Silly "workaround" would be blocking TCP RST whie haproxy is restarting so clients that would hit server when nothing is listening would just resend SYN. Or fire 2 instances and use iptables to switch between them (load configuration into 2nd, switch to it, then reload 1st one, and switch back)
2011/10/20 Adam Kocoloski <[email protected]>: > Yes, that works, but there's a window during which neither the old process > nor the new process is listening on the socket. > > Simon Horman implemented a master/worker model to avoid that downtime, but I > guess it's not quite ready to be integrated into the mainline yet. I believe > you can find his work at > http://haproxy.1wt.eu/git?p=haproxy.git;a=shortlog;h=refs/heads/horms-rebased7 > > Best, Adam > > On Oct 19, 2011, at 9:00 PM, John Singleton wrote: > >> Hi Kevin, >> >> I believe you can just send it a -HUP signal. eg: kill -HUP <pid> >> >> Best, >> JLS >> >> On Wed, Oct 19, 2011 at 4:37 PM, Kevin Lindsay <[email protected]> wrote: >> Can HA Proxy make graceful configuration changes? Or does the entire proxy >> need to be restarted leaving space for downtime? >> >> Thanks, >> Kevin >> >> >> >> >> > > >
