I'm running HAProxy in HTTP mode, behind another load balancer which is not an HTTP load balancer but a TCP load balancer. A CDN may connect and start an HTTP session in keepalive mode and send a large number of requests, which is fine by me, but in the interest of even load balancing and orderly shutdowns, I would like to place a limit on the duration of a connection. It should be a soft limit — the timer begins when the connection is opened; if it expires while waiting for a request, then the connection is closed without error (as with "timeout http-keep-alive"), but if it expires while a request is being handled, the connection is closed after handling that request instead.
I could of course just disable keepalive, but I don't want the overhead of a new TCP connection for every HTTP request; just every now and then. Is there a way to do what I want already? From the documentation, I don't think so, but it's always possible that I missed something. If not, please consider this a feature request. Thanks, Andrew
