Merge pull request #668 from eero-t/gpu-plugin-monitoring-to-operator

Add "-enable-monitoring" GPU plugin option operator support

Author: rojkov

deployments/operator/crd/bases/deviceplugin.intel.com_gpudeviceplugins.yaml

@@ -51,6 +51,10 @@ spec:
           spec:
             description: GpuDevicePluginSpec defines the desired state of GpuDevicePlugin.
             properties:
+              enableMonitoring:
+                description: EnableMonitoring enables the monitoring resource ('i915_monitoring')
+                  which gives access to all GPU devices on given node.
+                type: boolean
               image:
                 description: Image is a container image with GPU device plugin executable.
                 type: string

pkg/apis/deviceplugin/v1/gpudeviceplugin_types.go

@@ -35,13 +35,17 @@ type GpuDevicePluginSpec struct {
 	// +kubebuilder:validation:Minimum=1
 	SharedDevNum int `json:"sharedDevNum,omitempty"`
 
-	// ResourceManager handles the fractional resource management for multi-GPU nodes
-	ResourceManager bool `json:"resourceManager,omitempty"`
-
 	// LogLevel sets the plugin's log level.
 	// +kubebuilder:validation:Minimum=0
 	LogLevel int `json:"logLevel,omitempty"`
 
+	// ResourceManager handles the fractional resource management for multi-GPU nodes
+	ResourceManager bool `json:"resourceManager,omitempty"`
+
+	// EnableMonitoring enables the monitoring resource ('i915_monitoring')
+	// which gives access to all GPU devices on given node.
+	EnableMonitoring bool `json:"enableMonitoring,omitempty"`
+
 	// NodeSelector provides a simple way to constrain device plugin pods to nodes with particular labels.
 	NodeSelector map[string]string `json:"nodeSelector,omitempty"`
 }

pkg/controllers/gpu/controller.go

@@ -393,6 +393,10 @@ func getPodArgs(gdp *devicepluginv1.GpuDevicePlugin) []string {
 	args := make([]string, 0, 4)
 	args = append(args, "-v", strconv.Itoa(gdp.Spec.LogLevel))
 
+	if gdp.Spec.EnableMonitoring {
+		args = append(args, "-enable-monitoring")
+	}
+
 	if gdp.Spec.SharedDevNum > 0 {
 		args = append(args, "-shared-dev-num", strconv.Itoa(gdp.Spec.SharedDevNum))
 	} else {