Floating ip get set on the wrong physical server

Nux! nux at li.nux.ro
Wed Apr 3 11:48:58 UTC 2013


On 03.04.2013 12:42, Sandro "red" Mathys wrote:
> On Wed, Apr 3, 2013 at 1:40 PM, Gary Kotton <gkotton at redhat.com> 
> wrote:
> 
>> On 04/03/2013 02:36 PM, Nux! wrote:
>> 
>>> On 03.04.2013 12:27, Gary Kotton wrote:
>>> 
>>>> On 04/03/2013 01:35 PM, Nux! wrote:
>>>> 
>>>>> Hello,
>>>>> 
>>>>> I'm currently testing Grizzly on EL6 from
>>>>> http://repos.fedorapeople.org/**repos/openstack/openstack-**
>>>>> grizzly/epel-6/<http://repos.fedorapeople.org/repos/openstack/openstack-grizzly/epel-6/>
>>>>> I'm using packstack to deploy it on 3 physical machines, one of 
>>>>> them
>>>>> acts also as controller, the other 2 are just nova-compute 
>>>>> instances.
>>>>> The problem starts when adding a floating IP to an instance, in my 
>>>>> case
>>>>> the floating (public) IP gets set on the controller, whereas the 
>>>>> instance
>>>>> is running on another physical server...
>>>>> 
>>>>> Any ideas?
>>>>> 
>>>> 
>>>> On which host is the nova-network running? This is what performs 
>>>> the
>>>> SNAT (i.e. the floating IP support). In nova you can run this on 
>>>> each
>>>> compute node for HA and scalability.
>>>> 
>>> 
>>> Hi Gary!
>>> 
>>> Openstack-nova-network ran only on the controller initially (this is 
>>> how
>>> packstack set it up). I did try to set up more IPs in the answers 
>>> file
>>> afterwards, but I got this:
>>> "Parameter CONFIG_NOVA_NETWORK_HOST failed validation: Given host is 
>>> not
>>> in IP address format: 
>>> 192.168.203.66,192.168.203.67,**192.168.203.68"
>>> I take it it doesn't like comma separated stuff (though it worked 
>>> for the
>>> COMPUTE hosts).
>>> 
>> 
>> This may be a bug (sorry I am not familiar with it)
> 
> 
> Not a bug per se (it's not supposed to work) but you can still open a 
> rhbz
> to request the functionality.

I sure will, not only is it not scalable to have all traffic routed 
through just one "node", not to mention HA issues, but in my case while 
incoming traffic gets NAT-ed through the nodes to the specific VM, 
outgoing traffic doesn't go anywhere. There is no SNAT rules on the node 
running the actual VM.

I have installed openstack-nova-network manually, but have not modified 
anything in nova.conf, I'll check this shortly.

Thanks Gary & Sandro!

Lucian

-- 
Sent from the Delta quadrant using Borg technology!

Nux!
www.nux.ro


More information about the cloud mailing list