Add label app:nvidia-dra-driver-gpu to all project components

Signed-off-by: Carlos Eduardo Arango Gutierrez <[email protected]>

Author: ArangoGutierrez

cmd/compute-domain-controller/computedomain.go

@@ -35,7 +35,10 @@ type GetComputeDomainFunc func(uid string) (*nvapi.ComputeDomain, error)
 const (
 	informerResyncPeriod = 10 * time.Minute
 
-	computeDomainLabelKey  = "resource.nvidia.com/computeDomain"
+	computeDomainLabelKey = "resource.nvidia.com/computeDomain"
+	appLabelKey           = "app"
+	appLabelValue         = "nvidia-dra-driver-gpu"
+
 	computeDomainFinalizer = computeDomainLabelKey
 
 	computeDomainDefaultChannelDeviceClass = "compute-domain-default-channel.nvidia.com"

cmd/compute-domain-controller/daemonset.go

@@ -47,6 +47,8 @@ type DaemonSetTemplateData struct {
 	Finalizer                 string
 	ComputeDomainLabelKey     string
 	ComputeDomainLabelValue   types.UID
+	AppLabelKey               string
+	AppLabelValue             string
 	ResourceClaimTemplateName string
 	ImageName                 string
 }
@@ -180,6 +182,8 @@ func (m *DaemonSetManager) Create(ctx context.Context, namespace string, cd *nva
 		Finalizer:                 computeDomainFinalizer,
 		ComputeDomainLabelKey:     computeDomainLabelKey,
 		ComputeDomainLabelValue:   cd.UID,
+		AppLabelKey:               appLabelKey,
+		AppLabelValue:             appLabelValue,
 		ResourceClaimTemplateName: rct.Name,
 		ImageName:                 m.config.imageName,
 	}
@@ -189,24 +193,24 @@ func (m *DaemonSetManager) Create(ctx context.Context, namespace string, cd *nva
 		return nil, fmt.Errorf("failed to parse template file: %w", err)
 	}
 
-	var deploymentYaml bytes.Buffer
-	if err := tmpl.Execute(&deploymentYaml, templateData); err != nil {
+	var daemonSetYaml bytes.Buffer
+	if err := tmpl.Execute(&daemonSetYaml, templateData); err != nil {
 		return nil, fmt.Errorf("failed to execute template: %w", err)
 	}
 
 	var unstructuredObj unstructured.Unstructured
-	err = yaml.Unmarshal(deploymentYaml.Bytes(), &unstructuredObj)
+	err = yaml.Unmarshal(daemonSetYaml.Bytes(), &unstructuredObj)
 	if err != nil {
 		return nil, fmt.Errorf("failed to unmarshal yaml: %w", err)
 	}
 
-	var deployment appsv1.DaemonSet
-	err = runtime.DefaultUnstructuredConverter.FromUnstructured(unstructuredObj.UnstructuredContent(), &deployment)
+	var daemonSet appsv1.DaemonSet
+	err = runtime.DefaultUnstructuredConverter.FromUnstructured(unstructuredObj.UnstructuredContent(), &daemonSet)
 	if err != nil {
 		return nil, fmt.Errorf("failed to convert unstructured data to typed object: %w", err)
 	}
 
-	d, err := m.config.clientsets.Core.AppsV1().DaemonSets(deployment.Namespace).Create(ctx, &deployment, metav1.CreateOptions{})
+	d, err := m.config.clientsets.Core.AppsV1().DaemonSets(daemonSet.Namespace).Create(ctx, &daemonSet, metav1.CreateOptions{})
 	if err != nil {
 		return nil, fmt.Errorf("error creating DaemonSet: %w", err)
 	}

cmd/compute-domain-controller/resourceclaimtemplate.go

@@ -49,6 +49,8 @@ type ResourceClaimTemplateTemplateData struct {
 	Finalizer               string
 	ComputeDomainLabelKey   string
 	ComputeDomainLabelValue types.UID
+	AppLabelKey             string
+	AppLabelValue           string
 	TargetLabelKey          string
 	TargetLabelValue        string
 	DeviceClassName         string
@@ -298,6 +300,8 @@ func (m *DaemonSetResourceClaimTemplateManager) Create(ctx context.Context, name
 		Finalizer:               computeDomainFinalizer,
 		ComputeDomainLabelKey:   computeDomainLabelKey,
 		ComputeDomainLabelValue: cd.UID,
+		AppLabelKey:             appLabelKey,
+		AppLabelValue:           appLabelValue,
 		TargetLabelKey:          computeDomainResourceClaimTemplateTargetLabelKey,
 		TargetLabelValue:        computeDomainResourceClaimTemplateTargetDaemon,
 		DeviceClassName:         computeDomainDaemonDeviceClass,

cmd/gpu-kubelet-plugin/device_state.go

@@ -22,7 +22,7 @@ import (
 	"slices"
 	"sync"
 
-	resourceapi "k8s.io/api/resource/v1beta1"
+	resourceapi "k8s.io/api/resource/v1beta2"
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/dynamic-resource-allocation/kubeletplugin"
 	"k8s.io/klog/v2"

cmd/gpu-kubelet-plugin/driver.go

@@ -21,7 +21,7 @@ import (
 	"fmt"
 	"time"
 
-	resourceapi "k8s.io/api/resource/v1beta1"
+	resourceapi "k8s.io/api/resource/v1beta2"
 	"k8s.io/apimachinery/pkg/types"
 	coreclientset "k8s.io/client-go/kubernetes"
 	"k8s.io/dynamic-resource-allocation/kubeletplugin"

deployments/helm/nvidia-dra-driver-gpu/values.yaml

@@ -61,6 +61,8 @@ resources:
     enabled: true
 
 controller:
+  labels:
+    app: nvidia-dra-driver-gpu
   priorityClassName: "system-node-critical"
   podAnnotations: {}
   podSecurityContext: {}
@@ -82,6 +84,8 @@ controller:
             operator: "Exists"
 
 kubeletPlugin:
+  labels:
+    app: nvidia-dra-driver-gpu
   priorityClassName: "system-node-critical"
   updateStrategy:
     type: RollingUpdate

templates/compute-domain-daemon-claim-template.tmpl.yaml

@@ -8,6 +8,7 @@ metadata:
     - {{ .Finalizer }}
   labels:
     {{ .ComputeDomainLabelKey }}: {{ .ComputeDomainLabelValue }}
+    {{ .AppLabelKey }}: {{ .AppLabelValue }}
     {{ .TargetLabelKey }}: {{ .TargetLabelValue }}
 spec:
   spec:

templates/compute-domain-daemon.tmpl.yaml

@@ -8,6 +8,7 @@ metadata:
     - {{ .Finalizer }}
   labels:
     {{ .ComputeDomainLabelKey }}: {{ .ComputeDomainLabelValue }}
+    {{ .AppLabelKey }}: {{ .AppLabelValue }}
 spec:
   selector:
     matchLabels:

templates/compute-domain-workload-claim-template.tmpl.yaml

@@ -8,6 +8,7 @@ metadata:
     - {{ .Finalizer }}
   labels:
     {{ .ComputeDomainLabelKey }}: {{ .ComputeDomainLabelValue }}
+    {{ .AppLabelKey }}: {{ .AppLabelValue }}
     {{ .TargetLabelKey }}: {{ .TargetLabelValue }}
 spec:
   spec: