So, I've got a region controller, and two cluster controllers. (All separate machines.) Cluster #1 -- the local cluster, on a 172.23.x.0/23 subnet -- works great. The remote cluster, on 10.2.0.x/16, doesn't work so hot.
- Systems PXE boot, provision, and commission great.
- Juju sees the new system(s) just fine.
- At some point during the process -- as is standard for MAAS -- the IP address changes from the "temporary" IP used for new devices, to the "permanent" address.
- The trouble starts when I do a "juju add-unit nova-compute". The config files that get created on the new compute node reflect the original IP address, not the one that the system is currently using. E.g., "ifconfig" shows an IP of 10.2.0.20, but here's what a grep in /etc/ returns:
rgrep "10\.\." * 2> /dev/null | grep -v ":#" neutron/plugins/ml2/ml2_conf.ini:local_ip = 10.2.1.1 nova/nova.conf:my_ip = 10.2.1.1 nova/nova.conf:vncserver_proxyclient_address = 10.2.1.1
I've deleted and re-created the network twice, to no avail (including using different ranges). On the verge of deleting and re-creating the cluster controller, etc., but wanted to post here in case anyone had any ideas. Oh -- last thing: this is 1.9.1+bzr4543-0ubuntu1~trusty1 on Ubuntu 14.04.
Any pointers gratefully accepted...
UPDATE: deleted and re-created cluster controller; still the same deal. Interestingly, if I attempt to deploy to an LXC, THOSE config files are fine (though LXC fails as a compute node, for reasons I assume are unrelated). Also, if I hardcode the correct values into the two files, everything works great.