Hi, On Mon, Oct 15, 2018 at 11:53 AM Lukas Tribus <[email protected]> wrote:
> Hello, > > > On Sat, 13 Oct 2018 at 10:34, Sébastien Kurtzemann <[email protected]> wrote: > > > > Hi, > > > > I’ve got a question about haproxy "seamless reload" : when this > > operation is perform does all backend servers connections be reset ? > > No. Only *restart* closes existing front and backend connections. > Reload (both seamless and regular) closes them gracefully, so no > request is lost. > Okay. I think I confound connections and servers sessions... :( Are sessions purged after a seamless reload ? So is there a way to keep sessions states or to configure haproxy TCP backends to not send a request to a server that is currently handling a connection ? > > > > Our use case: > > - we use Voyager Ingress for Kubernetes > > (https://appscode.com/products/voyager/) this last use haproxy to > > perform balancing > > - we have configure each servers backend with "maxconn 1" (we only > > what one connection to be handle by one pod) > > - when a new pod is schedule, Voyager update backend servers then do a > > haproxy seamless reload which cause that we lost previous connections > > states for servers > > That's not expected. How exactly does Voyager reload haproxy? > Under the hood Voyager launch the following command to reload haproxy: haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -x /var/run/ haproxy.sock -sf $(cat /var/run/haproxy.pid) > > Regards, > Lukas > Regards, Sébastien
