Add test case for unit test and delete the duplicated docker file.

.github/workflows/ci.yaml

@@ -16,11 +16,6 @@ jobs:
     steps:
     - uses: actions/checkout@v3
     - run: docker build --build-arg=KUBERNETES_MINOR_VERSION=latest --file Dockerfile.kubetest2 .
-  build-nccl:
-    runs-on: ubuntu-latest
-    steps:
-    - uses: actions/checkout@v3
-    - run: docker build --file e2e2/test/images/Dockerfile.aws-efa-nccl-tests .
   build-neuronx:
     runs-on: ubuntu-latest
     steps:

e2e2/test/cases/neuron/neuron_test.go

@@ -32,7 +32,7 @@ func TestMPIJobPytorchTraining(t *testing.T) {
 		WithLabel("hardware", "gpu").
 		Setup(func(ctx context.Context, t *testing.T, cfg *envconf.Config) context.Context {
 			if *neuronTestImage == "" {
-				t.Fatal(fmt.Errorf("neuronTestImage must be set to run neuron single node test, use https://github.com/aws/aws-k8s-tester/blob/main/e2e2/test/images/Dockerfile.neuronx-tests to build the image and -neuronTestImage to set the image url"))
+				t.Fatal(fmt.Errorf("neuronTestImage must be set to run neuron single node test, use https://github.com/aws/aws-k8s-tester/blob/main/e2e2/test/images/neuron/Dockerfile to build the image and -neuronTestImage to set the image url"))
 			}
 			var err error
 			renderedNeuronSingleNodeManifest, err = fwext.RenderManifests(neuronSingleNodeManifest, neuronSingleNodeManifestTplVars{

e2e2/test/cases/nvidia/main_test.go

@@ -23,13 +23,13 @@ import (
 )
 
 var (
-	testenv       env.Environment
-	nodeType      *string
-	efaEnabled    *bool
-	ncclTestImage *string
-	nodeCount     int
-	gpuPerNode    int
-	efaPerNode    int
+	testenv         env.Environment
+	nodeType        *string
+	efaEnabled      *bool
+	nvidiaTestImage *string
+	nodeCount       int
+	gpuPerNode      int
+	efaPerNode      int
 )
 
 var (
@@ -43,7 +43,7 @@ var (
 
 func TestMain(m *testing.M) {
 	nodeType = flag.String("nodeType", "", "node type for the tests")
-	ncclTestImage = flag.String("ncclTestImage", "", "nccl test image for nccl tests")
+	nvidiaTestImage = flag.String("nvidiaTestImage", "", "nccl test image for nccl tests")
 	efaEnabled = flag.Bool("efaEnabled", false, "enable efa tests")
 	cfg, err := envconf.NewFromFlags()
 	if err != nil {

e2e2/test/cases/nvidia/manifests/mpi-job-nccl-test-multi-node.yaml

@@ -16,7 +16,7 @@ spec:
         spec:
           restartPolicy: OnFailure
           containers:
-          - image: {{.NcclTestImage}}
+          - image: {{.NvidiaTestImage}}
             imagePullPolicy: Always
             name: nccl-test-launcher
             env:
@@ -82,7 +82,7 @@ spec:
             emptyDir:
               medium: Memory
           containers:
-          - image: {{.NcclTestImage}}
+          - image: {{.NvidiaTestImage}}
             imagePullPolicy: Always
             name: nccl-test-worker
             volumeMounts:

e2e2/test/cases/nvidia/mpi_test.go

@@ -31,7 +31,7 @@ type ncclTestManifestTplVars struct {
 	WorkerNodeCount     int
 	WorkerNodeGpuCount  int
 	GpuPerNode          int
-	NcclTestImage       string
+	NvidiaTestImage     string
 	EfaInterfacePerNode int
 	EfaUseDeviceRdma    int
 }
@@ -77,8 +77,8 @@ func TestMPIJobPytorchTraining(t *testing.T) {
 		WithLabel("hardware", "gpu").
 		WithLabel("hardware", "efa").
 		Setup(func(ctx context.Context, t *testing.T, cfg *envconf.Config) context.Context {
-			if *ncclTestImage == "" {
-				t.Fatal(fmt.Errorf("efaImage must be set to run nccl test, use https://github.com/aws/aws-k8s-tester/blob/main/e2e2/test/images/Dockerfile.aws-efa-nccl-tests to build the image and -efaImage to set the image url"))
+			if *nvidiaTestImage == "" {
+				t.Fatal(fmt.Errorf("nvidiaTestImage must be set to run unit test, use https://github.com/aws/aws-k8s-tester/blob/main/e2e2/test/images/nvidia/Dockerfile to build the image and -nvidiaTestImage to set the image url"))
 			}
 			// https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/efa-start-nccl-base.html#nccl-start-base-test
 			var EfaUseDeviceRdma int
@@ -90,7 +90,7 @@ func TestMPIJobPytorchTraining(t *testing.T) {
 				WorkerNodeCount:     nodeCount - 1,
 				WorkerNodeGpuCount:  (nodeCount - 1) * gpuPerNode,
 				GpuPerNode:          gpuPerNode,
-				NcclTestImage:       *ncclTestImage,
+				NvidiaTestImage:     *nvidiaTestImage,
 				EfaInterfacePerNode: efaPerNode,
 				EfaUseDeviceRdma:    EfaUseDeviceRdma,
 			})

e2e2/test/cases/nvidia/unit_test.go

@@ -0,0 +1,71 @@
+package nvidia
+
+import (
+	"context"
+	_ "embed"
+	"fmt"
+	"testing"
+	"time"
+
+	fwext "github.com/aws/aws-k8s-tester/e2e2/internal/framework_extensions"
+	"sigs.k8s.io/e2e-framework/klient/wait"
+	"sigs.k8s.io/e2e-framework/pkg/envconf"
+	"sigs.k8s.io/e2e-framework/pkg/features"
+
+	batchv1 "k8s.io/api/batch/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+)
+
+var (
+	//go:embed manifests/job-unit-test-single-node.yaml
+	jobUnitTestSingleNodeManifest         []byte
+	renderedJobUnitTestSingleNodeManifest []byte
+)
+
+type unitTestManifestTplVars struct {
+	NvidiaTestImage string
+}
+
+func TestSingleNodeUnitTest(t *testing.T) {
+	unitTest := features.New("unit-test").
+		WithLabel("suite", "nvidia").
+		WithLabel("hardware", "gpu").
+		Setup(func(ctx context.Context, t *testing.T, cfg *envconf.Config) context.Context {
+			if *nvidiaTestImage == "" {
+				t.Fatal(fmt.Errorf("nvidiaTestImage must be set to run unit test, use https://github.com/aws/aws-k8s-tester/blob/main/e2e2/test/images/nvidia/Dockerfile to build the image and -nvidiaTestImage to set the image url"))
+			}
+			var err error
+			renderedJobUnitTestSingleNodeManifest, err = fwext.RenderManifests(jobUnitTestSingleNodeManifest, unitTestManifestTplVars{
+				NvidiaTestImage: *nvidiaTestImage,
+			})
+			if err != nil {
+				t.Fatal(err)
+			}
+			err = fwext.ApplyManifests(cfg.Client().RESTConfig(), renderedJobUnitTestSingleNodeManifest)
+			if err != nil {
+				t.Fatal(err)
+			}
+			return ctx
+		}).
+		Assess("Unit test Job succeeds", func(ctx context.Context, t *testing.T, cfg *envconf.Config) context.Context {
+			job := &batchv1.Job{
+				ObjectMeta: metav1.ObjectMeta{Name: "unit-test-job", Namespace: "default"},
+			}
+			err := wait.For(fwext.NewConditionExtension(cfg.Client().Resources()).JobSucceeded(job),
+				wait.WithTimeout(time.Minute*20))
+			if err != nil {
+				t.Fatal(err)
+			}
+			return ctx
+		}).
+		Teardown(func(ctx context.Context, t *testing.T, cfg *envconf.Config) context.Context {
+			err := fwext.DeleteManifests(cfg.Client().RESTConfig(), renderedJobUnitTestSingleNodeManifest)
+			if err != nil {
+				t.Fatal(err)
+			}
+			return ctx
+		}).
+		Feature()
+
+	testenv.Test(t, unitTest)
+}