[openstack-dev] Combination of ComputeCapabilitiesFilter and AggregateInstanceExtraSpecsFilter
unicell at gmail.com
Sat Jul 6 00:14:52 UTC 2013
Should ComputeCapabilitiesFilter also be restricted to use scoped
format only? Currently it recognize and compare BOTH scoped and
non-scoped key, which is causing the conflict.
I've already submitted a bug and patch review before.
On Sat, Jul 6, 2013 at 2:04 AM, Russell Bryant <rbryant at redhat.com> wrote:
> On 07/05/2013 12:42 PM, Jérôme Gallard wrote:
>> Hi all,
>> I'm trying to combine ComputeCapabilitiesFilter and
>> AggregateInstanceExtraSpecsFilter. However I probably missed
>> something, because it does not work :-)
>> Both filters are activated with the following order:
>> ComputeCapabilitiesFilter, AggregateInstanceExtraSpecsFilter.
>> I created a flavor with the following extra_spec:
>> * capabilities:hypervisor_hostname=node1
>> * class=good
>> I created an aggregate containing node1 with an extra_spec:
>> * class=good
>> When I start a new instance with the previously created flavor, the
>> ComputeCapabilitiesFilter can't find an available node. I put some
>> debug inside the filter. From my understanding, it seems that,
>> ComputeCapabilitiesFilter manage to find the first spec
>> "capabilities:hypervisor_hostname=node1" into the list of metadata
>> provided by the host node1 : the first iteration of the loop is OK.
>> Then this filter continues with the "class=good" spec and, of course,
>> it fails and the filter returns that there is no available host.
>> Do you have an idea about what I'm missing? How to tell to
>> ComputeCapabilitiesFilter that the "class" key is not for it?
>> I read the detailed documentation about filter_scheduler (
>> ). But I didn't manage to solve the issue.
> The AggregateInstanceExtraSpecs filter needs to have support added for
> scoped extra specs. That way you can specify something like
> 'aggregate_capabilities:class=good', and the other filter will ignore it.
> I'll fix it up. It should be pretty easy.
> Russell Bryant
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
More information about the OpenStack-dev