adding e2e

generator/test_case_generator.go

@@ -217,8 +217,7 @@ var testTypeToTestConfig = map[string][]testConfig{
 		{testDir: "./test/fluent", terraformDir: "terraform/eks/daemon/fluent/windows/2022"},
 		{
 			testDir: "./test/gpu", terraformDir: "terraform/eks/daemon/gpu",
-			targets:      map[string]map[string]struct{}{"arc": {"amd64": {}}},
-			instanceType: "g4dn.xlarge",
+			targets: map[string]map[string]struct{}{"arc": {"amd64": {}}},
 		},
 	},
 	"eks_deployment": {

terraform/eks/daemon/gpu/main.tf

@@ -2,13 +2,13 @@
 // SPDX-License-Identifier: MIT
 
 module "common" {
-  source             = "../../../common"
+  source             = "../common"
   cwagent_image_repo = var.cwagent_image_repo
   cwagent_image_tag  = var.cwagent_image_tag
 }
 
 module "basic_components" {
-  source = "../../../basic_components"
+  source = "../basic_components"
 
   region = var.region
 }

terraform/eks/daemon/gpu/variables.tf

@@ -8,7 +8,7 @@ variable "region" {
 
 variable "test_dir" {
   type    = string
-  default = "./test/gpu"
+  default = "../test/gpu"
 }
 
 variable "cwagent_image_repo" {

terraform/gpu/main.tf

@@ -0,0 +1,146 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: MIT
+
+module "common" {
+  source             = "../common"
+  cwagent_image_repo = var.cwagent_image_repo
+  cwagent_image_tag  = var.cwagent_image_tag
+}
+
+module "basic_components" {
+  source = "../basic_components"
+
+  region = var.region
+}
+
+
+data "aws_eks_cluster_auth" "this" {
+  name = aws_eks_cluster.this.name
+}
+
+locals {
+  role_arn = format("%s%s", module.basic_components.role_arn, var.beta ? "-eks-beta" : "")
+  aws_eks  = format("%s%s", "aws eks --region ${var.region}", var.beta ? " --endpoint ${var.beta_endpoint}" : "")
+}
+
+resource "aws_eks_cluster" "this" {
+  name     = "cwagent-operator-eks-integ-${module.common.testing_id}"
+  role_arn = local.role_arn
+  version  = var.k8s_version
+  vpc_config {
+    subnet_ids         = module.basic_components.public_subnet_ids
+    security_group_ids = [module.basic_components.security_group]
+  }
+}
+
+# EKS Node Groups
+resource "aws_eks_node_group" "this" {
+  cluster_name    = aws_eks_cluster.this.name
+  node_group_name = "cwagent-operator-eks-integ-node"
+  node_role_arn   = aws_iam_role.node_role.arn
+  subnet_ids      = module.basic_components.public_subnet_ids
+
+  scaling_config {
+    desired_size = 1
+    max_size     = 1
+    min_size     = 1
+  }
+
+  ami_type       = "AL2_x86_64_GPU"
+  capacity_type  = "ON_DEMAND"
+  disk_size      = 20
+  instance_types = ["g4dn.xlarge"]
+
+  depends_on = [
+    aws_iam_role_policy_attachment.node_AmazonEC2ContainerRegistryReadOnly,
+    aws_iam_role_policy_attachment.node_AmazonEKS_CNI_Policy,
+    aws_iam_role_policy_attachment.node_AmazonEKSWorkerNodePolicy,
+    aws_iam_role_policy_attachment.node_CloudWatchAgentServerPolicy
+  ]
+}
+
+# EKS Node IAM Role
+resource "aws_iam_role" "node_role" {
+  name = "cwagent-operator-eks-Worker-Role-${module.common.testing_id}"
+
+  assume_role_policy = <<POLICY
+{
+  "Version": "2012-10-17",
+  "Statement": [
+    {
+      "Effect": "Allow",
+      "Principal": {
+        "Service": "ec2.amazonaws.com"
+      },
+      "Action": "sts:AssumeRole"
+    }
+  ]
+}
+POLICY
+}
+
+resource "aws_iam_role_policy_attachment" "node_AmazonEKSWorkerNodePolicy" {
+  policy_arn = "arn:aws:iam::aws:policy/AmazonEKSWorkerNodePolicy"
+  role       = aws_iam_role.node_role.name
+}
+
+resource "aws_iam_role_policy_attachment" "node_AmazonEKS_CNI_Policy" {
+  policy_arn = "arn:aws:iam::aws:policy/AmazonEKS_CNI_Policy"
+  role       = aws_iam_role.node_role.name
+}
+
+resource "aws_iam_role_policy_attachment" "node_AmazonEC2ContainerRegistryReadOnly" {
+  policy_arn = "arn:aws:iam::aws:policy/AmazonEC2ContainerRegistryReadOnly"
+  role       = aws_iam_role.node_role.name
+}
+
+resource "aws_iam_role_policy_attachment" "node_CloudWatchAgentServerPolicy" {
+  policy_arn = "arn:aws:iam::aws:policy/CloudWatchAgentServerPolicy"
+  role       = aws_iam_role.node_role.name
+}
+
+
+resource "null_resource" "kubectl" {
+  depends_on = [
+    aws_eks_cluster.this,
+    aws_eks_node_group.this
+  ]
+  provisioner "local-exec" {
+    command = <<-EOT
+      ${local.aws_eks} update-kubeconfig --name ${aws_eks_cluster.this.name}
+      ${local.aws_eks} list-clusters --output text
+      ${local.aws_eks} describe-cluster --name ${aws_eks_cluster.this.name} --output text
+    EOT
+  }
+}
+
+resource "aws_eks_addon" "this" {
+  depends_on = [
+    null_resource.kubectl
+  ]
+  addon_name   = var.addon_name
+  cluster_name = aws_eks_cluster.this.name
+  addon_version = var.addon_version
+}
+
+resource "null_resource" "validator" {
+  depends_on = [
+      aws_eks_node_group.this,
+      aws_eks_addon.this
+  ]
+
+  provisioner "local-exec" {
+    command = <<EOT
+      go test ${var.test_dir} -eksClusterName ${aws_eks_cluster.this.name} -computeType=EKS -v -eksDeploymentStrategy=DAEMON -eksGpuType=nvidia
+
+      # Get all pods and describe them
+      kubectl get pods --all-namespaces -o wide > pods.txt
+      kubectl describe pods --all-namespaces > pods_describe.txt
+
+      # Log the contents of the files
+      cat pods.txt
+      cat pods_describe.txt
+    EOT
+  }
+}
+

terraform/gpu/providers.tf

@@ -0,0 +1,20 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: MIT
+
+provider "aws" {
+  region = var.region
+  endpoints {
+    eks = var.beta ? var.beta_endpoint : null
+  }
+}
+
+provider "kubernetes" {
+  exec {
+    api_version = "client.authentication.k8s.io/v1beta1"
+    command     = "aws"
+    args        = ["eks", "get-token", "--cluster-name", aws_eks_cluster.this.name]
+  }
+  host                   = aws_eks_cluster.this.endpoint
+  cluster_ca_certificate = base64decode(aws_eks_cluster.this.certificate_authority.0.data)
+  token                  = data.aws_eks_cluster_auth.this.token
+}

terraform/gpu/variables.tf

@@ -0,0 +1,58 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: MIT
+
+variable "region" {
+  type    = string
+  default = "us-west-2"
+}
+
+variable "test_dir" {
+  type    = string
+  default = "../../test/gpu"
+}
+
+variable "addon_name" {
+  type    = string
+  default = "amazon-cloudwatch-observability"
+}
+
+variable "addon_version" {
+  type = string
+  default = "v1.6.0-eksbuild.1"
+}
+
+
+variable "cwagent_image_repo" {
+  type    = string
+  default = "public.ecr.aws/cloudwatch-agent/cloudwatch-agent"
+}
+
+variable "cwagent_image_tag" {
+  type    = string
+  default = "latest"
+}
+
+variable "k8s_version" {
+  type    = string
+  default = "1.28"
+}
+
+variable "ami_type" {
+  type    = string
+  default = "AL2_x86_64_GPU"
+}
+
+variable "instance_type" {
+  type    = string
+  default = "g4dn.xlarge"
+}
+
+variable "beta" {
+  type    = bool
+  default = true
+}
+
+variable "beta_endpoint" {
+  type    = string
+  default = "https://api.beta.us-west-2.wesley.amazonaws.com"
+}

test/metric/container_insights_util.go

@@ -58,6 +58,15 @@ func ValidateMetrics(env *environment.MetaData, metricFilter string, expectedDim
 		}
 		results = append(results, validateMetricsAvailability(dims, metrics, actual))
 		for _, m := range metrics {
+			// this is to prevent panic with rand.Intn when metrics are not yet ready in a cluster
+			if _, ok := actual[m]; !ok {
+				results = append(results, status.TestResult{
+					Name:   dims,
+					Status: status.FAILED,
+				})
+				log.Printf("ValidateMetrics failed with missing metric: %s", m)
+				continue
+			}
 			// pick a random dimension set to test metric data OR test all dimension sets which might be overkill
 			randIdx := rand.Intn(len(actual[m]))
 			results = append(results, validateMetricValue(m, actual[m][randIdx]))
@@ -123,7 +132,7 @@ func validateMetricsAvailability(dims string, expected []string, actual map[stri
 		Name:   dims,
 		Status: status.FAILED,
 	}
-	log.Printf("expected metrics: %d, actual metrics: %d", len(expected), len(actual))
+	log.Printf("expected metrics: %d, actual metrics: %d", len(expected), 3*len(actual))
 	if compareMetrics(expected, actual) {
 		testResult.Status = status.SUCCESSFUL
 	} else {