[Rdo-list] VM instance will not start

brian lee brian at brianlee.org
Fri Jul 24 17:59:01 UTC 2015


Thanks for the follow up. I think I am getting it. When you say multi host,
you are refereing to the setting in nova.conf multi_host, correct?

If I want the traffic to be routed through the controller, I should set
that to false, and not install the nova-network on the compute hosts.

--Brian

On Fri, Jul 24, 2015 at 12:26 PM, Brent Eagles <beagles at redhat.com> wrote:

> On Thu, Jul 23, 2015 at 04:26:08PM -0500, brian lee wrote:
> > So I have made headway on this problem. It was related to my networking.
> In
> > order to get nova networking working you have to install the
> > openstack-nova-network and openstack-nova-api packages on your compute
> > nodes as well. You did not have to do this in Icehouse.
> >
> > Once that is installed, you then need to configure the nova.conf per the
> > doc:
> >
> http://docs.openstack.org/kilo/install-guide/install/yum/content/nova-networking-compute-node.html
>
> Note that if you are using multi-host networks then each compute
> node is also a network controller, so nova-network and nova-api will be
> required on each compute node.
>
> > Once I have done that I am now able to get the instances started. On the
> > compute node it does create the br100 bridge device. But it does not
> create
> > it on the controller.
>
> You can check if openstack-nova-network running on the controller - but
> if you are using multi-host networking this is probably irrelevant.
>
> > Now I am stuck where I can get the instance up, but I can not ping it
> from
> > the conrtoller/outside network. Any idea what needs to be done to get the
> > controller to start its bridge so they can talk together?
> >
> > --Brian
>
> IIRC, if you are using multi-host networks you need to keep in mind that
> while each compute node is a network controller - the reverse is also
> true. i.e. a node is only a network controller for an instance's tenant
> network IF that node has an instance for that tenant running on it. If
> there isn't an instance for a particular tenant on a given node, there
> may be no bridge for that tenant network, etc. on it. This has to do
> with how the networks are provisioned - the bridges are setup where a
> tenant network is required, i.e. where an instance has been booted. Also
> of course, there is no standalone network-controller.
>
> To get access to your guest, try going through the multi-host node
> instead of the "controller" (which isn't a network controller in this
> case).
>
> If you *don't* use multi-host then you the network service should only
> be required on one host.
>
> Cheers,
>
> Brent
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rdoproject.org/pipermail/dev/attachments/20150724/9f635b38/attachment.html>


More information about the dev mailing list