Minutes interim-2022-core-07: Wed 16:00
|Meeting Minutes||Constrained RESTful Environments (core) WG|
|Title||Minutes interim-2022-core-07: Wed 16:00|
CoRE Virtual interim - 2022-05-25 - 14:00 UTC
- Marco Tiloca, RISE
- Jaime Jiménez, Ericsson
- Carsten Bormann, TZI
Minute takers: Christian, Marco Tiloca (helping)
Jabber scribes: Marco Tiloca
Remember that the note well applies, for IPR but also for WG
processes and code of conduct. Please be nice to each other.
Jabber & Minutes / Agenda bashing
MT doing introductions.
Concise Problem Details For CoAP APIs
TF (p2): Draft went to multi-WG WGLC (CoRE, CBOR, HTTPAPI) until
TF (p3): Surprisingly no pushback on the upgrade of CBOR tag 38.
Positive comments from Michael Richardson and Christian Amsüss.
TF (p4): Jaime is on the shepherd write-up; Esko gave positive feedback
on media type registration.
TF (p6): Next steps are submitting a new version -04 and proceed with AD
review. That would meet the June 2022 milestone.
TF (p6): For discussion: Should we ask for area reviews already to
happen in parallel?
CB\: It's not us who ask for area reviews but ADs, so next step for us
is to pass it on to Francesca -- but we can suggest that she requests
Carsten's extra slides:
CB going through WGLC comments and their processing on Github.
CB (p3-4): Mostly editorial (11/18 commits) and IANA-related fixes.
CB (p5): base direction auto can now be explicit. (If you don't have
more complex environment where direction can be inherited, treat absence
like null). All "how to do it" is a draft and clearly not a W3C
recommendation, so we're defining this into uneven landscape. Purpose is
to deliver information for the algorithm (which is not defined yet, so
we can't normatively reference there). Side chat with i18n directorate,
there is rough consensus but not written up.
CA (on chat): Thumbs up on the new i18n text.
CB (p6): New term "Problem shape". Previously we could enumerate problem
types, we can't any more. Still possible to import from RFC 7807 if you
have that information. Added explanations about that, it should be
CB (p7): If there is an embedded URI reference, the base URI might not
be clear and needs to be specified. Not really needed when exchanging
messages on the wire, but it becomes needed when storing a problem
details data iterm. More context, i.e., the base URI, should also be
part of what is stored for later processing.
CB (p7-8): We have now defined a base URI Standard Problem Details entry
and explain how to use that. Meshes with response-code -- would be lost
when storing just data item. Based on a mailing list thread, we should
(probably elsewehere) define what needs to be persisted when storing a
representation, but for here we can do it this way.
CB (p9-10): Ignoring unknown entries is a common JSON extensibility
pattern (forward compatibility, can add details to items where the item
will still be understood by old implementations). False interoperability
can still be prevented if required (with new Custom Problem Details
entries); this keeps new data from influencing old receivers. Can't
write up all of it, what's in should largely suffice.
CB (p10): Unknown map keys are easy to ignore, but what if data looks
different? "internal ignore unknown" is a special case of this. There is
a danger of overvalidation (which blocks extensibility / forward
evolution). Do we need to define that part of the JSON ignore-unknown
pattern? Expectation around enums are easy; numeric restrictions are
often handled intuitively. I don't think we should include text here,
but be aware that it's not fully defined in the extensibility story.
CB (p11): (careful, slide wrong in the last character of the body,
s/]%/}/). Prefer to keep it, because it's more natural in CBOR
environment. But it's a change, and depends on meaning of TS29112. But
the purpose of this is not to do 29112, but example of custom problem
TF\: Initially skeptical on having the base URI in the set of Standard
Problem Details entry, but it makes sense now that I see it in the light
of error response codes.
CA (on chat): My comments have been addressed, thanks.
JJ (p6): Discussed with Ari. Looking for 1:1 mapping to RFC 7807. It
could be helpful to put more on RFC 7807 type mapping in here;
yesterday's text was unclear.
CB\: We have the whole appendix B on the topic of RFC 7807, it also
JJ\: We can take it offline.
JJ (p11): Soft agree with Carsten (the array makes sense), but whatever
authors and the group decide is fine.
TF\: As Carsten mentioned, this is just an example. Do you want a change
JJ\: If it's a non-contentuous issue, the second form would be more
clear for casual readers (especially in the context of 3GPP).
CB\: The first form does not use any strings for identifying things;
that's what we normally want in concise environment.
TF\: It took some poetic license for translating the API into CBOR. It
seemed like a normal choice in this environment.
JJ\: Ambivalent; it's a bikeshed issue. If you justify what you put in
there, you can keep that.
JJ\: Timeline-wise, what's left? We check with Francesca to start the AD
parts; anything else that needs to be done?
CB\: I will submit a new version based on this PR today, and then we can
submit to the IESG at the next possible time (Friday this week). I don't
think we should wait a lot, there's still the whole IETF LC ahead.
MT\: So Shepherd write-up ready by Friday?
JJ\: Ready now. I may do changes depending on v04 today / Friday.
CB\: I will submit new version of the draft in next hour.
MT\: Good and fast process on a urgent item; good job everyone.
Proxy Operations for CoAP Group Communication
- Ready for WG Adoption ?
ED (p2, p3): recap; concrete solutions to proxy issues raised in
draft-ietf-core-groupcomm-bis. It builds on two new CoAP Options used by
client and proxy. Group OSCORE would work end-to-end between client and
servers. Need for security between client and proxy; if OSCORE, that's
defined in draft-tiloca-core-oscore-capable-proxies.
ED (p4): Renamed the two new Options and revised their semantics based
on comments from Christian and Carsten.
ED (p5): A reverse-proxy acts like an origin server; depending on
configurations and information at the client, it is admitted to have a
default timeout at the proxy, rather than a timeout negotiated with the
ED (p5): Notes on proxy-servers response revalidation; problem if also
Group OSCORE is used; might be solved with an "outer ETag", which may be
good to define in draft-amsuess-core-cachable-oscore (since that would
be needed in the first place for caching)
ED (p6): Added new example with a reverse-proxy, building on the URI
construction of RFC 8075. Indications used for forwarding are in
Uri-Path options. That means less addresses/ports that the reverse-proxy
has to reserve for group forwarding.
ED (p6): Sketched proposal to further extend the HTTP-CoAP proxy. We can
further use the Transfer-Coding "Chunked" at proxy, to relay responses
back to the client a stream like in the CoAP-CoAP care, rather then as a
single final HTTP response containing them. It seems doable, but not
CB\: Wondering about Chunked. How are responses delimited? Chunk
boundaries are not preserved if you have more HTTP-HTTP proxies in the
chain doing some juggling.
CA\: It's already in the draft -- it's multipart (and the chunks may or
may not align with the multipart boundaries).
CB\: Ok good.
ED (p7): Summary of features at a glance: addressing groupcomm-bis
issues with proxies; two new CoAP options; defined response caching at
the proxy; new CoAP option Group-ETag for Client-Proxy response
revalidation; CoAP-CoAP and HTTP-CoAP proxies; forward/reverse proxies;
chain of proxies.
ED (p8): Related to other documents. This is a case of non-traditional
responses (see draft-bormann-core-responses); it's a use case for using
OSCORE between client and proxy, possible together with Group OSCORE
end-to-end yielding to nested OSCORE (see
heavily relies on this document to address issues with proxies, better
to have it adopted before requesting publication of groupcomm-bis, see
also the WGLC review at
ED (p9): Overview of next steps, main features of this protocol are
stable. Plan to add also security considerations specific to HTTP-CoAP
proxies (building on RFC 8075).
ED\: It should be ready to consider for WG Adoption.
RH\: General comment: I've been following this work and working on an
implementation (considering a forward proxy at the moment), based on
Californium (which also has Group OSCORE support).
CA\: on p5, a default timeout for a reverse-proxy sounds problematic;
the client may know it's multicast and then opt in. It can be otherwise
problematic. The client might have no idea it's a reverse-proxy (that's
usually the case), so it's better to leave it for negotiation with the
CoAP option. Otherwise, the client can run into strange things due to
something unexpected happening as to receiving relayed responses.
CA\: For the client, the reverse-proxy is just a regular server.
CB\: The state machine of the protocol can not depend on information
that might be unevenly distributed. A specific thing is token retirement
at the client: if I send a request to someone and get a response, I'd
reuse the token, but if more responses come in, that's not great. I
Really need to know.
ED\: So on the protocol level the right tokens need to be generated. Not
want to take the risk that CoAP client is unaware of this and uses wrong
tokens. Maybe some default from application if requested?
CA\: The reverse-proxy could still not know the application.
CA\: on p6, using the Uri-Path approach of RFC 8075 might not be a good
idea here; need to look into that. We have Uri-Host! Could you elaborate
on packing the address in Uri-Path?
ED\: We had a similar discussion in groupcomm-bis to express the names
CA\: Relying on Uri-Host would work and it should be more appropriate
than Uri-Path, thus avoiding related problems.
ED\: Need to build some examples. We can take it on the mailing list.
CA\: And none of that would be in the way of a WG Adoption, which I
CB\: Christian already took some of what I was about to say. it seems
that we do agree on the fundamental approach of this document, and it
looks like we have energy for working on it. All the prerequirements for
WG Adoption are there. Are we waiting for anything specific, or do we do
ED\: Not waiting for anything specific; I agree that we have energy.
CB\: With Marco a co-author and Jaime not being in the meeting any more,
the Chairs should look into this soon.
MT\: reminder: we have a still ongoing WG Last Call for
draft-ietf-core-conditional-attributes; it was planned to last until
today, but I can see only my own review on the mailing list. Really need
for more feedback before closing WG Last Call. Please send feedback.
("I'm fine with it as is" is also valuable).
Don't forget to bring a towel!