ietf
[Top] [All Lists]

RE: [dhcwg] RE : Gen-art review: draft-ietf-dhc-triggered-reconfigure-05

2013-04-29 10:49:54
FYI - Stable storage or some kind of redundancy solution may be another way to 
recover state. So, I don't think there is a hard requirement for RFC 5460.

I don't think there is a strong reason to avoid e.g., it is "for example" so it 
is only one of many possible. And "such as" is basically saying the same thing.

---

But then we come to why this is even important ... The original text was:

   This setup assumes the relay has a record of the client, so that it
   has enough information to send the Reconfigure-Request message to the
   server.  How the state is recorded in the relay is out of scope.
   Furthermore, means to recover state in failure events must be
   supported, but are not discussed in this document.

I'm a bit skeptical why this draft even brings the 'recover state' up. Yes, the 
relay does need to record the client in order to initiate a 
Reconfigure-Request. But I think the sentence "Furthermore, means to recover 
state in failure events must be supported, but are not discussed in this 
document." is not needed. Sure, if the relay has no way of recovering this 
state, traffic to/from clients will likely not work and certainly if the relay 
is reconfigured, it will be unable to generate a Reconfigure-Request. But I 
don't really see this draft needing this requirement. Reconfigure-request could 
still work until the state is lost. And if the state is lost, there is a lot 
more than is potentially broken.

But I don't really see this as a big issue and the "must" is the lower-case 
variant anyway.

- Bernie

-----Original Message-----
From: Robert Sparks [mailto:rjsparks(_at_)nostrum(_dot_)com] 
Sent: Monday, April 29, 2013 9:45 AM
To: mohamed(_dot_)boucadair(_at_)orange(_dot_)com
Cc: Bernie Volz (volz); dhcwg(_at_)ietf(_dot_)org; General Area Review Team; 
ietf(_at_)ietf(_dot_)org; 
draft-ietf-dhc-triggered-reconfigure(_at_)tools(_dot_)ietf(_dot_)org
Subject: Re: [dhcwg] RE : Gen-art review: 
draft-ietf-dhc-triggered-reconfigure-05

On 4/29/13 12:59 AM, mohamed(_dot_)boucadair(_at_)orange(_dot_)com wrote:
Robert,

Ted suggested a wording which is better than the one I proposed. I made the 
following changes my local copy:

(1)

OLD:
    When retransmission is required, the relay may decide to correct the
    content of RECONFIGURE-REQUEST message it issues (e.g., update the
    Client Identifier list).

NEW:
    The relay MAY remove clients from the client identifier list in
    subsequent retransmissions, but MUST NOT add clients to the client
    identifier list.

(2)

OLD:
    Furthermore, means to recover state in failure events must be
    supported, but are not discussed in this document.

NEW:
    Furthermore, means to recover state in failure events (e.g.,
    [RFC5460]) must be supported, but are not discussed in this document.

Is this better?
1 is better.

2 is an improvement, I might say "such as" instead of e.g. to even more 
strongly avoid someone thinking you are _requiring_ implementation of RFC5460.  
Even then, it's still not clear whether you're trying to say

"Something that doesn't do this is does not conform to this specification"

or

"Something that doesn't do this isn't as good a tool as it can be and may 
create a condition that operators and users will not like much."

You chose not to use MUST in that sentence. Can you make it less likely that 
someone will assume you meant to?

Cheers,
Med


-----Message d'origine-----
De : Bernie Volz (volz) [mailto:volz(_at_)cisco(_dot_)com] Envoyé : samedi 
27 
avril 2013 06:24 À : Robert Sparks; BOUCADAIR Mohamed OLNC/OLN Cc : 
dhcwg(_at_)ietf(_dot_)org; General Area Review Team; 
ietf(_at_)ietf(_dot_)org; draft-ietf- 
dhc-triggered-reconfigure(_at_)tools(_dot_)ietf(_dot_)org
Objet : RE: [dhcwg] RE : Gen-art review: draft-ietf-dhc-triggered-
reconfigure-05

Robert:

The reason to allow this is that otherwise client A will be 
unnecessarily reconfigured many times. (It is also possible that a 
client might Renew on its own just as this is happening and thus it 
can also be removed from the
Reconfigure.)

I think the text should be cleaned up to indicate that allowing 
removal of already reconfigured clients is recommended (to prevent 
unnecessary
reconfigures) when retransmitting the Reconfigure-Request.

Note that if clients are added, that is not a retransmission but 
requires a "new" message (new XID).

- Bernie

-----Original Message-----
From: dhcwg-bounces(_at_)ietf(_dot_)org 
[mailto:dhcwg-bounces(_at_)ietf(_dot_)org] On 
Behalf Of Robert Sparks
Sent: Friday, April 26, 2013 12:19 PM
To: mohamed(_dot_)boucadair(_at_)orange(_dot_)com
Cc: dhcwg(_at_)ietf(_dot_)org; General Area Review Team; 
ietf(_at_)ietf(_dot_)org; 
draft-ietf- dhc-triggered-reconfigure(_at_)tools(_dot_)ietf(_dot_)org
Subject: Re: [dhcwg] RE : Gen-art review: draft-ietf-dhc-triggered-
reconfigure-05

On 4/26/13 10:58 AM, mohamed(_dot_)boucadair(_at_)orange(_dot_)com wrote:
Dear Robert,

Thank you for the review.
Please see inline.

Cheers,
Med
________________________________________
De : dhcwg-bounces(_at_)ietf(_dot_)org [dhcwg-bounces(_at_)ietf(_dot_)org] 
de la part de 
Robert Sparks [rjsparks(_at_)nostrum(_dot_)com] Date d'envoi : vendredi 26 
avril
2013 17:42 À : dhcwg(_at_)ietf(_dot_)org; ietf(_at_)ietf(_dot_)org; General 
Area Review 
Team; draft-ietf-dhc-triggered-reconfigure(_at_)tools(_dot_)ietf(_dot_)org
Objet : [dhcwg] Gen-art review:
draft-ietf-dhc-triggered-reconfigure-05

I am the assigned Gen-ART reviewer for this draft. For background on 
Gen-ART, please see the FAQ at


<http://wiki.tools.ietf.org/area/gen/trac/wiki/GenArtfaq><http://wiki
.tools .ietf.org/area/gen/trac/wiki/GenArtfaq>.
Please resolve these comments along with any other Last Call 
comments you may receive.

Document: draft-ietf-dhc-triggered-reconfigure-05
Reviewer:  Robert Sparks
Review Date: April 26, 2013
IETF LC End Date: May 6, 2013
IESG Telechat date: May 16, 2013

Summary: This draft is on the right track but has open issues, 
described
in
        the review.

Major issues:

Overall, this document is solid, but I think there is a bug in 
section
6.3.
I could be wrong, but If I'm right, this paragraph:

When retransmission is required, the relay may decide to correct the
content of RECONFIGURE-REQUEST message it issues (e.g., update the 
Client Identifier list).  This decision is local to the relay (e.g., 
it may be based on observed events such as one or more clients were 
reconfigured on their own).
introduces a race-condition that could lead to an erroneous state. 
If a
relay's first message included client A, then the retransmission 
included clients A and B, but that retransmission crosses with a 
success RECONFIGURE-REPLY to the request that only included client A, 
the relay will think it succeeded in asking A and B to be reconfigured.
Med: This example does not apply for that text.
Really? What text constrains how you change what's in the retransmission?
   In fact, the example should be the other way around. Perhaps, 
this can
be made clearer if we change "(e.g., update the Client Identifier 
list)" to "(e.g., remove a client from the Client Identifier list)".
If I understand you correctly, you need more than just changing a 
parenthetical e.g.. I think you're saying that you are constraining 
the message changes to be such that if anything earlier in the 
retransmission sequence succeeded, the request in the retransmission 
would also have succeeded. But why do you need that much complexity? 
Do you have it already with any other request?
Minor issues:

This sentence:

Furthermore, means to recover state in failure events must be 
supported,
but are not discussed in this document.
places a requirement on a relay (even though it's not using a 2119 MUST).
Is there some other document that defines this requirement that you 
can reference?
Med: I'm not aware of any; but if there is one we can cite it.

   If not, the requirement should be discussed in this document.
Alternatively, you could change the sentence to talk about the 
consequences of not having a proprietary means for recovering state.
Med: Will consider that option if you think this is really needed.

Nits/editorial comments:
_______________________________________________
dhcwg mailing list
dhcwg(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/dhcwg