Skip to content

Latest commit

 

History

History
378 lines (194 loc) · 13.7 KB

minutes.md

File metadata and controls

378 lines (194 loc) · 13.7 KB

httpbis WG minutes IETF-92 Dallas -- Tue afternoon 1300h

post-wg specs

no discussion

Current extension spec issues

alt-svc issue 44 - think we have agreement on it

pat mcmanus(pm) is fine with it

alt-svc issue 43 - ALPN Identifiers

martin thomson (mt) explains the detailed context ought to document the more broad connotations in the spec/doc

mnot: we should clarify that http/1.1 means HTTP/1.1 over TLS

mnot: we do have several issues open that aren't controversial -- need editors to update draft(s) and close out the issues

pm: ff37 impls this stuff, in beta, so folks can try that

OppSec issue 33

is open and needs to be incorp'd, then goto WGLC with this for 2 wks

mnot: moz impl'g, akamai on server side, other bwsrs?

desires to advance this in compnay of alternative svcs

tunnel protocol for CONNECT

has been discusssion, but no open issues, can goto WGLC

martin thompson(mt) there is not as much context in the into matterial so see the open pull request. Had priv feedback that they were unhappy with this but webrtc apps will have this header field forced on them

mnot: yes, in general can't make new header fields mandatory in http' but apps such as webrtc can do so, thus making that explict is nice/useful

Go ahead and put the text in, will issue WGLC

ekr: might need some wordsmithing that some traffice analysis of this stuff is possible -- will suggest text

Potential Work

See http wiki page "WatchList"

mnot: only want to keep WG open if there's useful work to do. Might be some of these the WG may want to dive into...

http signature

draft-cavege-http-signatures-04

tad early to adopt this?

HTTP client hints

have heard browser folk express interest in this

larry masinter(lm): there was work in this area 10 yrs ago -- already extant RFCs that cover much of this territory

mnot: agree that the above should be eval'd in that light

ekr: relates to client fingerprinting -- we ought to minimize passive fp surface and not worry so much about active so this draft may surface that struggle

interest in spec? some yes...

pm: not every client is a browser; its not like you require a ietf wg to mint a header -- could experiment actively with it not just hack spec

chris bentzel goog: interested in this; if you have tcp proxy this might be useful

peter griess FB (pg): interested in this too, eg for mobile website

?: might be useful for DASH

mnot: am concerned about privacy concerns tho

key http response header field

is a way to establish a secondary cache key that VARY selects. Vary is coarse, in usual case it is difficult to use to establish a secondary cache key. This spec is way to derive a more fine-grained secondary cache key.

might be either do this spec or kill 2ndry cache keys altogether

Ilya G likes this; if we adopted client hints -- we'd need this.

stuart cheshire (sc): iTunes team would like this

mt: concern w/this is the complexity -- trim it and leave open door for enhancing in future

sc: will ask them for specific needs

mnot: right not is kitchen sink-y (...describes...)

part of issues with this is header syntax

json encoding the http header fields

idea here is use json to describe header field, then serialize into header field value, could be a win

jr: it also makes internet aka i18n easier

phb: in long term we'll move to using json for everything. presently have class of headers for routing etc and a class for content -- would want to teases apart

eliot: how does jr view the transition to this working?

mnot: if i define a new header, then use this going forward

roy fielding: this is not an opportunity to define http/3

jr: trying to address defining new http header fields easier

phb: need a header field schema...

mnot: pushes back

www-authenticate I-D that uses the JSON HF definition technique

mt: would need something more concrete than "just use json". Suggests like two practices for how to use

pg: having doc on header definition best practices would be helpful

mnot: putting aside the specific proposal -- have a HUM on a doc defining a convention for defining header definition conventions

decent hum on interest

almost none on harmful

a little on don't know

mike bishop via jabber: this could encourage data smuggling via headers

oiwa: : something about the www-authn example

mnot: agree that was/is confusing

451 status code to report legal obstacles

mnot: discussion on list. A few folks think the semantics not well enough sorted out. Would like to discuss now.

barry leiba (bl) participant: is ok with what the WG thinks (formerly opposed). What were the args wrt insufficient semantics ? think they are pretty clear

paul hoffman (ph): tim puts up examples of folks using this. So, it can be used. It has shown to be sorta useful, but maybe not super useful.

mnot: some folks said just use 403 with a response body

ph: but that would change semantics of 403

lm: I pushed back on this purpose seems to be political theatre why not be informational?

mnot: a use case is if you want to crawl web and get indications of censorship

phb: this status code is supplying addttnl info. It isn't law here is the issue itself, but in brings the law into the picture....

jon peterson(jp): could imagine getting a diff page with a 406

wendy seltzer: chilling effects project thinks this useful potentially

mnot: we could define header fields that'd be used in conjunction with this status code

?: something about could indicate the proxy or origin blocking access to this content

mnot: two status codes perhaps

bl: anybody say "no"?

(no objection)

indicating char encoding and lang for http header field parameters

this'd be just adopting and finishing the spec.

should be close to done

"It was a WG spec" jr says -- but recants -- mnot celebrates

client init'd content encoding

another jr spec

allows client to use content-encoding on requests

server can use accept-encoding in responses

mnot asks client implementations -- would u use this?

jr: has addressed the issues from last meeting

mnot: is an incremental optional addition to http, some folks could use it, makes sense to adopt

lm: how's the interact with http header scoping?

mnot: have to treat as a hint

no objections otherwise

2NN Patch Status Code

mnot: esoteric use case

mt makes face

mt: I could see how it'd be useful in narrow cases. Could annoy api creators

roy: Yuk.

?: we currently do this when running commercial proxy

roy: effectively make all requests VARY

lm: don't you have to tie it to etag ?

mnot: yes, strong etag

?: w/intermediates, will be like 206

(roy later says it won't make all requests vary)

origin cookies and 1st party cookies

mnot: Seems a number of folks are interested, but they are limited given how deployment will happen. couldn't rely on them being there and so limited utility?

mt: ...various gnarlyness wrt cookies...

?: discussing with someone and intend to continue to work on this draft (which one?)

mt: in terms of csp, compliant UAs gain a bit of resistance to stuff like CSRF

mnot: need to have chat with Mike West and webappsec folk

mnot: some interest but need more discussion

encrypted content-encoding for http

mnot: working on this w/MT

mt: coupl use cases: webpush, need to push stuff thru intermediary, encrypt doc on server

mt: This is just a simple cut down version of tls record layer and some key mgmt

mt: there's some impls -- pretty interested in getting this going

mt: using jose didn't work in http semantics. eg a msg being processed continuously rather than a single blob

mt: no optionality in the format -- simple

mt: will engage w/sec folk

lm: there's desire afoot to secure http headers and optionally body and so ....

ph: why wont CMS w/smime eg smime 3 not work? pls engage smime list

mnot: so this spec is an fyi at this point

Issues against the httpbis docset

mnot: eventually take it to full std. once http/2 is done, align with http2, then try full std. Can't see us starting this now, but maybe later this year?

jr: no further thoughts

H2 (aka HTTP2) deployment advice

We have a faq on wiki, but info is thin.

There is perf increase but to really get in full need to tweak lots of deployment configs. Would be helpful to write down advice.

lm: have chatted w/folks this week. 2 aspects:

  1. What does it mean to benchmark http on the real internet? How do you measure whether one deployment is better or worse?

  2. and then deployment advice for how to attain improvements

mnot: w3c webperf wg -- could engage with them on this?

lm: lots of folks at IETF who know internet perf too -- will the lower layer folks interface with the webperf wg ?

mnot: RUM is interesting because it measures end user perf perspective -- in real world settings -- thus we really ought to engage with them

ph: we have benchmarking wg -- would not have this go there -- long time leads. Thinks webperf & rum is beter route

Dan Druta: goal-based: what do u want to improve? latency? load? etc. Stakeholder-based: e.g. for web developers, for web masters, for service providers, etc... both dimensions do both.

mnot: suggests starting effort on the wiki -- need folks to be willing to do work thinks webperf folks will eventually address it, but would be ncce if folks here would kick it off

mpeg dash

video streaming over http -- iteresting stuff on http2

herve ruellan (hr): DASH and HTTP/2 (see preso in meeting materials)

FDH -- full duplex over http

Dan Druta: detailed issues wrt pushing msgs to the server dunno if that was considered, but need to. "how quickly can u adapt?"

hr: yes can adapt but quickly is always difficult

? detailed ques wrt push msg characteristics sees benebit as ? seems lots of complexity w/o lots value didn't understand range option at all

jr & roy: request push sounds like GET, how is it diff than pipelined GET ?

hr: explains

mnot: http2 designed such that can interleave lots of reqs together. Here client can hint the server about something the client knows.

craig tayler(ct): concerned how server push will affect the ? semantics - might be issue

hr: yes, there issue here, an intent to have the server respond w/header to client. Then client knows to freeze the signal ?

mt: if u know what u need just ask for it. Thinks dash advantage is when server knows more about pacing of segments. E.g., if server sends next segment w/o client asking for it there's advantages. Also, client can notice about net issues and ask for lower quality.... hmmm maybe this looks like multi-get which we didn't do....

?: involved in dash. Yes there's a manifest file, server can update the manifest. Not clear whether h2 stack offers. Stacks have looked at don't offer this stuff as yet. That's something we need to address. We need a solution for this. Other think is whether push strategies is going to be defined somewhere.

questions about flow control

mnot: http2 doesn't strictly define how server push works, defined as cache update, not notification to the webapp client. Wonder if we need to notify a listener about cache updates.

pm: how can you prioritize things -- we did some of that w/GETs in http2 -- could use that? something about pushing addtnl metadata to the client could be useful.

lm: implicit context is using mpeg-dash for realtime. If you have a series of pipelined GETs, you'll get em all. Timing is implicit. If you want to generalize this you might make the timing explicit.

?: not sure dash is useful for RT, but maybe progressive download?

mt: something about canceling an outstanding req there's some useful cases for this. Avoid an XHR ?

mnot: don't hear interest in making this a generic extension yet.

mnot: Dash folks also have use case to send to client that they have partial segment to send want to send it so client can do what they want

mnot: I steered them from partial content to use a different media type to label it.

mnot: I will discuss appropriate times for new items with BL