[openstack-dev] [Notifications] Glance, Quantum et al.

Gary Kotton gkotton at redhat.com
Thu Nov 29 16:41:33 UTC 2012


On 11/29/2012 06:33 PM, Russell Bryant wrote:
> On 11/29/2012 11:13 AM, Gary Kotton wrote:
>> On 11/29/2012 06:09 PM, Brian Waldon wrote:
>>> On Nov 29, 2012, at 7:56 AM, Gary Kotton wrote:
>>>
>>>> Hi,
>>>> Due to the patch https://github.com/openstack/glance/commit/3d532de
>>>> we have a problem running Quantum and Glance on devstack. This just
>>>> opens a pandoras box of problems. Glance sets the queue as durable,
>>>> Quantum uses the default value which is false - this causes an
>>>> exception "AMQPChannelException: (406, u"PRECONDITION_FAILED -
>>>> parameters for queue 'notifications.info' in vhost '/' not
>>>> equivalent", (50, 10)".
>>>> The underlying problem is the way in which the notifications are
>>>> implemented. I am not really sure if and how this can be addressed as
>>>> I am not familiar with the code.
>>> Glance itself will default to non-durable queues (config opt
>>> rabbit_durable_queues), but there's code in Devstack that will attempt
>>> to match Glance's durability to that of an existing queue. Can you
>>> verify that the relevant code in Devstack is functioning correctly?
>>> See 'rabbit_durable_queues()' in lib/glance.
>> Yes, this works correctly. The problem is that Quantum works with the
>> default (which is False). Hence the exception from the broker. This is
>> just one problem. At the moment the Quantum DHCP agent is built upon the
>> notification mechanism. If I understand correctly if Glance and Quantum
>> are sending to the same queue then the listener(s) will read the
>> messages in a round robin fashion. This may break the functionality of
>> the DHCP agent
> Why are they sending to the same queue?  Shouldn't quantum be using a
> queue on the quantum exchange and glance a queue on the glance exchange?
>   AFAIK, they should *not* be using the same queue by the intended design.
>

I may be mistaken here but I think that for the notifications the 
exchange may be ignored.




More information about the OpenStack-dev mailing list