- From: Ryan Hamilton <rch@google.com>
- Date: Fri, 3 Apr 2015 14:17:56 -0700
- To: Patrick McManus <mcmanus@ducksong.com>
- Cc: Martin Thomson <martin.thomson@gmail.com>, "ietf-http-wg@w3.org" <ietf-http-wg@w3.org>
- Message-ID: <CAJ_4DfR-kpiGmdDQXRG85E46jjNtYP82L7NCqWk8gXHy4=wYTw@mail.gmail.com>
On Fri, Apr 3, 2015 at 2:04 PM, Patrick McManus <mcmanus@ducksong.com> wrote: > > On Fri, Apr 3, 2015 at 4:50 PM, Ryan Hamilton <rch@google.com> wrote: > >> Consider the following scenario. There are two servers, >> internal.example.com and external.example.com. Inside the enterprise >> access to resources outside the firewall must go through a proxy, whereas >> resource inside the firewall can go direct. That would lead to a .pac file >> like: >> >> function FindProxyForURL(url, host) { >> >> if (host == "internal.example.com") { >> return "DIRECT"; >> } >> return "PROXY proxy.example.com"; >> } >> >> >> If Alt-Svc for internal says, "Alt-Svc: h2="external.exmple.com:443", >> then the proxy will say, "Sure, go direct to >> http://internal.example.com/" but then the browser will connect to >> external.example.com:443 and will avoid the proxy and the request will >> hang. This seems a bit unfortunate, but it's not clear that the alternative >> is much better, so I'm happy to do this, if that's the consensus of the >> group. >> > > > The browser is supposed to validate the alternate service (among other > things by checking that it has a cert valid for the origin) before using it > and fall back if it is unavailable. So it seems that > external.example.com:443 could not be validated in your example and if > perhaps that validation were stale, it should get automagically cleaned up. > Agreed! When I said, "the request will hang", I meant to say, "the connection will hang". You're totally right that this won't be a user facing problem. We won't use the alternative-service but it's not the end of the world. I would be very cautious about ever changing url based on alt-svc.. alt-svc > does not change origins or urls. If it did, it would get very hard to > reason about. but supplying the alt information explicitly seems a lot more > interesting. > Agreed. > Somebody should really write up a I-D for a modern PAC (that could be me, > but realistically a lot of other things would have to come off the todo > list first.) > SGTM. Especially the part about you doing the work :>
Received on Friday, 3 April 2015 21:18:24 UTC