US Patent 11153375 Using reinforcement learning to scale queue-based services

Techniques for adjusting a compute capacity of a cloud computing system. In an example, a compute scaling application accesses, from a cloud computing system, a compute capacity indicating a number of allocated compute instances of a cloud computing system and usage metrics indicating pending task requests in a queue of the cloud computing system. The compute scaling application determines, for the cloud computing system, a compute scaling adjustment by applying a machine learning model to the compute capability of the cloud computing system and the usage metrics. The compute scaling adjustment indicates an adjustment to a number of compute instances of the cloud computing system. The compute scaling application provides the compute scaling adjustment to the cloud computing system. The cloud computing system adjusts a number of allocated compute instances.