- From: Roberto Peon <grmocg@gmail.com>
- Date: Mon, 27 Jan 2014 16:26:37 -0800
- To: Martin Thomson <martin.thomson@gmail.com>
- Cc: HTTP Working Group <ietf-http-wg@w3.org>, Amos Jeffries <squid3@treenet.co.nz>
- Message-ID: <CAP+FsNcE5s8oX_wz1y8Vo8K3_1L2xe6ccsa6ZvcGdDLcKLizjw@mail.gmail.com>
Keeping a list of streams and their state is something that some proxies may not do, especially if only involved in load balancing-- keeping the last seen stream-id is trivial, however. On Jan 27, 2014 3:12 PM, "Martin Thomson" <martin.thomson@gmail.com> wrote: > On 27 January 2014 14:56, Amos Jeffries <squid3@treenet.co.nz> wrote: > > Both ends already have that tracking in place. The costs I here would be > the > > extra bytes on the GTFO frame and the cycles used to generate that list > in > > the sender for which streams it still wants to receive. The cycles to > > process it in recipient would be the same ones for clearing up that state > > anyway. > > Not exactly. You know what streams you have, but the streams that > have crossed from 'not idle' states into 'no way back from here' isn't > something that necessarily needs tracking. > > It's a line call either way. Multiple stream identifiers is more > complexity all around, and the gain is marginal at best. > > > A list of only one entry (as currently used) meaning the connection > close is to follow and everything outstanding is borked. > > Signaling using a single entry won't work. You could make the first > item in the list imply that all earlier streams are (or could be) > processed. That's an easy way to allow people to contain the > complexity. > > BTW, if you send GTFO because things are messed up badly enough that > explosions might occur, I don't think you want to keep on keeping on. > In most of the error scenarios, GTFO is followed immediately with a > connection close. Keeping the connection alive is usually reserved > for graceful shutdown. > >
Received on Tuesday, 28 January 2014 00:27:05 UTC