2013 RHEL Server subscription should have a guest_limit attribute with value “4” This attribute is important because it allows us differentiate between 2013 subscriptions with a virt guest limit, and only rhel server subscriptions. Additionally, this gives us a new attribute that can be modified for future pricing changes.
RHEV and OST subscriptions should have a guest_limit attribute with value “-1” which indicates unlimited virtual guests are allowed.
The guest_limit value is global, we will use the highest number (-1 being very high) and apply it to every subscription that has a guest_limit value when we calculate compliance. If there are more guests than supported, every subscription with guest_limit should become “insufficient.” There will be no changes to the number of virtual subpools created by datacenter subscriptions
This requires new attributes on (hypervisor) subscriptions, but allows us to futureproof. Without the new attribute, It will become increasingly difficult to support this in the future. It gives the added benefit of the possibility of a medium-density hypervisor, or a hypervisor subscription that doesn’t increase the limit. This gives us the ability to create hypervisors with different limits, so we won’t have to worry about product ID logic and installed product checks.
When compliance starts, loop through all entitlements to find the maximum guest_limit attribute. We can probably preserve the current structure of stackable attribute checking if we add a new hook for the starting value of the “currentStackValue” stack tracker attribute, and no-op when we update the attribute from a new pool.
Pre_entitlement should not check the virt max attribute, because like stackable subscriptions, it depends heavily upon other subscriptions. It may be worth rethinking pre_entitlement rules in order to support subscriptions whose compliance relies upon other attached subscriptions, as that case is becoming increasingly more common.
This attribute should be a “stackable attribute” in the code, even though it doesn’t actually stack. That way a bad subscription will invalidate an entire stack (just like Arch).
Autobind probably won’t fully support this feature because it would need to take into consideration that an entitlement or stack will be “healed” by adding another, unrelated, subscription that doesn’t necessarily provide any products. This is acceptable because if there’s no subscription already attached, the consumer probably isn’t actually using a hypervisor that supports more than 4 guests.
It looks like it’s only immediately important to report active/inactive on the kvm hypervisor. It’s probably a good idea to hold hypervisor-type in the GuestID class, in addition to isActive. We can check for the guest_limit capability before sending the extra information to avoid breaking old candlepins. We will default to the current behavior when the capability is not found.