I think that I understand your perceptions better. Prior to adoption of
coap-tcp-tls and before I was active in the WG, I recall discussions related to
the confusion over application vs transport reliability in CoAP especially as
related to CON and NON. What was intended?
Tim Carey outlined some concerns in:
https://tools.ietf.org/html/draft-carey-core-std-msg-vs-trans-adapt-00#section-2
This topic was presented in detail at IETF 93 -
https://www.ietf.org/proceedings/93/slides/slides-93-core-0.pdf - starting on
slide 23.
And in a related thread on the mailing list back in 2015 -
https://www.ietf.org/mail-archive/web/core/current/msg06280.html - Carsten
responded:
In any case, CON and NON are about message layer semantics, not about
application semantics
-- you gave them a meaning they don't have.
By IETF 94, the authors were reporting – “Most of the Confusion around
CON/NON was resolved”.
Where relevant, I’ve added clarifications - such as the Appendix related to
differences in Observe for reliable transports.
Both Carsten and Hannes could probably offer more context if needed.
From: Yoshifumi Nishida [mailto:nishida(_at_)sfc(_dot_)wide(_dot_)ad(_dot_)jp]
Sent: Friday, April 21, 2017 2:08 PM
To: Brian Raymor <Brian(_dot_)Raymor(_at_)microsoft(_dot_)com>
Cc: Yoshifumi Nishida <nishida(_at_)sfc(_dot_)wide(_dot_)ad(_dot_)jp>;
tsv-art(_at_)ietf(_dot_)org; draft-ietf-core-coap-tcp-tls(_at_)ietf(_dot_)org;
core(_at_)ietf(_dot_)org; ietf(_at_)ietf(_dot_)org
Subject: Re: TSV-ART review of draft-ietf-core-coap-tcp-tls-07
Hi Brian,
Just in case,
Reliable transports only provide reliability at transport level. It doesn't
provide reliability in application protocol level.
RFC7252 has reliability mechanisms in it since it uses UDP. This means it has
abilities to check both transport and app level reliability.
This draft only provides transport level reliability and apps will need to
detect app protocol failure by themselves.
This means 7252 and this draft are not totally equivalent from the viewpoint of
applications.
I am not saying this is wrong or bad, but I believe app developer should aware
this point.
--
Yoshi
On Fri, Apr 21, 2017 at 11:15 AM, Brian Raymor
<Brian(_dot_)Raymor(_at_)microsoft(_dot_)com<mailto:Brian(_dot_)Raymor(_at_)microsoft(_dot_)com>>
wrote:
Hi Yoshi,
OK. I also think we should state that the protocol should notify the failure
events to applications.
Since errors can happen not only in TCP, but also TLS and websocket level,
mentioning only TCP close or reset might not
be enough.
After reviewing with the authors, an additional clarification was appended to
3.4 Connection Health - https://github.com/core-wg/coap-tcp-tls/pull/140/files
The opinion of the authors (and Gengyu WEI’s recent response -
https://www.ietf.org/mail-archive/web/core/current/msg08622.html) is that
RFC6455 covers the WebSocket case and does not need to be repeated here.
When we use 7252, I think applications basically don't need to implement
timeouts or retry mechanisms as the protocol
provides such things.
RFC7252 provides timeouts and retries because it's implementing a TCP-like
reliability mechanism over UDP - https://tools.ietf.org/html/rfc7252#section-2.1
However, when we use this one, it seems applications will need to have such
mechanisms. Isn't it a bit confusing? I am thinking that
there need to be some guidance here.
BTW, PONG is one example.
For coap-tcp-tls, there are multiple early implementations. This has never been
reported as a source of confusion.
My sense is that we should treat this as an update to RFC7959 based on the
original language:
I don't have a strong opinion here. Updating 7959 is fine for me if it's
clearer to CoAP people.
I've merged the change - https://github.com/core-wg/coap-tcp-tls/pull/138/files
Thanks again for helping us to improve the quality of the draft,
…Brian