On 05/28/2018 05:00 PM, Chris Lamb wrote: > Hi Thomas, > >> it looks like I had this in my /etc/sysctl.conf: >> >> net.ipv6.conf.all.disable_ipv6 = 1 >> net.ipv6.conf.default.disable_ipv6 = 1 >> net.ipv6.conf.lo.disable_ipv6 = 1 > > Great debugging work! What led you to that? (Logs?)
A bit of strace foo! :) > But what should it do if it can't? IMO, it should bind on everything configured if it's there. If not, then the daemon should still start, and continue to attempt binding until the configured IP is available. Indeed, network configuration shouldn't be expected as a static thing. What if for some reason a configured IP isn't available yet at the time of server boot? This happens often with some nasty switched like Cisco Catalyst if configured without port-fast. > I think it's the expected behaviour > that if you configured a daemon to listen on an interface and it could > not do so, then the daemon should fail to start. (And failing to start > on Debian necessarily implies that it should fail an installation...) Right, but the default config on Debian doesn't work for everyone, and in my case, it even completely breaks the OpenStack CI. Probably then, the default should be 127.0.0.0 without ::1 if that can't always work. Cheers, Thomas Goirand (zigo)
