Open Stack

Hi,

On 2014-07-09 7:29 PM, Jonathan Proulx wrote:
> I can't seem to find an understandable explaination of how to
> translate my existing OVS config to ML2, so I suspect this is a simple
> (to someone) config issue.
>
> following upgrade steps at:
> http://docs.openstack.org/openstack-ops/content/upgrades_havana-icehouse-ubuntu.html
>
> This is mostly copy pasta from the ovs plugin ini with a few
> suggestions I've found online so is highly suspect, config on
> controller/network node is identical modulo local_ip
>
> anyone see what I got wrong there?
>

Here is what I found out when trying to migrate from linuxbridge plugin 
to ML2+linuxbridge.

The ml2_conf.ini config file contains everything needed by the core 
plugin running in neutron-server. This includes vlan ranges, tunnel ids, 
etc.

On the other hand, the linuxbridge agent running on the compute nodes 
reads 2 config files: neutron.conf and linuxbridge_conf.ini.

People thought: "This is wrong! I'm running ML2, it should read my 
ml2_conf.ini!".

And they went ahead and created a bunch of symlinks from 
linuxbridge_conf.ini to ml2_conf.ini in a hope to make it works. And 
they added linuxbridge agent specific configs in ml2_conf.ini and it 
worked. They were happy and slept well at night.

What I finally understood is that you don't need to create such symlinks 
from linuxbridge_conf.ini to ml2_conf.ini. The agent does NOT need any 
information found in ml2_conf.ini to run on the compute nodes. It is 
quite happy to run with neutron.conf and linuxbridge_conf.ini alone.

What I did on the compute node is the following:
- neutron.conf: core_plugin = ml2
- linuxbridge_conf.ini: Everything needed by the agent: vlans, 
linux_bridge, vxlan, agent and securitygroup sections.
- ml2_conf.ini: Nothing!

I then started the agent and everything went well.

I understanding is that the same should apply to ML2+OVS mechanism.

Here is a related review in puppet-neutron:
https://review.openstack.org/#/c/106144/

-- 
Mathieu