Roles with quota assigned can "game" the system to receive excessive resources.

The current implementation of quota allocation attempts to satisfy each resource quota for a role, but in doing so can far exceed the quota assigned to the role.

For example, if a role has quota for [30,20,10], it can consume up to: [∞, ∞, 10] or [∞, 20, ∞] or [30, ∞, ∞] as only once each resource in the quota vector is satisfied do we stop allocating agent's resources to the role!

As a first step for preventing gaming, we could consider quota satisfied once any of the resources in the vector has quota satisfied. This approach works reasonably well for resources that are required and are present on every agent (cpus, mem, disk). However, it doesn't work well for resources that are optional / only present on some agents (e.g. gpus) (a.k.a. non-ubiquitous / scarce resources). For this we would need to determine which agents have resources that can satisfy the quota prior to performing the allocation.