Re: [Lsr] draft-ietf-lsr-isis-rfc5306bis-01

["Acee Lindem (acee)" <[e-mail]>]

Hi Les, Uma,

Excuse the top-post but Outlook doesn’t do well with inline response for messages such as this one. AFAIK, no implementation defaulted to aborting graceful restart due to any topology change as recommended by RFC 3623.

Thanks,
Acee
From: Lsr <[e-mail]> on behalf of "Les Ginsberg (ginsberg)" <[e-mail]>
Date: Tuesday, May 28, 2019 at 9:02 PM
To: Uma Chunduri <[e-mail]>
Cc: "[e-mail]" <[e-mail]>
Subject: Re: [Lsr] draft-ietf-lsr-isis-rfc5306bis-01

Uma -

From: Uma Chunduri <[e-mail]>
Sent: Tuesday, May 28, 2019 5:09 PM
To: Les Ginsberg (ginsberg) <[e-mail]>
Cc: [e-mail]
Subject: Re: [Lsr] draft-ietf-lsr-isis-rfc5306bis-01

Les,


[Les:] The timers (T1, T2, T3) are NOT relevant to PR/PA.
PR is sent BEFORE a router does a restart to alert the neighbors that the signaling router’s control plane is going away for a time.
RR/RA are associated with what happens AFTER the router has restarted and now wants to reacquire adjacencies/LSDB.

Thx. Yes, I got that.


I do not know what text in the draft suggests to you that there is any relation between PR/PA and RR/RA.


Newly added Section 2.2.3 a says this:

       The "remaining time" transmitted according to (b)
       below MUST reflect the actual time after which the adjacency will
       now expire.

The above is same for section 2.2.1 a, which talks about RR and RA.
This is the reason, I asked, what is the implication of same timer value here for PR/PA.
For example, what are the implications of this new timer times out before the value specified in RA (as PR is obviously initiated before) ? Also see my original question.


[Les:] Tx timers do NOT apply to the neighbors of the restarting router – and they are the only routers whose control plane is alive while the restarting router is reloading.

No offense intended – but your question is bizarre – I really don’t understand what logic leads you to ask it. ☺

[Les:] What constitutes a topology change significant enough to trigger bringing down the adjacency is an implementation decision.
Definition of the conditions is NOT an interoperability issue and therefore does not fall within the scope of the draft.



I would have agreed with the above statement if this is the guidance for the restarting router. Here the topology change is detected by the neighboring router (see your text below)  and you do want a consistent behavior from neighboring node from any vendor.  No?
[Les:] The neighbor is free at any time to declare adjacency to the restarting as down. Obviously, if it does so, this will affect forwarding in the network. Whether the neighbor is making the “optimal choice” based on the topology change is an implementation decision and it is up to the customers to judge whether they are happy w the implementation choice or not. It does not affect interoperability.

We could have been more prescriptive – similar to https://tools.ietf.org/html/rfc3623#section-3.2 – but I think that is sub-optimal. It is possible for a topology change to occur which does not affect forwarding via the restarting node – in which case it isn’t helpful to bring the adjacency down.
Rather than try to detail all possible cases, we have left it as an implementation decision as to how “smart” an implementation wants to be. Given that the neighbor will send an updated LSP of its own reflecting the down adjacency in such a case, the rest of the network will know that there is no longer a path via the restarting router.

It simply isn’t necessary to mandate a behavior.

   Les



Section 2.2.3
"  While a control plane restart is in progress it is expected that the
   restarting router will be unable to respond to topology changes.  It
   is therefore useful to signal a planned restart (if the forwarding
   plane on the restarting router is maintained) so that the neighbors
   of the restarting router can determine whether it is safe to maintain
   the adjacency if other topology changes occur prior to the completion
   of the restart. "


--
Uma C.