In my newly deployed openstack environment, I noticed that CEPH was failing within the 3 LXC containers.
016-05-03 02:18:38 INFO juju-log Unknown hook update-status - skipping.
Reading package lists... Donete-status
Building dependency tree -status
Reading state information... Donetatus
2016-05-03 02:18:40 INFO update-status sudo: unable to resolve host juju-machine-1-lxc-3
Any ideas? Is it standard that the physical nodes wouldn't be able to DNS the containers? Seems highly unlikely, and also, when I deployed the same setup before, I didn't have any issue. I'm guessing the DNS is why I am getting the HEALTH_WARN error.
Update: I've discovered the DNS error isn't actually why CEPH is failing, so I'm interested in both. Should DNS entries be added to Bind for the containers, not just physical hosts? Any idea on why Ceph might be failing?
Update 2: Ceph is failing because the placement groups exceed the PG max set by ceph. The fix is increase the OSDs without increasing the placement groups. I've added an answer that indicates the bug. The suggestion at the time of this edit, is to create some tuning in the juju charm to 'calibrate' settings after things are deployed.
Update 3: Adding OSD's does decrease the placement groups count against the maximum PG limit warning value.
