How to correctly configure nova-compute to use FlatDHCPManager

Question

I am trying to deploy Openstack using FlatDHCPManager. I am using MAAS+Juju to deploy all of my nodes (HP Moonshot) running trusty.

I want eth1 to be the public interface.

Here are the relevant parts of my configuration yaml:

nova-cloud-controller:
  network-manager: 'FlatDHCPManager'
nova-compute:
  config-flags: 'public_interface=br100'
  flat-interface: eth0
  multi-host: 'yes'
  virt-type: kvm

I have tried setting flat-interface to both eth0 and eth1. I have tried specifying public_interface to be = eth1, br100, and not setting anything at all to take the default value.

Using the config settings shown above, I can SSH into a running VM ONLY from the nova-compute host that the VM is running on. Otherwise, I can't get to it.

Once I SSH into the VM, I can't get out to anything.

I know most things are moving to Neutron, but that's not an option yet for what I am working with. I need to stick with nova-network for a while.

I have seen articles/bug reports and am not sure if the problem is related or not.

Any help is appreciated.

Answer 1

I am using the following configuration now, and it is working:

nova-cloud-controller:
  network-manager: 'FlatDHCPManager'
nova-compute:
  config-flags: 'public_interface=juju-br0'
  flat-interface: eth0
  multi-host: 'yes'
  virt-type: kvm

This is necessary because of the juju-created bridge that is already on eth0, and prevents another bridge from being created.