Merge pull request #894 from NVIDIA/cherry-pick_fixes-for-openshift

Cherry-pick fixes for OpenShift

Author: tariq1890

assets/state-node-status-exporter/0200_role.yaml

@@ -19,3 +19,11 @@ rules:
   verbs:
   - get
   - list
+- apiGroups:
+  - apps
+  resources:
+  - daemonsets
+  verbs:
+  - get
+  - list
+  - watch

assets/state-node-status-exporter/0700_daemonset.yaml

@@ -40,6 +40,10 @@ spec:
           valueFrom:
             fieldRef:
               fieldPath: spec.nodeName
+        - name: OPERATOR_NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.namespace
         ports:
         - name: node-status
           containerPort: 8000

cmd/gpu-operator/main.go

@@ -27,6 +27,7 @@ import (
 	// to ensure that exec-entrypoint and run can make use of them.
 	"go.uber.org/zap/zapcore"
 	_ "k8s.io/client-go/plugin/pkg/client/auth"
+	"sigs.k8s.io/controller-runtime/pkg/cache"
 
 	apiconfigv1 "github.com/openshift/api/config/v1"
 	apiimagev1 "github.com/openshift/api/image/v1"
@@ -49,6 +50,7 @@ import (
 	nvidiav1alpha1 "github.com/NVIDIA/gpu-operator/api/nvidia/v1alpha1"
 	"github.com/NVIDIA/gpu-operator/controllers"
 	"github.com/NVIDIA/gpu-operator/controllers/clusterinfo"
+	"github.com/NVIDIA/gpu-operator/internal/consts"
 	"github.com/NVIDIA/gpu-operator/internal/info"
 	// +kubebuilder:scaffold:imports
 )
@@ -104,13 +106,24 @@ func main() {
 		Port: 9443,
 	})
 
+	operatorNamespace := os.Getenv("OPERATOR_NAMESPACE")
+	openshiftNamespace := consts.OpenshiftNamespace
+	cacheOptions := cache.Options{
+		DefaultNamespaces: map[string]cache.Config{
+			operatorNamespace: {},
+			// Also cache resources in the openshift namespace to retrieve ImageStreams when on an openshift  cluster
+			openshiftNamespace: {},
+		},
+	}
+
 	options := ctrl.Options{
 		Scheme:                 scheme,
 		Metrics:                metricsOptions,
 		HealthProbeBindAddress: probeAddr,
 		LeaderElection:         enableLeaderElection,
 		LeaderElectionID:       "53822513.nvidia.com",
 		WebhookServer:          webhookServer,
+		Cache:                  cacheOptions,
 	}
 
 	if enableLeaderElection && int(renewDeadline) != 0 {

controllers/clusterinfo/clusterinfo.go

@@ -341,7 +341,7 @@ func getOpenshiftDTKImages(ctx context.Context, c *rest.Config) map[string]strin
 	logger := log.FromContext(ctx)
 
 	name := "driver-toolkit"
-	namespace := "openshift"
+	namespace := consts.OpenshiftNamespace
 
 	ocpImageClient, err := imagesv1.NewForConfig(c)
 	if err != nil {

controllers/object_controls.go

@@ -52,6 +52,7 @@ import (
 	"sigs.k8s.io/yaml"
 
 	gpuv1 "github.com/NVIDIA/gpu-operator/api/nvidia/v1"
+	"github.com/NVIDIA/gpu-operator/internal/consts"
 	"github.com/NVIDIA/gpu-operator/internal/utils"
 )
 
@@ -3705,7 +3706,7 @@ func ocpHasDriverToolkitImageStream(n *ClusterPolicyController) (bool, error) {
 	ctx := n.ctx
 	found := &apiimagev1.ImageStream{}
 	name := "driver-toolkit"
-	namespace := "openshift"
+	namespace := consts.OpenshiftNamespace
 	err := n.client.Get(ctx, types.NamespacedName{Namespace: namespace, Name: name}, found)
 	if err != nil {
 		if apierrors.IsNotFound(err) {

internal/consts/consts.go

@@ -39,6 +39,9 @@ const (
 	// Containerd runtime
 	Containerd = "containerd"
 
+	// OpenshiftNamespace indicates the main namespace of an  Openshift cluster
+	OpenshiftNamespace = "openshift"
+
 	OcpDriverToolkitVersionLabel        = "openshift.driver-toolkit.rhcos"
 	OcpDriverToolkitIdentificationLabel = "openshift.driver-toolkit"
 	NfdOSTreeVersionLabelKey            = "feature.node.kubernetes.io/system-os_release.OSTREE_VERSION"

validator/metrics.go

@@ -238,6 +238,7 @@ func (nm *NodeMetrics) watchDriverValidation() {
 			nm.driverValidation.Set(1)
 			nm.driverValidationLastSuccess.Set(float64(time.Now().Unix()))
 		} else {
+			log.Errorf("failed to validate driver: %v", err)
 			nm.driverValidation.Set(0)
 		}
 		time.Sleep(driverValidationCheckDelaySeconds * time.Second)