DeviceFromName returns the device object for a given device name.

DiscoverGPUDevices discovers GPUs attached to the node, and updates `gpuDevices` map.

GetDevicesForAllContainers returns a map with container as the key and the list of devices allocated to that container as the value.

AcceleratorRequests reports the number of GPU devices requested by the container.

DutyCycle reports the percent of time when the GPU was actively processing per container.

DutyCycleNodeGpu reports the percent of time when the GPU was actively processing per Node.

MemoryTotal reports the total memory available on the GPU per container.

MemoryTotalNodeGpu reports the total memory available on the GPU per Node.

MemoryUsed reports GPU memory allocated per container.

MemoryUsedNodeGpu reports GPU memory allocated per Node.

ContainerID uniquely identifies a container.

MetricServer exposes GPU metrics for all containers and nodes in prometheus format on the specified port.