Add single node Neuron test to the e2e tester

.github/workflows/ci.yaml

@@ -20,4 +20,9 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@v3
-    - run: docker build --file e2e2/test/images/Dockerfile.aws-efa-nccl-tests .
+    - run: docker build --file e2e2/test/images/Dockerfile.aws-efa-nccl-tests .
+  build-neuronx:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v3
+    - run: docker build --file e2e2/test/images/Dockerfile.neuronx-tests .

e2e2/internal/framework_extensions/conditions.go

@@ -4,6 +4,7 @@ import (
 	"context"
 
 	appsv1 "k8s.io/api/apps/v1"
+	batchv1 "k8s.io/api/batch/v1"
 	apimachinerywait "k8s.io/apimachinery/pkg/util/wait"
 
 	"sigs.k8s.io/e2e-framework/klient/k8s"
@@ -41,3 +42,17 @@ func (c *ConditionExtension) DaemonSetReady(daemonset k8s.Object) apimachinerywa
 		return
 	}
 }
+
+func (c *ConditionExtension) JobSucceeded(job k8s.Object) apimachinerywait.ConditionWithContextFunc {
+	return func(ctx context.Context) (done bool, err error) {
+		if err := c.resources.Get(ctx, job.GetName(), job.GetNamespace(), job); err != nil {
+			return false, err
+		}
+		status := job.(*batchv1.Job).Status
+		spec := job.(*batchv1.Job).Spec
+		if status.Succeeded == *spec.Completions {
+			done = true
+		}
+		return
+	}
+}

e2e2/test/cases/neuron/main_test.go

@@ -0,0 +1,80 @@
+package nvidia
+
+import (
+	"context"
+	_ "embed"
+	"flag"
+	"log"
+	"os"
+	"slices"
+	"testing"
+	"time"
+
+	fwext "github.com/aws/aws-k8s-tester/e2e2/internal/framework_extensions"
+	appsv1 "k8s.io/api/apps/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"sigs.k8s.io/e2e-framework/klient/wait"
+	"sigs.k8s.io/e2e-framework/pkg/env"
+	"sigs.k8s.io/e2e-framework/pkg/envconf"
+)
+
+var (
+	testenv         env.Environment
+	neuronTestImage *string
+)
+
+var (
+	//go:embed manifests/k8s-neuron-device-plugin-rbac.yml
+	neuronDevicePlugiRbacManifest []byte
+	//go:embed manifests/k8s-neuron-device-plugin.yml
+	neuronDevicePluginManifest []byte
+)
+
+func TestMain(m *testing.M) {
+	neuronTestImage = flag.String("neuronTestImage", "", "image for neuron single node test")
+	cfg, err := envconf.NewFromFlags()
+	if err != nil {
+		log.Fatalf("failed to initialize test environment: %v", err)
+	}
+	testenv = env.NewWithConfig(cfg)
+
+	// all NVIDIA tests require the device plugin and MPI operator
+	manifests := [][]byte{
+		neuronDevicePluginManifest,
+		neuronDevicePlugiRbacManifest,
+	}
+
+	testenv.Setup(
+		func(ctx context.Context, config *envconf.Config) (context.Context, error) {
+			err := fwext.ApplyManifests(config.Client().RESTConfig(), manifests...)
+			if err != nil {
+				return ctx, err
+			}
+			return ctx, nil
+		},
+		func(ctx context.Context, config *envconf.Config) (context.Context, error) {
+			ds := appsv1.DaemonSet{
+				ObjectMeta: metav1.ObjectMeta{Name: "neuron-device-plugin-daemonset", Namespace: "kube-system"},
+			}
+			err := wait.For(fwext.NewConditionExtension(config.Client().Resources()).DaemonSetReady(&ds),
+				wait.WithTimeout(time.Minute*5))
+			if err != nil {
+				return ctx, err
+			}
+			return ctx, nil
+		},
+	)
+
+	testenv.Finish(
+		func(ctx context.Context, config *envconf.Config) (context.Context, error) {
+			slices.Reverse(manifests)
+			err = fwext.DeleteManifests(config.Client().RESTConfig(), manifests...)
+			if err != nil {
+				return ctx, err
+			}
+			return ctx, nil
+		},
+	)
+
+	os.Exit(testenv.Run(m))
+}

e2e2/test/cases/neuron/manifests/k8s-neuron-device-plugin-rbac.yml

@@ -0,0 +1,58 @@
+---
+kind: ClusterRole
+apiVersion: rbac.authorization.k8s.io/v1
+metadata:
+  name: neuron-device-plugin
+rules:
+- apiGroups:
+  - ""
+  resources:
+  - nodes
+  verbs:
+  - get
+  - list
+  - watch
+- apiGroups:
+  - ""
+  resources:
+  - events
+  verbs:
+  - create
+  - patch
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - update
+  - patch
+  - get
+  - list
+  - watch
+- apiGroups:
+  - ""
+  resources:
+  - nodes/status
+  verbs:
+  - patch
+  - update
+---
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: neuron-device-plugin
+  namespace: kube-system
+---
+kind: ClusterRoleBinding
+apiVersion: rbac.authorization.k8s.io/v1
+metadata:
+  name: neuron-device-plugin
+  namespace: kube-system
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: ClusterRole
+  name: neuron-device-plugin
+subjects:
+- kind: ServiceAccount
+  name: neuron-device-plugin
+  namespace: kube-system

e2e2/test/cases/neuron/manifests/k8s-neuron-device-plugin.yml

@@ -0,0 +1,97 @@
+apiVersion: apps/v1
+kind: DaemonSet
+metadata:
+  name: neuron-device-plugin-daemonset
+  namespace: kube-system
+spec:
+  selector:
+    matchLabels:
+      name:  neuron-device-plugin-ds
+  updateStrategy:
+    type: RollingUpdate
+  template:
+    metadata:
+      # Uncomment the annotation below if k8s version is 1.13 or lower
+      # annotations:
+      #  scheduler.alpha.kubernetes.io/critical-pod: ""
+      labels:
+        name: neuron-device-plugin-ds
+    spec:
+      serviceAccount: neuron-device-plugin
+      tolerations:
+      - key: CriticalAddonsOnly
+        operator: Exists
+      - key: aws.amazon.com/neuron
+        operator: Exists
+        effect: NoSchedule
+      # Mark this pod as a critical add-on; when enabled, the critical add-on
+      # scheduler reserves resources for critical add-on pods so that they can
+      # be rescheduled after a failure.
+      # See https://kubernetes.io/docs/tasks/administer-cluster/guaranteed-scheduling-critical-addon-pods/
+      priorityClassName: "system-node-critical"
+      affinity:
+        nodeAffinity:
+          requiredDuringSchedulingIgnoredDuringExecution:
+            nodeSelectorTerms:
+              # Uncomment following matchExpressions if using k8s 1.16 or lower
+              #- matchExpressions:
+              #    - key: "beta.kubernetes.io/instance-type"
+              #      operator: In
+              #      values:
+              #        - inf1.xlarge
+              #        - inf1.2xlarge
+              #        - inf1.6xlarge
+              #        - inf1.24xlarge
+              #        - inf2.xlarge
+              #        - inf2.8xlarge
+              #        - inf2.24xlarge
+              #        - inf2.48xlarge
+              #        - trn1.2xlarge
+              #        - trn1.32xlarge
+              #        - trn1n.32xlarge
+              - matchExpressions:
+                  - key: "node.kubernetes.io/instance-type"
+                    operator: In
+                    values:
+                      - inf1.xlarge
+                      - inf1.2xlarge
+                      - inf1.6xlarge
+                      - inf1.24xlarge
+                      - inf2.xlarge
+                      - inf2.8xlarge
+                      - inf2.24xlarge
+                      - inf2.48xlarge
+                      - trn1.2xlarge
+                      - trn1.32xlarge
+                      - trn1n.32xlarge
+      containers:
+        # Find all neuron-device-plugin images at https://gallery.ecr.aws/neuron/neuron-device-plugin
+      - image: public.ecr.aws/neuron/neuron-device-plugin:2.19.16.0
+        imagePullPolicy: Always
+        name: neuron-device-plugin
+        env:
+        - name: KUBECONFIG
+          value: /etc/kubernetes/kubelet.conf
+        - name: NODE_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+        securityContext:
+          allowPrivilegeEscalation: false
+          capabilities:
+            drop: ["ALL"]
+        volumeMounts:
+          - name: device-plugin
+            mountPath: /var/lib/kubelet/device-plugins
+          - name: infa-map
+            mountPath: /run
+      volumes:
+        - name: device-plugin
+          hostPath:
+            path: /var/lib/kubelet/device-plugins
+        - name: infa-map
+          hostPath:
+            path: /run
+
+
+

e2e2/test/cases/neuron/manifests/single-node-test-neuronx.yaml

@@ -0,0 +1,30 @@
+kind: Job
+apiVersion: batch/v1
+metadata:
+  name: neuronx-single-node
+  labels:
+    app: neuronx-single-node
+spec:
+  template:
+    metadata:
+      labels:
+        app: neuronx-single-node
+    spec:
+      containers:
+      - name: neuronx-single-node-test
+        image: {{.NeuronTestImage}}
+        command:
+        - /bin/bash
+        - ./pytorch_tests/singleNodeTest.sh
+        imagePullPolicy: Always
+        resources:
+          limits:
+            cpu: "4"
+            memory: 4Gi
+            aws.amazon.com/neuron: "1"
+          requests:
+            cpu: "1"
+            memory: 1Gi
+            aws.amazon.com/neuron: "1"
+      restartPolicy: Never
+  backoffLimit: 4

e2e2/test/cases/neuron/neuron_test.go

@@ -0,0 +1,70 @@
+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/single-node-test-neuronx.yaml
+	neuronSingleNodeManifest         []byte
+	renderedNeuronSingleNodeManifest []byte
+)
+
+type neuronSingleNodeManifestTplVars struct {
+	NeuronTestImage string
+}
+
+func TestMPIJobPytorchTraining(t *testing.T) {
+	singleNode := features.New("single-node").
+		WithLabel("suite", "neuron").
+		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"))
+			}
+			renderedNeuronSingleNodeManifest, err := fwext.RenderManifests(neuronSingleNodeManifest, neuronSingleNodeManifestTplVars{
+				NeuronTestImage: *neuronTestImage,
+			})
+			if err != nil {
+				t.Fatal(err)
+			}
+			err = fwext.ApplyManifests(cfg.Client().RESTConfig(), renderedNeuronSingleNodeManifest)
+			if err != nil {
+				t.Fatal(err)
+			}
+			return ctx
+		}).
+		Assess("Single node test Job succeeds", func(ctx context.Context, t *testing.T, cfg *envconf.Config) context.Context {
+			job := &batchv1.Job{
+				ObjectMeta: metav1.ObjectMeta{Name: "neuronx-single-node", 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(), renderedNeuronSingleNodeManifest)
+			if err != nil {
+				t.Fatal(err)
+			}
+			return ctx
+		}).
+		Feature()
+
+	testenv.Test(t, singleNode)
+}

e2e2/test/images/Dockerfile.neuronx-tests

@@ -0,0 +1,5 @@
+# Start with the Neuron base image
+FROM public.ecr.aws/neuron/pytorch-training-neuronx:2.1.2-neuronx-py310-sdk2.18.2-ubuntu20.04
+
+WORKDIR /
+COPY e2e2/test/images/pytorch_tests/ ./pytorch_tests

e2e2/test/images/pytorch_tests/singleNodeTest.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+
+torchrun --nproc_per_node=2 --nnodes=1 pytorch_tests/testNeuronSingleAllReduce.py
+torchrun --nproc_per_node=2 --nnodes=1 pytorch_tests/testNeuronParallelState.py
+torchrun --nproc_per_node=2 --nnodes=1 pytorch_tests/testNeuronMlp.py