can nova-manage be executed remotely ?

I'm not sure about all of the interactions with quantum, but in general nova-manage shouldn't be trying to kill dnsmasq at all. The network commands in nova-manage should only do db only setup. Ultimately, yes nova-manage should be replaced with something that is only using apis, but deleting in use networks is not something that is supported anywhere yet.

Vish

On Mar 2, 2012, at 9:25 AM, JC Martin wrote:

> New question #189460 on OpenStack Compute (nova):
> https://answers.launchpad.net/nova/+question/189460
>
> In the following setup :
> node 1: Nova Scheduler
> node 2: Nova Network, Nova API, Quantum
>
> If I execute nova-manage network delete on node 1, it tries to kill dnsmasq on node 1, but dnsmasq is on node 2.
> Creation of the network worked as expected (from node 1, to node 2)
> This is leading to strange situations.
>
> So the questions are really :
> 1/ is this setup supported at all ?
> 2/ should nova-manage always use APIs and not execute commands locally ?
>
> I was tempted to file a bug because of point 2.
>
> Opinion ?
>
> JC
>
>
> --
> You received this question notification because you are a member of Nova
> Core, which is an answer contact for OpenStack Compute (nova).

The offending code is in :
nova/network/quantum/manager.py :
       # Get rid of dnsmasq
        if FLAGS.quantum_use_dhcp:
            dev = self._generate_gw_dev(quantum_net_id)
            self.driver.kill_dhcp(dev)

However, quantum (and nova-network) is not on the machine where nova-manage is executed.
This section of the code should be moved out to the quantum manager, in the plugin implementing the delete_network function.

If this is not done, the removal of a network fails to kill the dnsmasq processes.
There are some other issues I've seen that may be related to this too (remaining bridges/interfaces)

JC