Removing extras

Signed-off-by: Vishesh Tanksale <vtanksale@nvidia.com>

Author: visheshtanksale

.github/workflows/e2e.yml

@@ -32,7 +32,6 @@ jobs:
   e2e-tests:
     runs-on: linux-amd64-cpu4
     if: ${{ github.event.workflow_run.conclusion == 'success' }} && ${{ github.event.workflow_run.event == 'push' }}
-    # if: false # TODO: Disabled until e2e test infra is fixed
     steps:
       - name: Check out code
         uses: actions/checkout@v4

api/apps/v1alpha1/nimcache_types.go

@@ -303,9 +303,9 @@ func (n *NIMCache) GetTolerations() []corev1.Toleration {
 
 // GetNodeSelectors returns nodeselectors configured for the NIMCache Job.
 func (n *NIMCache) GetNodeSelectors() map[string]string {
-	/*if n.Spec.NodeSelector == nil {
+	if n.Spec.NodeSelector == nil {
 		return map[string]string{"feature.node.kubernetes.io/pci-10de.present": "true"}
-	}*/
+	}
 	return n.Spec.NodeSelector
 }
 

internal/controller/nimcache_controller.go

@@ -608,33 +608,27 @@ func (r *NIMCacheReconciler) reconcileModelManifest(ctx context.Context, nimCach
 		logger.Error(err, "failed to create", "pod", pod.Name)
 		return false, err
 	}
-	logger.Info("Created pod for model manifest extraction", "pod", pod.Name)
 
 	existingPod := &corev1.Pod{}
 	err = r.Get(ctx, client.ObjectKey{Name: pod.Name, Namespace: nimCache.Namespace}, existingPod)
 	if err != nil {
 		logger.Error(err, "failed to get pod for model selection", "pod", pod.Name)
-		return true, err
+		return false, err
 	}
-	logger.Info("HERE", "pod", pod.Name)
 	if existingPod.Status.Phase != corev1.PodRunning {
 		// requeue request with delay until the pod is ready
-		logger.Info("HERE", "pod", existingPod.Status)
 		return true, nil
 	}
-	logger.Info("HERE1", "pod", pod.Name)
 	// Extract manifest file
 	output, err := k8sutil.GetPodLogs(ctx, existingPod, NIMCacheContainerName)
 	if err != nil {
 		logger.Error(err, "failed to get pod logs for parsing model manifest file", "pod", pod.Name)
 		return false, err
 	}
-	logger.Info("HERE2", "pod", pod.Name)
 	if output == "" {
 		logger.Info("Requeuing to wait for the manifest to be copied from the container")
 		return true, nil
 	}
-	logger.Info("HERE3", "pod", pod.Name)
 	parser := nimparserutils.GetNIMParser([]byte(output))
 	// Parse the file
 	manifest, err := parser.ParseModelManifestFromRawOutput([]byte(output))
@@ -643,14 +637,12 @@ func (r *NIMCacheReconciler) reconcileModelManifest(ctx context.Context, nimCach
 		return false, err
 	}
 	logger.Info("manifest file", "nimcache", nimCache.Name, "manifest", manifest)
-	logger.Info("HERE4", "pod", pod.Name)
 	// Create a ConfigMap with the model manifest file for re-use
 	err = r.createManifestConfigMap(ctx, nimCache, &manifest)
 	if err != nil {
 		logger.Error(err, "Failed to create model manifest config map")
 		return false, err
 	}
-	logger.Info("HERE5", "pod", pod.Name)
 	// Model manifest is successfully extracted, cleanup temporary pod
 	err = r.Delete(ctx, existingPod)
 	if err != nil && !errors.IsNotFound(err) {
@@ -664,7 +656,6 @@ func (r *NIMCacheReconciler) reconcileModelManifest(ctx context.Context, nimCach
 
 func (r *NIMCacheReconciler) reconcileModelSelection(ctx context.Context, nimCache *appsv1alpha1.NIMCache) error {
 	logger := r.GetLogger()
-	logger.Info("Reconciling model selection", "nimcache", nimCache.Name)
 
 	// reconcile model selection pod
 	if isModelSelectionRequired(nimCache) && !isModelSelectionDone(nimCache) {
@@ -710,7 +701,6 @@ func (r *NIMCacheReconciler) reconcileModelSelection(ctx context.Context, nimCac
 
 func (r *NIMCacheReconciler) reconcileJob(ctx context.Context, nimCache *appsv1alpha1.NIMCache) error {
 	logger := r.GetLogger()
-	logger.Info("Reconciling job", "nimcache", nimCache.Name)
 	// reconcile model caching job
 	job := &batchv1.Job{}
 	jobName := types.NamespacedName{Name: getJobName(nimCache), Namespace: nimCache.GetNamespace()}

test/e2e/data/nimservice.yml

@@ -4,8 +4,8 @@ metadata:
   name: meta-llama3-8b-instruct
 spec:
   image:
-    repository: nvcr.io/nim/meta/llama3-8b-instruct
-    tag: 1.0.0
+    repository: nvcr.io/nim/meta/llama-3.2-1b-instruct
+    tag: 1.8
     pullPolicy: IfNotPresent
     pullSecrets:
       - ngc-secret

test/e2e/nim-operator_test.go

@@ -31,7 +31,6 @@ import (
 	helmValues "github.com/mittwald/go-helm-client/values"
 	"helm.sh/helm/v3/pkg/repo"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
-	clientset "k8s.io/client-go/kubernetes"
 	kubernetes "k8s.io/client-go/kubernetes"
 	"sigs.k8s.io/yaml"
 
@@ -65,21 +64,6 @@ var _ = Describe("NIM Operator", Ordered, func() {
 
 			err = diagnosticsCollector.Collect(ctx)
 			Expect(err).NotTo(HaveOccurred())
-
-			cli, err := versioned.NewForConfig(clientConfig)
-			Expect(err).NotTo(HaveOccurred())
-
-			nimCacheObject, _ := cli.AppsV1alpha1().NIMCaches(testNamespace.Name).Get(ctx, "meta-llama3-8b-instruct", metav1.GetOptions{})
-			fmt.Printf("NIMCache object: %#v\n", nimCacheObject)
-
-			clientSet, err = clientset.NewForConfig(clientConfig)
-			Expect(err).NotTo(HaveOccurred())
-			pods, err := clientSet.CoreV1().Pods(testNamespace.Name).List(ctx, metav1.ListOptions{})
-			Expect(err).NotTo(HaveOccurred())
-			for _, pod := range pods.Items {
-				fmt.Println("Pod name:", pod.Name)
-			}
-
 		}
 	})
 
@@ -137,9 +121,6 @@ var _ = Describe("NIM Operator", Ordered, func() {
 			cli, err := versioned.NewForConfig(clientConfig)
 			Expect(err).NotTo(HaveOccurred())
 
-			clientSet, err = clientset.NewForConfig(clientConfig)
-			Expect(err).NotTo(HaveOccurred())
-
 			nimCache := &v1alpha1.NIMCache{}
 			data, err := os.ReadFile(filepath.Join(cwd, "data", "nimcache.yml"))
 			Expect(err).NotTo(HaveOccurred())
@@ -153,16 +134,6 @@ var _ = Describe("NIM Operator", Ordered, func() {
 			By("Checking the NIMCache object state is ready")
 			Eventually(func() bool {
 				nimCacheObject, _ := cli.AppsV1alpha1().NIMCaches(testNamespace.Name).Get(ctx, nimCache.Name, metav1.GetOptions{})
-				fmt.Println("NIMCache object conditions:", nimCacheObject.Status.Conditions)
-				fmt.Println("NIMCache object state:", nimCacheObject.Status.State)
-				//pod, err := clientSet.CoreV1().Pods(testNamespace.Name).Get(ctx, fmt.Sprintf("%s-pod", nimCache.GetName()), metav1.GetOptions{})
-
-				err := DescribePod(clientSet.(*kubernetes.Clientset), testNamespace.Name, fmt.Sprintf("%s-pod", nimCache.GetName()))
-				if err != nil {
-					fmt.Println("Error:", err)
-				}
-
-				fmt.Println("***************111")
 				return nimCacheObject.Status.State == v1alpha1.NimCacheStatusReady
 			}, Timeout, 5*time.Second).Should(BeTrue())
 
@@ -407,7 +378,7 @@ func installEntitystoreDependencies() {
 		CreateNamespace: false,
 		Wait:            true,
 		WaitForJobs:     true,
-		Timeout:         2 * time.Minute,
+		Timeout:         10 * time.Minute,
 		CleanupOnFail:   true,
 		ValuesOptions:   values}
 	_, err = helmClient.InstallOrUpgradeChart(ctx, chartSpec, nil)