When changing back end config, and reloading haproxy using “-sf” it often seems to take a long time (~120 sec)for the old process to go away. During that period, refreshing an existing browser page returns 503 because it’s presumably connected to the old haproxy process, which no longer has a valid back end. I’m guessing this is because the TCP connection is still open due to HTTP keep alive?
Is there a way to specify a timeout in the config, so a reload kills the old process to prevent them hanging around for more than a few sec? I know there is an -st option, but that is quite brutal. A gracefully reload with a cutoff would be the best solution I think?