apis/nfd: Add Status to NodeFeature CRD

Signed-off-by: Oleg Zhurakivskyy <oleg.zhurakivskyy@intel.com>
kubernetes-sigs · Apr 26, 2024 · 078619c · 078619c
1 parent a7c58b1
commit 078619c
Show file tree

Hide file tree

Showing 8 changed files with 140 additions and 13 deletions.
diff --git a/api/generated/clientset/versioned/typed/nfd/v1alpha1/fake/fake_nodefeature.go b/api/generated/clientset/versioned/typed/nfd/v1alpha1/fake/fake_nodefeature.go
diff --git a/api/generated/clientset/versioned/typed/nfd/v1alpha1/nodefeature.go b/api/generated/clientset/versioned/typed/nfd/v1alpha1/nodefeature.go
diff --git a/api/nfd/v1alpha1/types.go b/api/nfd/v1alpha1/types.go
@@ -41,6 +41,7 @@ type NodeFeature struct {
 	metav1.ObjectMeta `json:"metadata,omitempty"`
 
 	Spec NodeFeatureSpec `json:"spec"`
+	Status NodeFeatureStatus `json:"status"`
 }
 
 // NodeFeatureSpec describes a NodeFeature object.
@@ -53,6 +54,22 @@ type NodeFeatureSpec struct {
 	Labels map[string]string `json:"labels"`
 }
 
+// Status of a NodeFeature object.
+type NodeFeatureStatus struct {
+	// UTC time when the NodeFeature object was last updated.
+	// +optional
+	LastAppliedAt metav1.Time `json:"lastAppliedAt,omitempty"`
+	// +optional
+	// Number of features discovered.
+	NumberOfFeatures int `json:"numberOfFeatures,omitempty"`
+	// +optional
+	// Number of errors during last feature discovery.
+	NumberOfFeatureErrors int `json:"numberOfFeatureErrors,omitempty"`
+	// +optional
+	// Number of labels created.
+	NumberOfLabels int `json:"numberOfLabels,omitempty"`
+}
+
 // Features is the collection of all discovered features.
 //
 // +protobuf=true

diff --git a/api/nfd/v1alpha1/zz_generated.deepcopy.go b/api/nfd/v1alpha1/zz_generated.deepcopy.go
diff --git a/deployment/base/nfd-crds/nfd-api-crds.yaml b/deployment/base/nfd-crds/nfd-api-crds.yaml
@@ -109,8 +109,26 @@ spec:
                   be created.
                 type: object
             type: object
+          status:
+            description: Status of a NodeFeature object.
+            properties:
+              lastAppliedAt:
+                description: UTC time when the NodeFeature object was last updated.
+                format: date-time
+                type: string
+              numberOfFeatureErrors:
+                description: Number of errors during last feature discovery.
+                type: integer
+              numberOfFeatures:
+                description: Number of features discovered.
+                type: integer
+              numberOfLabels:
+                description: Number of labels created.
+                type: integer
+            type: object
         required:
         - spec
+        - status
         type: object
     served: true
     storage: true

diff --git a/deployment/helm/node-feature-discovery/crds/nfd-api-crds.yaml b/deployment/helm/node-feature-discovery/crds/nfd-api-crds.yaml
@@ -109,8 +109,26 @@ spec:
                   be created.
                 type: object
             type: object
+          status:
+            description: Status of a NodeFeature object.
+            properties:
+              lastAppliedAt:
+                description: UTC time when the NodeFeature object was last updated.
+                format: date-time
+                type: string
+              numberOfFeatureErrors:
+                description: Number of errors during last feature discovery.
+                type: integer
+              numberOfFeatures:
+                description: Number of features discovered.
+                type: integer
+              numberOfLabels:
+                description: Number of labels created.
+                type: integer
+            type: object
         required:
         - spec
+        - status
         type: object
     served: true
     storage: true

diff --git a/pkg/nfd-worker/nfd-worker.go b/pkg/nfd-worker/nfd-worker.go
@@ -121,18 +121,19 @@ type ConfigOverrideArgs struct {
 }
 
 type nfdWorker struct {
-	args                Args
-	certWatch           *utils.FsWatcher
-	clientConn          *grpc.ClientConn
-	configFilePath      string
-	config              *NFDConfig
-	kubernetesNamespace string
-	grpcClient          pb.LabelerClient
-	healthServer        *grpc.Server
-	nfdClient           *nfdclient.Clientset
-	stop                chan struct{} // channel for signaling stop
-	featureSources      []source.FeatureSource
-	labelSources        []source.LabelSource
+	args                        Args
+	certWatch                   *utils.FsWatcher
+	clientConn                  *grpc.ClientConn
+	configFilePath              string
+	config                      *NFDConfig
+	kubernetesNamespace         string
+	grpcClient                  pb.LabelerClient
+	healthServer                *grpc.Server
+	nfdClient                   *nfdclient.Clientset
+	stop                        chan struct{} // channel for signaling stop
+	featureSources              []source.FeatureSource
+	numberOfFeatureSourceErrors int
+	labelSources                []source.LabelSource
 }
 
 // This ticker can represent infinite and normal intervals.
@@ -218,9 +219,11 @@ func (w *nfdWorker) startGrpcHealthServer(errChan chan<- error) error {
 // Run feature discovery.
 func (w *nfdWorker) runFeatureDiscovery() error {
 	discoveryStart := time.Now()
+	w.numberOfFeatureSourceErrors = 0
 	for _, s := range w.featureSources {
 		currentSourceStart := time.Now()
 		if err := s.Discover(); err != nil {
+			w.numberOfFeatureSourceErrors++
 			klog.ErrorS(err, "feature discovery failed", "source", s.Name())
 		}
 		klog.V(3).InfoS("feature discovery completed", "featureSource", s.Name(), "duration", time.Since(currentSourceStart))
@@ -747,6 +750,12 @@ func (m *nfdWorker) updateNodeFeatureObject(labels Labels) error {
 				Labels:   labels,
 			},
 		}
+		nfr.Status = nfdv1alpha1.NodeFeatureStatus{
+			LastAppliedAt:         metav1.Time{Time: time.Now().UTC()},
+			NumberOfFeatures:      len(m.featureSources),
+			NumberOfFeatureErrors: m.numberOfFeatureSourceErrors,
+			NumberOfLabels:        len(m.labelSources),
+		}
 
 		nfrCreated, err := cli.NfdV1alpha1().NodeFeatures(namespace).Create(context.TODO(), nfr, metav1.CreateOptions{})
 		if err != nil {
@@ -765,7 +774,12 @@ func (m *nfdWorker) updateNodeFeatureObject(labels Labels) error {
 			Features: *features,
 			Labels:   labels,
 		}
-
+		nfrUpdated.Status = nfdv1alpha1.NodeFeatureStatus{
+			LastAppliedAt:         metav1.Time{Time: time.Now().UTC()},
+			NumberOfFeatures:      len(m.featureSources),
+			NumberOfFeatureErrors: m.numberOfFeatureSourceErrors,
+			NumberOfLabels:        len(m.labelSources),
+		}
 		if !apiequality.Semantic.DeepEqual(nfr, nfrUpdated) {
 			klog.InfoS("updating NodeFeature object", "nodefeature", klog.KObj(nfr))
 			nfrUpdated, err = cli.NfdV1alpha1().NodeFeatures(namespace).Update(context.TODO(), nfrUpdated, metav1.UpdateOptions{})

diff --git a/test/e2e/node_feature_discovery_test.go b/test/e2e/node_feature_discovery_test.go
@@ -500,6 +500,19 @@ var _ = NFDDescribe(Label("nfd-master"), func() {
 					}
 					eventuallyNonControlPlaneNodes(ctx, f.ClientSet).Should(MatchLabels(expectedLabels, nodes))
 
+					By("Verifying the CRD status")
+					nf, err := nfdClient.NfdV1alpha1().NodeFeatures(f.Namespace.Name).Get(ctx, targetNodeName, metav1.GetOptions{})
+					Expect(err).NotTo(HaveOccurred(), "Error getting NodeFeature object: %q", err)
+
+					expectedStatus := nfdv1alpha1.NodeFeatureStatus{
+						LastAppliedAt:         nf.Status.LastAppliedAt,
+						NumberOfFeatures:      nf.Status.NumberOfFeatures,
+						NumberOfFeatureErrors: 0,
+						NumberOfLabels:        len(expectedLabels[targetNodeName]) - 1,
+					}
+					isEqual := (expectedStatus == nf.Status)
+					Expect(isEqual).To(BeTrue())
+
 					By("Deleting nfd-worker daemonset")
 					err = f.ClientSet.AppsV1().DaemonSets(f.Namespace.Name).Delete(ctx, workerDS.Name, metav1.DeleteOptions{})
 					Expect(err).NotTo(HaveOccurred())