9-11 October 2013 - Seattle, WA USA
Chair: Mark Nottingham
Participants:
- Martin Thomson (Microsoft)
- Gabriel Montenegro (Microsoft)
- Patrick McManus (Mozilla)
- William Chan (Google)
- Hasan Khalil (Google)
- Salvatore Loreto (Ericcson)
- Roberto Peon (Google)
- Fred Akalin (Google)
- Stephen Ludin (Akamai)
- Nicholas Hurley (Mozilla)
- Shigeki Ohtsu (IIJ)
- Brian Raymor (Microsoft Open Tech)
- Rob Trace (Microsoft)
- Mike Bishop (Microsoft)
- Osama Mazahir (Microsoft)
- Matthew Cox (Microsoft)
- Michelle Lai (Microsoft)
- Jeff Mendoza (Microsoft)
- Bryan Sullivan (AT&T)
- James Godfrey (Bluecoat)
- Birendra Keshari (Bluecoat)
- Aruna Balasubramanian (U Washington)
- Sophia Wang (U Washington)
- Saurabh Kulkarni (Akamai)
- Jason Duell (Mozilla)
- Jeff Pinner (Twitter)
- Herve Ruellan (Canon)
- Martin Nilsson (Opera)
- William Chow (Mobolize)
Remote Participants:
- Tony Hansen (AT&T)
- Gabor Molnar
- Emile Stephan (Orange)
- Giri Mandyam (Qualcomm)
- Leif Hedstrom (Apple)
- Ilya Grigorik (Google)
- Kalyani Bogineni (Verizon Wireless)
- Blue sheets
- Introductions
- Agenda Bashing
- NOTE WELL
- HTTP/2.0 Protocol Test draft
Michelle Lai presented her draft on HTTP/2.0 Protocol Test Principles. We then gathered implementation experience regarding draft-06; generally, feedback was good, although most problems were seen about header compression.
We then broke into informal groups, centred around protocol testing (Michelle Lai), header compression testing (Herve Ruellan) and general interop work between implementations.
- Blue sheets
- Introductions
- Scribe selection
- Agenda bashing
- NOTE WELL
Upgrade issue: Move the Upgrade mechanism from the 2.0 draft to a different spec until it will be fully tested ? Upgrade is a complicated mechanism to setup a plain text 2.0 connection.
Lots of discussion about Alternate-Protocol and the potential for that to replace the use of Upgrade.
There will be a separate draft describing the Alternate mechanism and it will be referenced in the next implementation draft version as experimental.
Header compression: discussion about removing "substitution operation" from the spec as it is quite complex to implement and brings little benefits. No clients at moment support it (is it true?)
Consensus to drop it for the time being and to evaluate it again in the future
Move the new header addition to the top (index 0)
The initial header table is currently mutable. Consensus to make the initial header table completely immutable and always present.
Discussion on having separate tables for client and server at all, or simply having a single table containing common header names for both. Consensus on one Big Table.
We're not going to define what header field values are or aren't. We're going to say that the compressor carries octet sequences and avoid the issue entirely.
Roberto described the use of this for doing things like reclaiming header compression table for long-lived connections.
This is one particular solution to the problem posed by wanting to change the size of the header compression table.
Proposal from Mike Bishop: https://github.com/MikeBishop/http2-spec/commit/fb59b5517105867f8fa681aecf6868da 9bc2a6e3
There is consensus on work on it. Discussion if include the mechanism for it in the next implementation spec or keep it in a separate spec. The proposal will be discussed in the list.
discussion on other possible reference sets to be used. Work on a draft proposal to be discussed in Vancouver.
Consensus on proposal #2. Remove the END_STREAM flag from CONTINUATION and determine whether a stream is ended based on the initial HEADERS frame flags.
FRAME_SIZE_ERROR
lets discuss it in Vancouver
Discussion on :host and host, potential chance to rename :host to :authority. Proposal to apply the "do not set and MUST ignore" rule for colon headers.
omit colon-prefixed headers to be treated as a malformed HTTP message. Intermediares should not act on interpret them just forwarding them.
I think some discussions occurred after issue 257
Some discussion about non-final status codes: issue raised #264 and assigned to mnot
Jeff Pinner wants to make the 2^14 limit apply only to DATA frames.
James Snell wants to make the 2^14 limit apply only to all except HEADERS/PUSH_PROMISE/CONTINUATION frames.
Roberto recognizes the layering violation, and would address it by making 2^14 the limit at the framing layer.
Patrick isn't concerned with the layering violation, but sees the possibility for avoiding CONTINUATIONS frames on large headers to be nice.
The coin toss in Seattle decided that it would be a fixed 2^14-1 limit for ALL frames, regardless of layer.
People like explicit disable PUSH_PROMISE as a binary on/off
clarify in the draft, push is allowed only for safe cachable methods + no request body.
it has been discussed the possibilty to use SETTINGS ACK + reset streams to limit server resource usage, instead of using profiles as proposed in the issue.
put the issue on hold until we get more implementation/interop experience.
closed
We talked about heuristics for avoiding DoS. Not a lot of enthusiasm. We've talked about PING and empty, acknowledged SETTINGS. We've talked about a new explicit GOAWAY code for "enhance your calm".
closed
closed
closed
closed. no interest in carrying the reason phrase.
discussion summarized in the issue
closed
closed.
closed.
How a proxy maps priorities from different connections/browser within a single connection towards the Origin Server. People agree that this is something worth to work on.
Priority dependencies, i.e. grouping, tree, list etc.
a new Issue #270 has been create to track the discussion.
- Blue sheets
- Introductions
- Scribe selection
- Agenda bashing
- NOTE WELL
We need to retain :authority and host as separate, because there are 1.1 cases that break without both. OAuth in particular.
lots of discussion on frame types
lots of discussion on alpn strings
some snoozing by the notetaker
Mark wants to collect input
Will is concerned about how many bytes this can use, as is Patrick
Jeff wants to use NPN