EKS Nvidia Test Use A GPU Instance Type (#407)

aws · May 3, 2024 · c4a3a4a · c4a3a4a
1 parent a273a24
commit c4a3a4a
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 4 deletions.
diff --git a/generator/resources/eks_daemon_test_matrix.json b/generator/resources/eks_daemon_test_matrix.json
@@ -2,7 +2,7 @@
   {
     "k8s_version": "1.24",
     "ami": "AL2_x86_64",
-    "instanceType":"t3a.medium",
+    "instanceType":"t3.medium",
     "arc": "amd64"
   },
   {

diff --git a/generator/test_case_generator.go b/generator/test_case_generator.go
@@ -41,6 +41,7 @@ type testConfig struct {
 	// e.g. statsd can have a multiple terraform module sets for difference test scenarios (ecs, eks or ec2)
 	testDir       string
 	terraformDir  string
+	instanceType  string
 	runMockServer bool
 	// define target matrix field as set(s)
 	// empty map means a testConfig will be created with a test entry for each entry from *_test_matrix.json
@@ -216,7 +217,8 @@ 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": {}}},
+			targets:      map[string]map[string]struct{}{"arc": {"amd64": {}}},
+			instanceType: "g4dn.xlarge",
 		},
 	},
 	"eks_deployment": {
@@ -304,6 +306,10 @@ func genMatrix(testType string, testConfigs []testConfig, ami []string) []matrix
 				log.Panicf("can't decode map test %v to metric line struct with error %v", testConfig, err)
 			}
 
+			if testConfig.instanceType != "" {
+				row.InstanceType = testConfig.instanceType
+			}
+
 			if len(ami) != 0 && !slices.Contains(ami, row.Ami) {
 				continue
 			}

diff --git a/terraform/eks/daemon/gpu/main.tf b/terraform/eks/daemon/gpu/main.tf
@@ -47,10 +47,10 @@ resource "aws_eks_node_group" "this" {
     min_size     = 1
   }
 
-  ami_type       = "AL2_x86_64"
+  ami_type       = var.ami_type
   capacity_type  = "ON_DEMAND"
   disk_size      = 20
-  instance_types = ["t3.medium"]
+  instance_types = [var.instance_type]
 
   depends_on = [
     aws_iam_role_policy_attachment.node_AmazonEC2ContainerRegistryReadOnly,