feat: add test cases to check vGPU license status (#696)

* feat: add test cases to check vGPU license status

Author: shvbsle

test/cases/nvidia/unit_test.go

@@ -7,6 +7,7 @@ import (
 	_ "embed"
 	"fmt"
 	"testing"
+	"time"
 
 	fwext "github.com/aws/aws-k8s-tester/internal/e2e"
 	"sigs.k8s.io/e2e-framework/klient/wait"
@@ -66,7 +67,8 @@ func TestSingleNodeUnitTest(t *testing.T) {
 				ObjectMeta: metav1.ObjectMeta{Name: "unit-test-job", Namespace: "default"},
 			}
 			err := wait.For(fwext.NewConditionExtension(cfg.Client().Resources()).JobSucceeded(job),
-				wait.WithContext(ctx))
+				wait.WithContext(ctx),
+				wait.WithTimeout(10*time.Minute))
 			if err != nil {
 				t.Fatal(err)
 			}

test/images/nvidia/gpu_unit_tests/tests/common.sh

@@ -53,3 +53,12 @@ generate_data()
     eval "$cmd" > $expected
     _assert_data "$expected" "$cmd" "$msg"
 }
+
+function is_vgpu()
+{
+  local instance_type=${EC2_INSTANCE_TYPE:-$(get_instance_type)}
+  case "${instance_type}" in
+    g6f.*|gr6f.*) return ;;
+    *) return 1 ;;  # Not supported
+  esac
+}

test/images/nvidia/gpu_unit_tests/tests/test_basic.sh

@@ -31,6 +31,11 @@ test_03_nvbandwidth()
 
 test_04_dcgm_diagnostics()
 {
+    # This test is not applicable for vGPU instance types.
+    if is_vgpu; then
+        skip "This test does not apply to vGPU instances (g6f.*, gr6f.*)"
+    fi
+
     # https://docs.nvidia.com/datacenter/dcgm/latest/user-guide/dcgm-diagnostics.html#run-levels-and-tests
     if [[ $EC2_INSTANCE_TYPE == g* ]]; then
         # The G series instance don't have nvlink and GPU p2p communication

test/images/nvidia/gpu_unit_tests/tests/test_sysinfo.sh

@@ -61,9 +61,26 @@ test_nvidia_gpu_unused()
 test_nvidia_gpu_throttled()
 {
 
+    # vGPU instances don't support GPU clock throttling detection.
+    # This test is not applicable for vGPU instance types.
+    if is_vgpu; then
+        skip "This test does not apply to vGPU instances (g6f.*, gr6f.*)"
+    fi
     # https://docs.nvidia.com/deploy/nvml-api/group__nvmlClocksEventReasons.html#group__nvmlClocksEventReasons
     # The only  bit allowed is nvmlClocksEventReasonGpuIdle 0x0000000000000001LL
     filter="egrep -v -e '(0x0000000000000000|0x0000000000000001|0x0000000000000004)'"
     cmd="nvidia-smi --query-gpu index,gpu_bus_id,gpu_uuid,clocks_throttle_reasons.active --format=csv,noheader"
     assert_status_code 1 "$cmd | $filter" "Throttled gpu detected"
 }
+
+
+test_nvidia_vgpu_license_status()
+{
+    if ! is_vgpu; then
+        skip "This test only applies to vGPU instances (g6f.*, gr6f.*)"
+    fi
+
+    assert_data $data/nvidia_vgpu_license_status.txt \
+          "nvidia-smi -q | grep 'vGPU Software' -A 2" \
+          "vGPU license status validation failed"
+}

test/images/nvidia/gpu_unit_tests/tests/test_sysinfo.sh.data/g6f.2xlarge/efa_count.txt

@@ -0,0 +1 @@
+0

test/images/nvidia/gpu_unit_tests/tests/test_sysinfo.sh.data/g6f.2xlarge/gpu_count.txt

@@ -0,0 +1,2 @@
+name, index, pci.bus_id
+NVIDIA L4-6Q, 0, 00000000:31:00.0

test/images/nvidia/gpu_unit_tests/tests/test_sysinfo.sh.data/g6f.2xlarge/numa_topo.txt

@@ -0,0 +1,2 @@
+/sys/devices/system/node/node0/cpulist:0-7
+/sys/devices/system/node/node0/distance:10

test/images/nvidia/gpu_unit_tests/tests/test_sysinfo.sh.data/g6f.2xlarge/nvidia_persistence_status.txt

@@ -0,0 +1,2 @@
+name, pci.bus_id, persistence_mode
+NVIDIA L4-6Q, 00000000:31:00.0, Enabled

test/images/nvidia/gpu_unit_tests/tests/test_sysinfo.sh.data/g6f.2xlarge/nvidia_smi_topo.txt

@@ -0,0 +1,2 @@
+	GPU0	CPU Affinity	NUMA Affinity	GPU NUMA ID
+GPU0	 X 	0-7	0		N/A

test/images/nvidia/gpu_unit_tests/tests/test_sysinfo.sh.data/g6f.2xlarge/nvidia_vgpu_license_status.txt

@@ -0,0 +1,3 @@
+    vGPU Software Licensed Product
+        Product Name                      : NVIDIA RTX Virtual Workstation
+        License Status                    : Licensed (Expiry: N/A)