[openstack-dev] Dynamically adding Extra Specs

Jay Pipes jaypipes at gmail.com
Fri Jan 22 13:47:49 UTC 2016


Hi! Comments inline...

On 01/22/2016 01:26 AM, Dhvanan Shah wrote:
> Hi,
>
> I had a few queries regarding adding extra specs for VM requests.
>
> According to my understanding if I want to add extra specs to requests
> then I need to change that in different flavors adding those
> capabilities by setting them in the flavors. But if the requests that I
> get have varying values for those extra capabilities then it seems to
> create an issue as the values in the flavors are static. Please correct
> me if I'm wrong.

Right, flavors are static blobs of both resources (amounts of things 
requested by the user) and capabilities (free-form key/value pairs in 
extra_specs). For every variation you may want to offer your cloud 
users, you need to create a new flavor in the system, and your cloud 
users must select that flavor when booting a VM.

Don Dugger from Intel has called this problem "flavor explosion", which 
I have comically re-labeled "the Skittles problem". We have a long-term 
plan to allow for the ability to list a set of capabilities present in 
the deployment and allow a cloud user to "mix and match" resource 
amounts and required capabilities in a more flexible way, but that work 
is likely 9-12 months out in reality.

> So I wanted to know as to how I could dynamically add those extra specs
> best suiting each request. Is there a way of mentioning the extra specs
> everytime I spawn a VM through the nova cli? Setting and unsetting the
> extra specs everytime I spawn VM's according to my need would be quite
> inefficient as it makes changes to the database.

Yes, it is a clunky and inflexible API right now, I agree. The genesis 
of the flavor concept comes from Rackspace Cloud Servers, which modeled 
its flavor concept on the Amazon EC2 instance types concept. This model 
works well for public clouds, because a) they use the flavor as a 
stock-keeping unit (SKU) so therefore this system makes billing easier, 
and b) it reduces the number of hardware configuration options that 
public cloud operators need to provide at scale to users.

If you are interested in this area, follow conversations and blueprints 
with the terms "flavor decomposition", "host capabilities", and 
"resource representation".

Best,
-jay



More information about the OpenStack-dev mailing list