Re: [Idr] John Scudder's No Objection on draft-ietf-idr-bfd-subcode-04: (with COMMENT)

["Dale W. Carder" <[e-mail]>]

Thus spake Jeffrey Haas ([e-mail]) on Tue, Dec 27, 2022 at 05:36:58PM -0500:
> On Mon, Nov 21, 2022 at 02:34:41PM -0800, John Scudder via Datatracker wrote:
> > ### Section 2, SHOULD v. MUST
> > 
> >    When a BGP connection is terminated due to a BFD session going into
> >    the Down state, the BGP speaker SHOULD send a NOTIFICATION message
> >    with the Error Code Cease and the Error Subcode "BFD Down".
> > 
> > Why isn't it a MUST? If the qualm is over the fact you might not always be
> > able to send a notification message, perhaps rewrite something like "the
> > NOTIFICATION message sent by the BGP speaker MUST use Error Code Cease and
> > Error Subcode 'BFD Down'"? Or "MUST attempt to send", "MUST send if
> > possible", etc.
> 
> (Note, I expect this to be the primary point of semantic wrangling in this
> review...)
> 
> The motivation to use SHOULD rather than MUST is an attempt to avoid
> creating unreconcilable conformance failure in the presence of multiple
> failure inputs.  Such headaches still haunt me from years of dealing with
> interop labs.
> 
> When BFD Down occurs, there may be additional failures that could be
> occurring.  For example, the interface for the peering session may have gone
> down.  Under any such conflicting circumstances, a MUST sets up the
> implementation to have to choose which is the best root cause to report.
> 
> Thus, does this document say that because the implementation saw a BFD Down
> it MUST use that code?  

In looking at RFC 4486 section 4 which describes usage of 8 other Cease 
NOTIFICATION subcodes, the majority of those (including administrative
shutdown) are also SHOULD vs MUST.  It looks like the only MUST for that
matter is for exceeding prefix limits.

> Put differently, what language would you want in this document to allow for
> "choose the best error code given the full knowledge available"?

When reported up to NMS systems, BFD Down is a leading indicator but to
me it would be expected that it is correlated with other data by an operator 
to determine root cause for a session being down.  We see similar issues
for example when BFD is used with IS-IS. 

Dale