container addressability When using openstack as provider lxc containers get a hardcoded ip

[Gema Gomez]

I am trying to deploy automatically 3 ubuntu servers with 3 lxc containers on each of the nodes:

    ubuntu:
      branch: lp:charms/trusty/ubuntu
      num_units: 3
      constraints: mem=2G
    rabbitmq-server:
      branch: lp:charms/trusty/rabbitmq-server
      num_units: 3
      constraints: mem=1G
      to: [ "lxc:ubuntu=0", "lxc:ubuntu=1", "lxc:ubuntu=2" ]

See full environment: http://bazaar.launchpad.net/~openstack-charm-testers/+junk/tempest/view/head:/rabbitHA.yaml

When deployed, the lxc containers are forced into:

LXC_DHCP_RANGE="10.0.3.2,10.0.3.254,infinite"

... instead of being plugged into the network of the deployment.

The problem is that these lxcs are not accessible. I was expecting the behaviour to be exactly like the maas provider.

This happens on OpenStack provider and in our opinion should be working on the same way MAAS provider is working.

[Jorge Niedbalski (niedbalski)]

The workaround used to fix this on OpenStack provider.

1. Add a new interface to your VM:

vi /etc/network/interfaces.d/juju-br0.cfg

# The primary network interface
auto juju-br0
iface juju-br0 inet dhcp
bridge_ports eth0

2. In /var/lib/lxc/juju-machine-23-lxc-0/config:

lxc.network.link = juju-br0

3. Go to your dhcp namespace (sudo ip netns exec
qdhcp-b9c1579d-79e2-4215-8ba0-ee4649d9c38f bash on this machine):

3.1 Add the new entries to
/var/lib/neutron/dhcp/b9c1579d-79e2-4215-8ba0-ee4649d9c38f/hosts

3.2 Restart the dnsmasq process with:
kill -HUP $PID

3.3 Restart the ubuntu hosts that contain the lxcs

[Dimiter Naydenov (dimitern)]

Addressable containers on OpenStack are not yet supported, so the workaround you're describing is a possible way to get close to what juju does on MAAS (juju-br0 is not even created on OpenStack).

We do plan to have the same level of support for addressable containers in OpenStack, as in MAAS and EC2 (currently only with the address-allocation feature flag set at bootstrap), but it's not considered a priority vs. AWS VPC support for example.

[Tim Penhey (thumper)]

This work is targetted for 2.3. A milestone will be set again once the work starts.