Hi,
2017-11-24 15:52 GMT+01:00 Willy Tarreau <[email protected]>: >> - "timeout http-keep-alive" is not used > > I thought about trying to use it instead of timeout.client but felt > uncertain about this. Maybe it would make more sense. What's your > opinion ? See below. >> Should http-keep-alive be limited to HTTP1? > > I really don't know and as mentionned above I have been hesitating as > well. One point to keep in mind though is that http-keep-alive timeouts > sometimes tend to be extremely short on certain systems to limit the > number of concurrent idle connections. But with H2 we have a single > connection per client, and re-creating it requires some exchanges, so > the incentive for keeping it open longer is real. I suspect that some > users might prefer to keep a short http-keep-alive for HTTP/1 and a > longer one for HTTP/2. But I could be totally wrong. No, I do agree with you on this. It makes sense to keep a HTTP/2 connection open for a longer time than we did with HTTP/1. "timeout client" will certainly do the job for now. I agree this is something that can be changed post release, should the need arise. Maybe adding a doc note regarding the current http2 timeout behavior in the timeout directives (http-keep-alive and client)? I can send a doc patch later today. Regards, Lukas
