Merge pull request #1636 from pwschuurman/add-support-nvme-tests

Add support for running tests on confidential VMs that use NVMe

Author: k8s-ci-robot

test/e2e/tests/setup_e2e_test.go

@@ -36,16 +36,19 @@ import (
 )
 
 var (
-	project          = flag.String("project", "", "Project to run tests in")
-	serviceAccount   = flag.String("service-account", "", "Service account to bring up instance with")
-	architecture     = flag.String("arch", "amd64", "Architecture pd csi driver build on")
-	zones            = flag.String("zones", "us-east4-a,us-east4-c", "Zones to run tests in. If there are multiple zones, separate each by comma")
-	machineType      = flag.String("machine-type", "n2-standard-2", "Type of machine to provision instance on")
-	imageURL         = flag.String("image-url", "projects/debian-cloud/global/images/family/debian-11", "OS image url to get image from")
-	runInProw        = flag.Bool("run-in-prow", false, "If true, use a Boskos loaned project and special CI service accounts and ssh keys")
-	deleteInstances  = flag.Bool("delete-instances", false, "Delete the instances after tests run")
-	cloudtopHost     = flag.Bool("cloudtop-host", false, "The local host is cloudtop, a kind of googler machine with special requirements to access GCP")
-	extraDriverFlags = flag.String("extra-driver-flags", "", "Extra flags to pass to the driver")
+	project                   = flag.String("project", "", "Project to run tests in")
+	serviceAccount            = flag.String("service-account", "", "Service account to bring up instance with")
+	vmNamePrefix              = flag.String("vm-name-prefix", "gce-pd-csi-e2e", "VM name prefix")
+	architecture              = flag.String("arch", "amd64", "Architecture pd csi driver build on")
+	minCpuPlatform            = flag.String("min-cpu-platform", "AMD Milan", "Minimum CPU architecture")
+	zones                     = flag.String("zones", "us-east4-a,us-east4-c", "Zones to run tests in. If there are multiple zones, separate each by comma")
+	machineType               = flag.String("machine-type", "n2d-standard-2", "Type of machine to provision instance on")
+	imageURL                  = flag.String("image-url", "projects/ubuntu-os-cloud/global/images/family/ubuntu-minimal-2310-amd64", "OS image url to get image from")
+	runInProw                 = flag.Bool("run-in-prow", false, "If true, use a Boskos loaned project and special CI service accounts and ssh keys")
+	deleteInstances           = flag.Bool("delete-instances", false, "Delete the instances after tests run")
+	cloudtopHost              = flag.Bool("cloudtop-host", false, "The local host is cloudtop, a kind of googler machine with special requirements to access GCP")
+	extraDriverFlags          = flag.String("extra-driver-flags", "", "Extra flags to pass to the driver")
+	enableConfidentialCompute = flag.Bool("enable-confidential-compute", false, "Create VMs with confidential compute mode. This uses NVMe devices")
 
 	testContexts        = []*remote.TestContext{}
 	computeService      *compute.Service
@@ -130,21 +133,24 @@ func getDriverConfig() testutils.DriverConfig {
 	}
 }
 
-func getRemoteInstanceConfig() *remote.InstanceConfig {
-	return &remote.InstanceConfig{
-		Project:        *project,
-		Architecture:   *architecture,
-		MachineType:    *machineType,
-		ServiceAccount: *serviceAccount,
-		ImageURL:       *imageURL,
-		CloudtopHost:   *cloudtopHost}
-}
-
 func NewTestContext(zone string) *remote.TestContext {
-	nodeID := fmt.Sprintf("gce-pd-csi-e2e-%s", zone)
+	nodeID := fmt.Sprintf("%s-%s", *vmNamePrefix, zone)
 	klog.Infof("Setting up node %s", nodeID)
 
-	i, err := remote.SetupInstance(getRemoteInstanceConfig(), zone, nodeID, computeService)
+	instanceConfig := remote.InstanceConfig{
+		Project:                   *project,
+		Architecture:              *architecture,
+		MinCpuPlatform:            *minCpuPlatform,
+		Zone:                      zone,
+		Name:                      nodeID,
+		MachineType:               *machineType,
+		ServiceAccount:            *serviceAccount,
+		ImageURL:                  *imageURL,
+		CloudtopHost:              *cloudtopHost,
+		EnableConfidentialCompute: *enableConfidentialCompute,
+		ComputeService:            computeService,
+	}
+	i, err := remote.SetupInstance(instanceConfig)
 	if err != nil {
 		klog.Fatalf("Failed to setup instance %v: %v", nodeID, err)
 	}

test/e2e/tests/single_zone_e2e_test.go

@@ -76,7 +76,7 @@ var _ = Describe("GCE PD CSI Driver", func() {
 		Expect(volumeLimit).To(Equal(defaultVolumeLimit))
 	})
 
-	It("Should create->attach->stage->mount volume and check if it is writable, then unmount->unstage->detach->delete and check disk is deleted", func() {
+	It("[NVMe] Should create->attach->stage->mount volume and check if it is writable, then unmount->unstage->detach->delete and check disk is deleted", func() {
 		testContext := getRandomTestContext()
 
 		p, z, _ := testContext.Instance.GetIdentity()
@@ -173,7 +173,7 @@ var _ = Describe("GCE PD CSI Driver", func() {
 		}()
 	})
 
-	It("Should automatically add a symlink between /dev/* and /dev/by-id if disk is not found", func() {
+	It("[NVMe] Should automatically add a symlink between /dev/* and /dev/by-id if disk is not found", func() {
 		testContext := getRandomTestContext()
 
 		p, z, _ := testContext.Instance.GetIdentity()
@@ -331,7 +331,7 @@ var _ = Describe("GCE PD CSI Driver", func() {
 		Entry("on pd-ssd", ssdDiskType),
 	)
 
-	DescribeTable("Should complete publish/unpublish lifecycle with underspecified volume ID and missing volume",
+	DescribeTable("[NVMe] Should complete publish/unpublish lifecycle with underspecified volume ID and missing volume",
 		func(diskType string) {
 			testContext := getRandomTestContext()
 
@@ -1317,15 +1317,7 @@ var _ = Describe("GCE PD CSI Driver", func() {
 		_, err := getRandomTestContext().Client.ListVolumes()
 		Expect(err).To(BeNil(), "no error expected when passed valid compute url")
 
-		zone := "us-central1-c"
-		nodeID := fmt.Sprintf("gce-pd-csi-e2e-%s", zone)
-		i, err := remote.SetupInstance(getRemoteInstanceConfig(), zone, nodeID, computeService)
-
-		if err != nil {
-			klog.Fatalf("Failed to setup instance %v: %v", nodeID, err)
-		}
-
-		klog.Infof("Creating new driver and client for node %s\n", i.GetName())
+		i := getRandomTestContext().Instance
 
 		// Create new driver and client with valid, empty endpoint
 		klog.Infof("Setup driver with empty compute endpoint %s\n", i.GetName())
@@ -1349,7 +1341,7 @@ var _ = Describe("GCE PD CSI Driver", func() {
 		Expect(err).To(BeNil(), "no error expected when passed valid compute url")
 	})
 
-	It("Should update readahead if read_ahead_kb passed on mount", func() {
+	It("[NVMe] Should update readahead if read_ahead_kb passed on mount", func() {
 		testContext := getRandomTestContext()
 
 		p, z, _ := testContext.Instance.GetIdentity()
@@ -1410,7 +1402,7 @@ var _ = Describe("GCE PD CSI Driver", func() {
 				Expect(err).To(BeNil(), "Failed to symlink devicePath")
 				devFsPathPieces := strings.Split(devFsPath, "/")
 				devName = devFsPathPieces[len(devFsPathPieces)-1]
-
+				break
 			}
 		}
 		Expect(validated).To(BeTrue(), "could not find device in %v that links to volume %s", devicePaths, volName)

test/e2e/utils/utils.go

@@ -293,11 +293,13 @@ func ValidateLogicalLinkIsDisk(instance *remote.InstanceInfo, link, diskName str
 
 	devFsPath, err := instance.SSH("find", link, "-printf", "'%l'")
 	if err != nil {
-		return false, fmt.Errorf("failed to find symbolic link for %s. Output: %v, errror: %v", link, devFsPath, err.Error())
+		// Skip over if there is no matching symlink.
+		return false, nil
 	}
 	if len(devFsPath) == 0 {
 		return false, nil
 	}
+
 	if sdx := sdRegex.FindString(devFsPath); len(sdx) != 0 {
 		fullDevPath := path.Join("/dev/", string(sdx))
 		scsiIDOut, err := instance.SSH("/lib/udev_containerized/scsi_id", "--page=0x83", "--whitelisted", fmt.Sprintf("--device=%v", fullDevPath))