test: fix NVIDIA OSS tests

Add more logging output.

Force non-UEFI boot.

Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>

.github/workflows/ci.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2025-04-10T08:42:48Z by kres d903dae.
+# Generated on 2025-04-15T15:57:38Z by kres fd5cab0.
 
 name: default
 concurrency:
@@ -634,7 +634,6 @@ jobs:
       - name: e2e-aws-nvidia-nonfree
         env:
           EXTRA_TEST_ARGS: -talos.extensions.nvidia
-          TEST_NUM_NODES: "4"
         run: |
           make e2e-aws
       - name: tf destroy
@@ -788,8 +787,7 @@ jobs:
           make e2e-cloud-tf
       - name: e2e-aws-nvidia-oss
         env:
-          EXTRA_TEST_ARGS: -talos.extensions.nvidia
-          TEST_NUM_NODES: "4"
+          EXTRA_TEST_ARGS: -talos.extensions.nvidia -talos.verifyukibooted=false
         run: |
           make e2e-aws
       - name: tf destroy

.github/workflows/integration-aws-nvidia-nonfree-cron.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2025-04-01T08:14:24Z by kres d903dae.
+# Generated on 2025-04-15T15:57:38Z by kres fd5cab0.
 
 name: integration-aws-nvidia-nonfree-cron
 concurrency:
@@ -144,7 +144,6 @@ jobs:
       - name: e2e-aws-nvidia-nonfree
         env:
           EXTRA_TEST_ARGS: -talos.extensions.nvidia
-          TEST_NUM_NODES: "4"
         run: |
           make e2e-aws
       - name: tf destroy

.github/workflows/integration-aws-nvidia-oss-cron.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2025-04-04T11:30:51Z by kres d903dae.
+# Generated on 2025-04-15T15:57:38Z by kres fd5cab0.
 
 name: integration-aws-nvidia-oss-cron
 concurrency:
@@ -143,8 +143,7 @@ jobs:
           make e2e-cloud-tf
       - name: e2e-aws-nvidia-oss
         env:
-          EXTRA_TEST_ARGS: -talos.extensions.nvidia
-          TEST_NUM_NODES: "4"
+          EXTRA_TEST_ARGS: -talos.extensions.nvidia -talos.verifyukibooted=false
         run: |
           make e2e-aws
       - name: tf destroy

.kres.yaml

@@ -2562,8 +2562,7 @@ spec:
         - name: e2e-aws-nvidia-oss
           command: e2e-aws
           environment:
-            TEST_NUM_NODES: 4
-            EXTRA_TEST_ARGS: -talos.extensions.nvidia
+            EXTRA_TEST_ARGS: "-talos.extensions.nvidia -talos.verifyukibooted=false"
         - name: tf destroy
           command: e2e-cloud-tf
           conditions:
@@ -2660,7 +2659,6 @@ spec:
         - name: e2e-aws-nvidia-nonfree
           command: e2e-aws
           environment:
-            TEST_NUM_NODES: 4
             EXTRA_TEST_ARGS: -talos.extensions.nvidia
         - name: tf destroy
           command: e2e-cloud-tf

hack/cloud-image-uploader/aws.go

@@ -384,7 +384,7 @@ func (au *AWSUploader) registerAMIArch(ctx context.Context, region string, svc *
 		log.Printf("aws: import into %s/%s, deregistered image ID %q", region, arch, *image.ImageId)
 	}
 
-	registerResp, err := svc.RegisterImage(ctx, &ec2.RegisterImageInput{
+	registerReq := &ec2.RegisterImageInput{
 		Name: aws.String(imageName),
 		BlockDeviceMappings: []types.BlockDeviceMapping{
 			{
@@ -404,8 +404,13 @@ func (au *AWSUploader) registerAMIArch(ctx context.Context, region string, svc *
 		Description:        pointer.To(fmt.Sprintf("Talos AMI %s %s %s", au.Options.Tag, arch, region)),
 		Architecture:       awsArchitectures[arch],
 		ImdsSupport:        types.ImdsSupportValuesV20,
-		BootMode:           types.BootModeValuesUefiPreferred,
-	})
+	}
+
+	if !au.Options.AWSForceBIOS {
+		registerReq.BootMode = types.BootModeValuesUefiPreferred
+	}
+
+	registerResp, err := svc.RegisterImage(ctx, registerReq)
 	if err != nil {
 		return err
 	}

hack/cloud-image-uploader/main.go

@@ -64,6 +64,7 @@ func run() error {
 	pflag.StringVar(&DefaultOptions.NamePrefix, "name-prefix", DefaultOptions.NamePrefix, "prefix for the name of the uploaded image")
 
 	pflag.StringSliceVar(&DefaultOptions.AWSRegions, "aws-regions", DefaultOptions.AWSRegions, "list of AWS regions to upload to")
+	pflag.BoolVar(&DefaultOptions.AWSForceBIOS, "aws-force-bios", DefaultOptions.AWSForceBIOS, "force BIOS boot mode for AWS images")
 
 	pflag.Parse()
 

hack/cloud-image-uploader/options.go

@@ -18,7 +18,8 @@ type Options struct {
 	TargetClouds  []string
 
 	// AWS options.
-	AWSRegions []string
+	AWSRegions   []string
+	AWSForceBIOS bool
 }
 
 // DefaultOptions used throughout the cli.

hack/test/e2e-aws-prepare.sh

@@ -9,6 +9,12 @@ REGION="us-east-1"
 function cloud_image_upload() {
   CLOUD_IMAGES_EXTRA_ARGS=("--name-prefix=${1}" "--target-clouds=aws" "--architectures=amd64" "--aws-regions=${REGION}")
 
+  case "${1}" in
+    talos-e2e-nvidia-oss)
+      CLOUD_IMAGES_EXTRA_ARGS+=("--aws-force-bios")
+      ;;
+  esac
+
   make cloud-images CLOUD_IMAGES_EXTRA_ARGS="${CLOUD_IMAGES_EXTRA_ARGS[*]}"
 }
 

hack/test/e2e.sh

@@ -37,7 +37,6 @@ export KUBERNETES_VERSION=${KUBERNETES_VERSION:-1.33.0-rc.0}
 
 export NAME_PREFIX="talos-e2e-${SHA}-${PLATFORM}"
 export TIMEOUT=1200
-export NUM_NODES=${TEST_NUM_NODES:-6}
 
 # default values, overridden by talosctl cluster create tests
 PROVISIONER=

internal/integration/api/extensions_nvidia.go

@@ -53,6 +53,8 @@ func (suite *ExtensionsSuiteNVIDIA) TearDownTest() {
 }
 
 // TestExtensionsNVIDIA verifies that a cuda workload can be run.
+//
+//nolint:gocyclo
 func (suite *ExtensionsSuiteNVIDIA) TestExtensionsNVIDIA() {
 	expectedModulesModDep := map[string]string{
 		"nvidia":         "nvidia.ko",
@@ -126,6 +128,14 @@ func (suite *ExtensionsSuiteNVIDIA) TestExtensionsNVIDIA() {
 				return retry.ExpectedErrorf("error getting pod: %s", listErr)
 			}
 
+			for _, pod := range podList.Items {
+				if pod.Status.Phase == corev1.PodFailed {
+					logData := suite.getPodLogs("default", pod.Name)
+
+					suite.T().Logf("pod %s logs:\n%s", pod.Name, logData)
+				}
+			}
+
 			if len(podList.Items) != 1 {
 				return retry.ExpectedErrorf("expected 1 pod, got %d", len(podList.Items))
 			}
@@ -147,19 +157,25 @@ func (suite *ExtensionsSuiteNVIDIA) TestExtensionsNVIDIA() {
 	suite.Require().Len(podList.Items, 1)
 
 	for _, pod := range podList.Items {
-		res := suite.Clientset.CoreV1().Pods("default").GetLogs(pod.Name, &corev1.PodLogOptions{})
-		stream, err := res.Stream(suite.ctx)
-		suite.Require().NoError(err)
+		logData := suite.getPodLogs("default", pod.Name)
 
-		defer stream.Close() //nolint:errcheck
-
-		logData, err := io.ReadAll(stream)
-		suite.Require().NoError(err)
-
-		suite.Require().Contains(string(logData), "Test PASSED")
+		suite.Require().Contains(logData, "Test PASSED")
 	}
 }
 
+func (suite *ExtensionsSuiteNVIDIA) getPodLogs(namespace, name string) string {
+	res := suite.Clientset.CoreV1().Pods(namespace).GetLogs(name, &corev1.PodLogOptions{})
+	stream, err := res.Stream(suite.ctx)
+	suite.Require().NoError(err)
+
+	defer stream.Close() //nolint:errcheck
+
+	logData, err := io.ReadAll(stream)
+	suite.Require().NoError(err)
+
+	return string(logData)
+}
+
 func (suite *ExtensionsSuiteNVIDIA) getNVIDIANodes(labelQuery string) []string {
 	nodes, err := suite.Clientset.CoreV1().Nodes().List(suite.ctx, metav1.ListOptions{
 		LabelSelector: labelQuery,