DNS tries to resolve kube-spawn... to external + root nameserver

[donbowman]

Somewhere there is a hardcoded 8.8.8.8 (because this is not my machines DNS).

I am seeing an A-record lookup for kube-spawn-...worker (no domain). This goes to 8.8.8.8. This is then forwarded to a.root-servers.net, which responds NXDOMAIN this is repeated for AAAA

to avoid this I think the master should maybe have /etc/hosts filled in w/ the workers name + IP.

$ machinectl list
MACHINE                          CLASS     SERVICE        OS      VERSION  ADDRESSES
kube-spawn-flannel-master-wm9g3x container systemd-nspawn flatcar 1828.0.0 10.22.0.149...
kube-spawn-flannel-worker-matpk7 container systemd-nspawn flatcar 1828.0.0 10.22.0.148...
kube-spawn-flannel-worker-o5xape container systemd-nspawn flatcar 1828.0.0 10.22.0.146...
kube-spawn-flannel-worker-zdt5mu container systemd-nspawn flatcar 1828.0.0 10.22.0.147...

4 machines listed.
$ sudo machinectl shell kube-spawn-flannel-master-wm9g3x
Connected to machine kube-spawn-flannel-master-wm9g3x. Press ^] three times within 1s to exit session.
Update Strategy: No Reboots
Failed Units: 1
  systemd-networkd-wait-online.service
kube-spawn-flannel-master-wm9g3x ~ # host kube-spawn-flannel-worker-matpk7
Host kube-spawn-flannel-worker-matpk7 not found: 3(NXDOMAIN)
kube-spawn-flannel-master-wm9g3x ~ # logout
Connection to machine kube-spawn-flannel-master-wm9g3x terminated.

shows the problem.

perhaps the --resolv-conf= switch to systemd-nspawn

[dongsupark]

Oh, I missed this issue. Yeah you're right. We have a hard-coded DNS server: https://github.com/kinvolk/kube-spawn/blob/v0.3.0/pkg/cluster/cluster.go#L284 I agree, at least we should provided ways for users to configure the DNS server, via cmdline parameter or a config file.

[donbowman]

the issue is the node-names we create. they need to be locally resolvable. Since we made them, its insufficient to allow local dns server. otherwise we are hammering the root name servers (a.root-servers.net etc) w/ unresolvable.