OpenStack Quantum integration feature page for 3.3

[Gary Kotton]

On 04/09/2013 10:16 AM, Mike Kolesnik wrote:
> ----- Original Message -----
>> On 04/08/2013 07:13 PM, Dan Kenigsberg wrote:
>>> On Fri, Mar 22, 2013 at 08:34:14AM +0200, Gary Kotton wrote:
>>>> On 03/21/2013 11:06 PM, Dan Kenigsberg wrote:
>>>>> On Thu, Mar 21, 2013 at 03:53:38PM +0200, Gary Kotton wrote:
>>>>>> On 03/21/2013 03:12 PM, Dan Kenigsberg wrote:
>>>>>>> On Mon, Mar 18, 2013 at 04:23:32AM -0400, Mike Kolesnik wrote:
>>>>>>>> ----- Original Message -----
>>>>>>>>> On Mon, Mar 18, 2013 at 03:20:45AM -0400, Mike Kolesnik wrote:
>>>>>>>>>> Hi all,
>>>>>>>>>>
>>>>>>>>>> The feature page for integrating OpenStack Quantum into oVirt is
>>>>>>>>>> available on the wiki:
>>>>>>>>>> http://www.ovirt.org/Features/Quantum_Integration
>>>>>>>>>>
>>>>>>>>> A quote from "Network discovery":
>>>>>>>>>> Currently, we assume that the networks provided by the provider are
>>>>>>>>>> available on all hosts in the data center...
>>>>>>>>> it is not completely clear who are "we" there. I suppose that you
>>>>>>>>> mean
>>>>>>>>> that Engine takes this assumption, and does not verify that a
>>>>>>>>> specific
>>>>>>>>> host is actually connectable by the network provider. That's not much
>>>>>>>>> worse than what Engine does elsewhere: it does not verify that
>>>>>>>>> network
>>>>>>>>> "green" in one host can actually connect with network "green" on
>>>>>>>>> another
>>>>>>>>> host.
>>>>>>>> "We" is oVirt as a whole.
>>>>>>>>
>>>>>>>> Currently we do know which network is available on which host since
>>>>>>>> the user set it up.
>>>>>>>> In the planned integration we wouldn't know, thus we will consider
>>>>>>>> external network as "provided".
>>>>>>> So *Engine* cannot tell if a host is controllable by the defined
>>>>>>> quantum
>>>>>>> server, and therefore it has to assume that everything is all right,
>>>>>>> putting this burden on the end user.
>>>>>>>
>>>>>>> Garry, does Quantum provide any means to tell - ahead of time - if
>>>>>>> provisioning a virtual network to a specific work is expected to
>>>>>>> succeed? Or, let's say, that the host is utterly misconfigured and is
>>>>>>> unknown to Quantum? Can it at least be done for plugins with host
>>>>>>> agents?
>>>>>> I am not really sure that I understand your question. Can you please
>>>>>> clarify what you mean by "if provisioning a virtual network ... is
>>>>>> expected to succeed".
>>>>>>
>>>>>> Lets take two open source examples:
>>>>>> openvswitch - a integration bridge is provisioned ahed of time, that
>>>>>> is, each host will need to provision an integration bridge on the
>>>>>> OVS. When the agent detects that new port is created on the bridge
>>>>>> then it will configure the relevant VLAN. The Quantum service will
>>>>>> be notified that the port is "UP"
>>>>>> linuxbridge - the nova host that is deplying the VM will create the
>>>>>> networkwork if it does not exist. If this fails the a VM will not be
>>>>>> spawned. Once the network is up the agent will configure the VLAN.
>>>>>> The quantum service will be notified that the port is "UP"
>>>>>>
>>>>>> The assumption is that the hosts will be configured correctly. If
>>>>> I would like to have a clue - ahead of time - if that assumption is
>>>>> valid. The two examples above require the host to notify the Quantum
>>>>> service. But what if the host agent is notifying an utterly different
>>>>> service? Is there some kind of liveliness check between the host and its
>>>>> Quantum service?
>>>> The communication between the host and the quantum service is done
>>>> by a message broker. If the host is updating a different service
>>>> then the port state will not be set as "UP"
>>>> In grizzy a feature was added where one is able to get the state of
>>>> the agents for example:
>>>>
>>>> openstack at dhcp-4-227:~/devstack$ quantum agent-list
>>>> +--------------------------------------+--------------------+---------------------------+-------+----------------+
>>>> | id                                   | agent_type         | host
>>>> | alive | admin_state_up |
>>>> +--------------------------------------+--------------------+---------------------------+-------+----------------+
>>>> | 1ab54438-9338-4050-a818-7faf28d31eb2 | DHCP agent         |
>>>> dhcp-4-227.tlv.redhat.com | :-)   | True           |
>>>> | 396f03eb-b3af-4ddf-b52f-6a0de65b0574 | L3 agent           |
>>>> dhcp-4-227.tlv.redhat.com | :-)   | True           |
>>>> | 876c7474-9e2e-4533-a27e-1de8ac3544c6 | Open vSwitch agent |
>>>> dhcp-4-227.tlv.redhat.com | :-)   | True           |
>>>> +--------------------------------------+--------------------+---------------------------+-------+----------------+
>>>>
>>>> If you get an XXX instead of a :) then the agent is down. The above
>>>> is used for scheduling networks and routers to the dhcp and l3
>>>> agents respectively.
>>>>
>>>> Does this help?
>>> Yes. It seems that this is done at the Quantum server. Is it exposed by
>>> the Quantum API or an extension thereof?
>> This is exposed by a Quantum extension. It is not part of the official
>> API and may not be supported by some plugins.
>> You are nonetheless able to query which extensions are supported by the
>> plugin.
>>
>>> This information is valuable for scheduling VMs, and I'd like to
>>> understand how oVirt can tap into it.
>> If the above extension is supported and the agent on the host is down
>> then the scheduler should exclude that host - if that host is selected
>> then the VM will not have network connectivity.
>>
>>> Also, what happens with agent-less plugins? Can Quantum tell who are the
>>> hosts that it controls?
>> This depends on the extensions supported by the plugin.
>>
>> At the moment there is a patch in progress (Havana) which will have the
>> extension for the port bindings store the host. Once again this may not
>> be supported by some plugins.
> Do you know if it will be supported for Linux Bridge&  OVS plugins?

Yes, this is currently supported by these plugins.

>
>>> Regards,
>>> Dan.
>>