to
>>    direct-address RSVP packets so that they would be IP-routed to R2 (or
>>    direct to R1) via R5.
>>
>>
>>
>> Cheers,
>>
>> Adrian
>>
>>
>>
>> *From:* Tuấn Anh Vũ <[e-mail]>
>> *Sent:* 22 October 2024 04:45
>> *To:* [e-mail]
>> *Subject:* [mpls] Comments abount rfc2205 Resource ReSerVation Protocol
>> (RSVP)
>>
>>
>>
>> Hi IETF team,
>>
>> I'm AnhVT from the SVTech company in VietNam, I have experienced some
>> RSVP issues in the IPv4 MPLS network.
>>
>> I suspect that RSVP has a point that needs to be enhanced. I
>> describe this point below:
>>
>>
>>
>> I./ Topology:
>>
>>     ---------LSP-------->
>>
>>     R1----R2----R3-----R4
>>
>>           |    /
>>
>>           |  /
>>
>>            R5
>>
>> II./ Issue
>>
>> 1./ Because of some bugs (exp: R3 experiences a flap link between R3-R2,
>> but R2 does not recognize the interface flap), R3 indicates that LSP is
>> down, then it deletes the LSP state and sends the PathTear downstream to R4.
>>
>> 2./Because R2 does not recognize the interface flap, R2 still keeps
>> it available. It does not know that the LSP should be deleted.
>>
>> 3./ Due to 1./ and 2./ R1 does not know that the LSP is stuck because R3
>> and R4 deleted the LSP state, and R1 continues forwarding traffic to the
>> LSP, This makes the service down.
>>
>>
>>
>> III./ My comment
>>
>> I think that RSVP needs a mechanic so that R3 signals to R2 to ensure
>> that R2 knows that R3 deleted the LSP. Based on that signal, R2 will bring
>> down the LSP and continue to send Reserve Tear to R1.
>>
>>
>>
>> I hope that you take a look at my comment.
>>
>>
>>
>> Regards,
>>
>> AnhVT
>> _______________________________________________
>> Teas mailing list -- [e-mail]
>> To unsubscribe send an email to [e-mail]
>>
>