Merge pull request #433 from flatcar/sayan/add-nvidia-test

Add tests for the NVIDIA GPU tests

CHANGELOG.md

@@ -11,6 +11,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - `DefaultUser` parameter when registering a test to use a user different from `core` ([#424](https://github.com/flatcar/mantle/pull/424))
 - `systemd.sysext.custom-oem` for testing the activation of the OEM sysext image ([#423](https://github.com/flatcar/mantle/pull/423))
 - Kubernetes 1.27 tests ([#441](https://github.com/flatcar/mantle/pull/441))
+- Add tests for testing the installation/integrity of the NVIDIA drivers ([#433](https://github.com/flatcar/mantle/pull/433))
 
 ### Changed
 
@@ -67,7 +68,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - Ignition v3 support and tests ([#301](https://github.com/flatcar-linux/mantle/pull/301), [#311](https://github.com/flatcar-linux/mantle/pull/311))
 - Butane config support ([#318](https://github.com/flatcar-linux/mantle/pull/318))
 - GCP: support testing with GVNIC ([#322](https://github.com/flatcar-linux/mantle/pull/322))
-- `networkd` Ignition translation test ([#344](https://github.com/flatcar-linux/mantle/pull/334)) 
+- `networkd` Ignition translation test ([#344](https://github.com/flatcar-linux/mantle/pull/334))
 - kola test `cl.misc.falco` that tests falco kmod building ([#339](https://github.com/flatcar-linux/mantle/pull/339))
 - Kubernetes test for release 1.24.1 ([#337](https://github.com/flatcar-linux/mantle/pull/337))
 - Added storage abstraction for Equinix Metal tests (SSH can be used in addition of Google Cloud Storage) ([#340](https://github.com/flatcar-linux/mantle/pull/340))

kola/tests/misc/nvidia.go

@@ -0,0 +1,52 @@
+package misc
+
+import (
+	"bytes"
+	"fmt"
+	"time"
+
+	"github.com/coreos/pkg/capnslog"
+	"github.com/flatcar/mantle/kola"
+	"github.com/flatcar/mantle/kola/cluster"
+	"github.com/flatcar/mantle/kola/register"
+	"github.com/flatcar/mantle/util"
+)
+
+const (
+	CmdTimeout = time.Second * 300
+)
+
+var plog = capnslog.NewPackageLogger("github.com/flatcar/mantle", "kola/tests/misc")
+
+func init() {
+	register.Register(&register.Test{
+		Name:        "cl.misc.nvidia",
+		Run:         verifyNvidiaInstallation,
+		ClusterSize: 1,
+		Distros:     []string{"cl"},
+		// This test is to test the NVIDIA installation, limited to AZURE for now
+		Platforms:     []string{"azure"},
+		Architectures: []string{"amd64"},
+		Flags:         []register.Flag{register.NoEnableSelinux},
+	})
+}
+
+func verifyNvidiaInstallation(c cluster.TestCluster) {
+	if kola.AzureOptions.Size != "Standard_NC6s_v3" {
+		c.Skip("skipping due to wrong instance size")
+	}
+	m := c.Machines()[0]
+
+	nvidiaStatusRetry := func() error {
+		out, err := c.SSH(m, "systemctl is-active nvidia.service")
+		if !bytes.Contains(out, []byte("inactive")) {
+			return fmt.Errorf("nvidia.service: %q: %v", out, err)
+		}
+		return nil
+	}
+
+	if err := util.Retry(40, 15*time.Second, nvidiaStatusRetry); err != nil {
+		c.Fatal(err)
+	}
+	c.AssertCmdOutputContains(m, "/opt/bin/nvidia-smi", "Tesla")
+}