Notes from a discussion about the current state and next steps of generic resource pools and move allocations.

Resource Pools

The main question about generic resources pools at this stage is: How does a compute node determine if it is using shared resources?

The solution here is for the compute node to determine the aggregates which it is associated with and the related resource providers that do not have the uuid of this compute node. From these associated resource providers it is then possible to get inventories for different resource classes.

So, we can determine that if there is, for example, a provider with inventory in a DISK_GB resource class, then this node is using shared disk and things like "local_gb" in the inventory should not be managed.

From an implementation standpoint is it is better that this inventory information is queried at node startup and/or on a periodic task so such that the knowledge of use of shared resources is cached rather than queried every time. We can query all this information in one go, but then make tests against the cached information as needed.

Move allocations

This work can proceed concurrently. I need to read the spec and get up to date on how this will work because at the moment I'm out of sync from focusing solely on the resource pools stuff.

The spec: https://review.openstack.org/#/c/300177/