- From: Willy Tarreau <w@1wt.eu>
- Date: Tue, 9 Sep 2014 10:27:16 +0200
- To: Poul-Henning Kamp <phk@phk.freebsd.dk>
- Cc: HTTP Working Group <ietf-http-wg@w3.org>
On Tue, Sep 09, 2014 at 08:24:56AM +0000, Poul-Henning Kamp wrote: > -------- > In message <20140909081126.GA4299@1wt.eu>, Willy Tarreau writes: > >On Tue, Sep 09, 2014 at 06:23:25AM +0000, Poul-Henning Kamp wrote: > >> -------- > >> In message <20140909055008.GB3403@1wt.eu>, Willy Tarreau writes: > >> > >> >> Managing TCP connections is out-of-band in HTTP/2.0, so sending > >> >> "Connection: close" or "Connection: keepalive" inside a HTTP/2.0 > >> >> multiplex doesn't make any sense, should not happen and I think > >> >> we should make that an explicit SHALL NOT. > >> >> > >> >> The general way to do that would be to make it a stream-error in > >> >> HTTP/2.0 if Connection: tries to make a non-existent header hop-by-hop. > >> > > >> >I don't think this extra check would provide any value, to be honnest. > >> > >> It would catch cases where people have messed up. > > > >Yes, but it's contrary to the principle of being liberal about what > >you accept. > > Can we at least implement the "be conservative" part and tell people > that "Connection: close" will not work ? Yes I think that's quite reasonable. I'd even say that the SHOULD NOT emit connection: close nor connection: keep-alive. This can easily happen in 1.x->2.0 gateways/proxies. Willy
Received on Tuesday, 9 September 2014 08:27:41 UTC