Expose device UUIDs to node label

Signed-off-by: Zubiao Xiong <[email protected]>

Author: xiongzubiao

internal/lm/nvml.go

@@ -85,6 +85,11 @@ func NewDeviceLabeler(manager resource.Manager, config *spec.Config) (Labeler, e
 		return nil, fmt.Errorf("error creating IMEX labeler: %v", err)
 	}
 
+	uuidLabler, err := newGPUUUIDLabeler(devices)
+	if err != nil {
+		return nil, fmt.Errorf("error creating UUID labeler: %v", err)
+	}
+
 	l := Merge(
 		machineTypeLabeler,
 		versionLabeler,
@@ -93,6 +98,7 @@ func NewDeviceLabeler(manager resource.Manager, config *spec.Config) (Labeler, e
 		resourceLabeler,
 		gpuModeLabeler,
 		imexLabeler,
+		uuidLabler,
 	)
 
 	return l, nil
@@ -261,3 +267,16 @@ func getDeviceClasses(devices []resource.Device) ([]uint32, error) {
 	}
 	return classes, nil
 }
+
+// newGPUUUIDLabeler creates a new labeler that reports the UUIDs of GPUs on the node.
+func newGPUUUIDLabeler(devices []resource.Device) (Labeler, error) {
+	labels := make(Labels, len(devices))
+	for idx, d := range devices {
+		uuid, err := d.GetUUID()
+		if err != nil {
+			return nil, err
+		}
+		labels[fmt.Sprintf("nvidia.com/gpu-%d.uuid", idx)] = uuid
+	}
+	return labels, nil
+}

internal/resource/device_mock.go

@@ -81,6 +81,15 @@ func (d nvmlDevice) GetName() (string, error) {
 	return name, nil
 }
 
+// GetUUID returns the device UUID.
+func (d nvmlDevice) GetUUID() (string, error) {
+	uuid, ret := d.Device.GetUUID()
+	if ret != nvml.SUCCESS {
+		return "", ret
+	}
+	return uuid, nil
+}
+
 // GetTotalMemoryMB returns the total memory on a device in MB
 func (d nvmlDevice) GetTotalMemoryMB() (uint64, error) {
 	info, ret := d.Device.GetMemoryInfo()

internal/resource/nvml-device.go

@@ -104,6 +104,15 @@ func (d nvmlMigDevice) GetName() (string, error) {
 	return resourceName, nil
 }
 
+// GetUUID returns the UUID of the nvmlMigDevice.
+func (d nvmlMigDevice) GetUUID() (string, error) {
+	uuid, ret := d.MigDevice.GetUUID()
+	if ret != nvml.SUCCESS {
+		return "", ret
+	}
+	return uuid, nil
+}
+
 // GetTotalMemoryMB returns the total memory on a device in MB
 func (d nvmlMigDevice) GetTotalMemoryMB() (uint64, error) {
 	attr, err := d.GetAttributes()

internal/resource/nvml-mig-device.go

@@ -51,6 +51,11 @@ func (d vfioDevice) GetName() (string, error) {
 	return d.nvidiaPCIDevice.DeviceName, nil
 }
 
+// GetUUID is unsupported for vfio devices
+func (d vfioDevice) GetUUID() (string, error) {
+	return "", fmt.Errorf("GetUUID is not supported for vfio devices")
+}
+
 // GetTotalMemoryMB returns the total memory on a device in MB
 func (d vfioDevice) GetTotalMemoryMB() (uint64, error) {
 	_, val := d.nvidiaPCIDevice.Resources.GetTotalAddressableMemory(true)
@@ -72,6 +77,7 @@ func (d vfioDevice) GetPCIClass() (uint32, error) {
 func (d vfioDevice) IsFabricAttached() (bool, error) {
 	return false, nil
 }
+
 func (d vfioDevice) GetFabricIDs() (string, string, error) {
 	return "", "", fmt.Errorf("GetFabricIDs is not supported for vfio devices")
 }

internal/resource/sysfs-device.go

@@ -37,6 +37,7 @@ type Device interface {
 	GetMigDevices() ([]Device, error)
 	GetAttributes() (map[string]interface{}, error)
 	GetName() (string, error)
+	GetUUID() (string, error)
 	GetTotalMemoryMB() (uint64, error)
 	GetDeviceHandleFromMigDeviceHandle() (Device, error)
 	GetCudaComputeCapability() (int, int, error)