I think this needs to be fixed in the Ubuntu packaging. Neutron-ovs-cleanup
needs to run after openvswitch-switch starts and before nova-compute starts.
But care needs to be taken for the case where neutron-dhcp-agent and/or
neutron-l3-agent also runs on a node with nova-compute. Cleanup should run
before these start too. It's easy to do this with the system V init system, but
I'm not sure how with upstart.
Re, Darragh.
On Thursday, 21 November 2013, 7:50, Jian Wen <[email protected]> wrote:
It is https://bugs.launchpad.net/neutron/+bug/1091605
>
>
>The patch fixes the above bug introduces another bug:
>https://bugzilla.redhat.com/show_bug.cgi?id=1010941
>
>
>
>
>On Thu, Nov 21, 2013 at 2:18 PM, Jean-Daniel BUSSY <[email protected]>
>wrote:
>
>Nice! Thanks Thiago for the details.
>>Looks like this bug should be reported.
>>
>>
>>Cheers
>>
>>
>>
>>BUSSY Jean-Daniel
>>Cloud Engineer | GREE
>>Mobile: +81 090-3317-1337
>>Email: [email protected]
>>
>>
>>
>>On Tue, Nov 19, 2013 at 7:01 PM, Darragh O'Reilly
>><[email protected]> wrote:
>>
>>Hi Thiago,
>>>
>>>I did a qucik test with Havana on Ubuntu 12.04 with QEMU, OVS 1.10.2 and the
>>>OVS hybrid driver. Started an instance, rebooted the compute node, and the
>>>instance goes to status STOPPED. Then ran 'nova start instance-name', and it
>>>starts ok, but DHCP does not work.
>>>
>>>In the logs I see OVS starting first and recreating the qvoxxxxxx-xx port
>>>and interface from its database. Then nova-compute starts and creates the
>>>qbr Linux bridge and qvb interface, but it is not recreating the qvo port
>>>and interface because they already exist. Everything looks ok, and the DHCP
>>>requests are getting to qvo, but they are not getting into br-int.
>>>
>>>It seems the qvo interface needs to be created by Nova (using the ip
>>>command) before the OVS port is created with the same name. Otherwise OVS
>>>does not seem to enslave the interface.
>>>
>>>I added this upstart script to run cleanup which will delete the ports on
>>>br-int before nova-compute and the OVS agent start:
>>>
>>>root@compute1:~# cat /etc/init/neutron-ovs-cleanup.conf
>>>start on starting nova-compute or neutron-plugin-openvswitch-agent
>>>script
>>> /usr/bin/neutron-ovs-cleanup
>>>end script
>>>
>>>Now nova-compute has to recreate everything, and the instance is getting an
>>>IP with DHCP. Maybe you can try this out in your test environment.
>>>
>>>Re, Darragh.
>>>
>>>
>>>
>>>
>>>>Guys,
>>>>
>>>>My Havana (Ubuntu based) Compute Node was restarted and lots of Instances
>>>>does not get an IP anymore.
>>>>
>>>>Tips?!
>>>>
>>>
>>>>It is ramdom, I mean, some instances of this same compute node are normal,
>>>>while others have no IP.
>>>>
>>>
>>>>I really need help here because my client's web site is completely off line
>>>>now...
>>>>
>>>>I'm using Per-Tenant router with private networks + VXLAN.
>>>>
>>>>Tks!
>>>>Thiago
>>>
>>>
>>>_______________________________________________
>>>Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>>>Post to : [email protected]
>>>Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>>>
>>
>>_______________________________________________
>>Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>>Post to : [email protected]
>>Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>>
>>
>>
>
>
>
>
>--
>
>Cheers,
>Jian
>
>
_______________________________________________
Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
Post to : [email protected]
Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
