[Openstack-operators] Live migration neutron issues

Adam Dibiase adibiase at digiumcloud.com
Thu Jan 28 15:17:55 UTC 2016


I have an update to my migration issue and thought I would share. It seems
this happens on instances that are CentOS 5.x. It does not happen on
instances that are CentOS 6.x. Does anyone know what neutron is sending to
the guest to start the network post migration and why it would not work in
CentOS 5.x?

Thanks,

Adam




On Thu, Jan 7, 2016 at 4:00 PM, Adam Dibiase <adibiase at digiumcloud.com>
wrote:

> Yes. I have tried hitting it from an instance on the same L2 network with
> the same result.
>
> Just to make things interesting, I cannot be sure yet, but I think I am
> seeing a correlation between instances that have been created in Liberty
> letting nova create the neutron port (always seems to migrate successfully)
> and instances that I have attached a manually created port (always seem to
> fail).
>
>
>
> Thanks,
>
> Adam
>
>
>
>
> On Thu, Jan 7, 2016 at 3:36 PM, Mathieu Gagné <mgagne at internap.com> wrote:
>
>> On 2016-01-07 3:28 PM, Adam Dibiase wrote:
>> > Greetings,
>> >
>> > I am having connectivity issues to instances after live migrating
>> > between compute nodes. Scenario is as follows:
>> >
>> >   * Start pinging instance IP
>> >   * Send nova live-migration <instancename> <computenode>
>> >   * Once nova reports the pause/resume on the new compute node after
>> >     migration, ping replies stop
>> >   * Instance migrates successfully and I can hit it via VNC console, but
>> >     no network connectivity
>> >   * If I live migrate it back to losing compute node, it will still NOT
>> >     restore network connectivity
>> >   * The only way to restore network is to nova stop/start or nova reboot
>> >     instance
>> >   * When the instance comes back up, I am able to hit it again on the
>> >     new compute node
>> >
>> > I have checked to see if the neutron port is showing the new
>> > binding:host_id, ovs shows the correct port for instance on new compute
>> > node, and no other errors present in the logs.
>> >
>> > I am running Liberty with cinder FYI. Anyone else having similar issues?
>> >
>>
>> Have you tried starting another instance on the same L2 network and ping
>> from there once it's migrated? This is to make sure it's not an issue
>> with ARP caching on a device between you and the instance.
>>
>> --
>> Mathieu
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20160128/d46eebd4/attachment.html>


More information about the OpenStack-operators mailing list