Re: [Pce] Shepherd's review of draft-ietf-pce-pcep-exp-codepoints

[Jonathan Hardwick <[e-mail]>]

I'm OK with downgrading the "must" to a "may" (in lowercase).
Cheers
Jon



From: Adrian Farrel [mailto:[e-mail]]
Sent: 19 November 2017 14:43
To: 'Julien Meuric' <[e-mail]>; 'Dhruv Dhody' <[e-mail]>; Jonathan Hardwick <[e-mail]>
Cc: [e-mail]; [e-mail]
Subject: RE: [Pce] Shepherd's review of draft-ietf-pce-pcep-exp-codepoints

That works for me, although I would probably lower-case the "MAY".

A

From: Pce [mailto:[e-mail]] On Behalf Of Julien Meuric
Sent: 17 November 2017 17:30
To: Dhruv Dhody; Jonathan Hardwick
Cc: [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
Subject: Re: [Pce] Shepherd's review of draft-ietf-pce-pcep-exp-codepoints

Hi,

IMHO, the correct wording lies in between. RFC 5440 set the default for PCEP ("A PCEP-ERROR object is used to report a PCEP error"). Further specification (e.g. RFC 8231) MAY add message-specific behavior, but I think it is wrong to mandate a new behavior for each new message. I would thus suggest :
   If the PCE does not understand or support an experimental object with
   the P flag set in the Object Header, in the Path Computation Request
   message (PCReq), the entire PCEP message is rejected and PCE responds
   with a PCErr message with Error-Type="Unknown Object" or "Not
   supported object" as described in [RFC5440]. Otherwise, the object is
   ignored. Message-specific behavior MAY be specified (e.g., [RFC8231]
   defines rules for a PCC to handle an unknown object in an Update
   (PCUpd) message).

My 2 cents,

Julien

Nov. 13, 2017 - [e-mail]<mailto:[e-mail]>:
<snip>


Section 5

The following paragraph does not tell the whole story.

   A PCE that does not recognize an experimental PCEP object, will
   reject the entire PCEP message and send a PCE error message with
   Error- Type="Unknown Object" or "Not supported object" as described
   in [RFC5440].

If the P flag is clear in the object header, then the PCE MAY ignore the object instead of generating this error message. Also, you do not discuss what a PCC would do on receipt of a PCUdp or PCInitiate containing an unrecognised experimental object - it is inconsistent that you don't cover these cases.  (FWIW, RFC 8231 is a bit ambiguous about what a PCC should do about the PCUpd. Section 6.2 says that a PCErr should be sent, but then it refers to section 7.3.3, which says that a PCRpt should be sent. Hmmm.)

[[[Dhruv Dhody]]] Yes. How about I update to this -

   If the PCE does not understand or support an experimental object with
   the P flag set in the Object Header, in the Path Computation Request
   message (PCReq), the entire PCEP message is rejected and PCE responds
   with a PCErr message with Error-Type="Unknown Object" or "Not
   supported object" as described in [RFC5440].  Otherwise the object is
   ignored.  In case of stateful PCE messages [RFC8231], the P flag is
   ignored and the unknown object handling is as per the stateful PCE
   extensions.

And let's try to handle the inconsistency in RFC 8231 with an errata perhaps? And handle PCE-initiated during AUTH48?

[Jon] I think this is OK, but if we are just going to point the reader at RFC8231, then we might as well do the same with RFC5440, rather than duplicate its text.  And we should write something that allows for the possibility that more message types may be relevant in future.  How about

   If a PCEP speaker does not understand or support an experimental object
   then the way it handles this situation depends on the message type.
   For example, a PCE handles an unknown object in the Path Computation Request
   (PCReq) message according to the rules of [RFC5440].  A PCC handles an
   unknown object in an Update (PCUpd) message according to the rules of [RFC8231]
   and, in an LSP Initiate Request (PCInitiate) message, according to the rules of
   [I-D.ietf-pce-pce-initiated-lsp].  Any document that adds a new PCEP message
   type must specify how to handle unknown objects on that message.

Note that this last sentence is not an RFC2119 MUST because it defines author behaviour, not device behaviour.