Hi Vincent, On Fri, Mar 07, 2014 at 09:10:21AM +0100, Vincent Bernat wrote: > ??? 6 mars 2014 16:15 CET, Jonathan Matthews <cont...@jpluscplusm.com> : > > > 1) On restart/reload/disabled-server-now-enabled-via-admin-interface, > > haproxy considers a server to be 1 health check away from going down, > > but considers it *initially* up. > > On reload, haproxy could wait for a whole round of healthcheck before > making the new process listen on the network instead of the old one.
This has already been proposed in the past. Unfortunately we can't do that, because the poll loop starts after the fork which starts after the new process has confirmed it could bind. In practice we'd need almost the ability to run some network loop before forking, and I really fail to see how we could do that without changing a bunch of things in the current architecture. Willy
