bee/audit/internal/collector/nvidia_test.go

package collector

import (
	"bee/audit/internal/schema"
	"testing"
)

func TestParseNVIDIASMIQuery(t *testing.T) {
	raw := "0, 00000000:65:00.0, NVIDIA H100 80GB HBM3, GPU-SERIAL-1, 96.00.1F.00.02, 54, 210.33, 0, 5, Not Active, 4, 4, 16, 16\n"
	byBDF, err := parseNVIDIASMIQuery(raw)
	if err != nil {
		t.Fatalf("parse failed: %v", err)
	}

	gpu, ok := byBDF["0000:65:00.0"]
	if !ok {
		t.Fatalf("gpu by normalized bdf not found")
	}
	if gpu.Name != "NVIDIA H100 80GB HBM3" {
		t.Fatalf("name: got %q", gpu.Name)
	}
	if gpu.Serial != "GPU-SERIAL-1" {
		t.Fatalf("serial: got %q", gpu.Serial)
	}
	if gpu.VBIOS != "96.00.1F.00.02" {
		t.Fatalf("vbios: got %q", gpu.VBIOS)
	}
	if gpu.ECCUncorrected == nil || *gpu.ECCUncorrected != 0 {
		t.Fatalf("ecc uncorrected: got %v", gpu.ECCUncorrected)
	}
	if gpu.HWSlowdown == nil || *gpu.HWSlowdown {
		t.Fatalf("hw slowdown: got %v, want false", gpu.HWSlowdown)
	}
	if gpu.PCIeLinkGenCurrent == nil || *gpu.PCIeLinkGenCurrent != 4 {
		t.Fatalf("pcie link gen current: got %v, want 4", gpu.PCIeLinkGenCurrent)
	}
	if gpu.PCIeLinkGenMax == nil || *gpu.PCIeLinkGenMax != 4 {
		t.Fatalf("pcie link gen max: got %v, want 4", gpu.PCIeLinkGenMax)
	}
}

func TestNormalizePCIeBDF(t *testing.T) {
	tests := []struct {
		in   string
		want string
	}{
		{"00000000:17:00.0", "0000:17:00.0"},
		{"0000:17:00.0", "0000:17:00.0"},
		{"17:00.0", "0000:17:00.0"},
	}
	for _, tt := range tests {
		got := normalizePCIeBDF(tt.in)
		if got != tt.want {
			t.Fatalf("normalizePCIeBDF(%q)=%q want %q", tt.in, got, tt.want)
		}
	}
}

func TestEnrichPCIeWithNVIDIAData_driverLoaded(t *testing.T) {
	vendorID := NvidiaVendorID
	bdf := "0000:65:00.0"
	manufacturer := "NVIDIA Corporation"
	status := "OK"
	devices := []schema.HardwarePCIeDevice{
		{
			HardwareComponentStatus: schema.HardwareComponentStatus{Status: &status},
			VendorID:                &vendorID,
			BDF:                     &bdf,
			Manufacturer:            &manufacturer,
		},
	}

	byBDF := map[string]nvidiaGPUInfo{
		"0000:65:00.0": {
			BDF:            "0000:65:00.0",
			Serial:         "GPU-ABC",
			VBIOS:          "96.00.1F.00.02",
			ECCUncorrected: ptrInt64(2),
			ECCCorrected:   ptrInt64(10),
			TemperatureC:   ptrFloat(55.5),
			PowerW:         ptrFloat(230.2),
		},
	}

	out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
	if out[0].SerialNumber == nil || *out[0].SerialNumber != "GPU-ABC" {
		t.Fatalf("serial: got %v", out[0].SerialNumber)
	}
	if out[0].Firmware == nil || *out[0].Firmware != "96.00.1F.00.02" {
		t.Fatalf("firmware: got %v", out[0].Firmware)
	}
	if out[0].Telemetry == nil || out[0].Telemetry["nvidia_gpu_index"] != 0 {
		t.Fatalf("telemetry nvidia_gpu_index: got %#v", out[0].Telemetry)
	}
	if out[0].Status == nil || *out[0].Status != statusWarning {
		t.Fatalf("status: got %v", out[0].Status)
	}
	if out[0].ECCUncorrectedTotal == nil || *out[0].ECCUncorrectedTotal != 2 {
		t.Fatalf("ecc_uncorrected_total: got %#v", out[0].ECCUncorrectedTotal)
	}
	if out[0].TemperatureC == nil || *out[0].TemperatureC != 55.5 {
		t.Fatalf("temperature_c: got %#v", out[0].TemperatureC)
	}
}

func TestEnrichPCIeWithNVIDIAData_driverMissingFallback(t *testing.T) {
	vendorID := NvidiaVendorID
	bdf := "0000:17:00.0"
	manufacturer := "NVIDIA Corporation"
	devices := []schema.HardwarePCIeDevice{
		{
			VendorID:     &vendorID,
			BDF:          &bdf,
			Manufacturer: &manufacturer,
		},
	}

	out := enrichPCIeWithNVIDIAData(devices, nil, false)
	if out[0].SerialNumber != nil {
		t.Fatalf("serial should stay nil without source data, got %v", out[0].SerialNumber)
	}
	if out[0].Status == nil || *out[0].Status != statusUnknown {
		t.Fatalf("fallback status: got %v", out[0].Status)
	}
}

func TestParseNVIDIAReliability(t *testing.T) {
	raw := "0000:65:00.0, No, 0, 2, No, Yes\n"
	byBDF, err := parseNVIDIAReliabilityCSV(raw)
	if err != nil {
		t.Fatalf("parse failed: %v", err)
	}

	gpu, ok := byBDF["0000:65:00.0"]
	if !ok {
		t.Fatalf("gpu by normalized bdf not found")
	}
	if gpu.ResetRequired == nil || *gpu.ResetRequired {
		t.Fatalf("reset_required: got %v, want false", gpu.ResetRequired)
	}
	if gpu.RemapCorrectable == nil || *gpu.RemapCorrectable != 0 {
		t.Fatalf("remap correctable: got %v", gpu.RemapCorrectable)
	}
	if gpu.RemapUncorrectable == nil || *gpu.RemapUncorrectable != 2 {
		t.Fatalf("remap uncorrectable: got %v", gpu.RemapUncorrectable)
	}
	if gpu.RemapPending == nil || *gpu.RemapPending {
		t.Fatalf("remap pending: got %v, want false", gpu.RemapPending)
	}
	if gpu.RemapFailure == nil || !*gpu.RemapFailure {
		t.Fatalf("remap failure: got %v, want true", gpu.RemapFailure)
	}
}

func TestEnrichPCIeWithNVIDIAData_resetRequiredCritical(t *testing.T) {
	vendorID := NvidiaVendorID
	bdf := "0000:65:00.0"
	devices := []schema.HardwarePCIeDevice{
		{VendorID: &vendorID, BDF: &bdf},
	}

	byBDF := map[string]nvidiaGPUInfo{
		"0000:65:00.0": {ResetRequired: ptrBool(true)},
	}

	out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
	if out[0].Status == nil || *out[0].Status != statusCritical {
		t.Fatalf("status: got %v, want %v", out[0].Status, statusCritical)
	}
	if out[0].ResetRequired == nil || !*out[0].ResetRequired {
		t.Fatalf("reset_required: got %v", out[0].ResetRequired)
	}
}

func TestEnrichPCIeWithNVIDIAData_remapFailureCritical(t *testing.T) {
	vendorID := NvidiaVendorID
	bdf := "0000:65:00.0"
	devices := []schema.HardwarePCIeDevice{
		{VendorID: &vendorID, BDF: &bdf},
	}

	byBDF := map[string]nvidiaGPUInfo{
		"0000:65:00.0": {RemapFailure: ptrBool(true)},
	}

	out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
	if out[0].Status == nil || *out[0].Status != statusCritical {
		t.Fatalf("status: got %v, want %v", out[0].Status, statusCritical)
	}
	if out[0].RemappedRowsFailure == nil || !*out[0].RemappedRowsFailure {
		t.Fatalf("remapped_rows_failure: got %v", out[0].RemappedRowsFailure)
	}
}

func TestEnrichPCIeWithNVIDIAData_preservesExistingPCIeLinkWarning(t *testing.T) {
	vendorID := NvidiaVendorID
	bdf := "0000:65:00.0"
	status := statusWarning
	desc := "PCIe link speed degraded: running at Gen1, capable of Gen4"
	devices := []schema.HardwarePCIeDevice{
		{
			HardwareComponentStatus: schema.HardwareComponentStatus{Status: &status, ErrorDescription: &desc},
			VendorID:                &vendorID,
			BDF:                     &bdf,
		},
	}

	// Clean ECC/remap/reset: NVIDIA enrichment itself sees nothing wrong.
	byBDF := map[string]nvidiaGPUInfo{
		"0000:65:00.0": {},
	}

	out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
	if out[0].Status == nil || *out[0].Status != statusWarning {
		t.Fatalf("status: got %v, want %v (must not be overwritten by clean NVIDIA telemetry)", out[0].Status, statusWarning)
	}
	if out[0].ErrorDescription == nil || *out[0].ErrorDescription != desc {
		t.Fatalf("error_description: got %v, want %q", out[0].ErrorDescription, desc)
	}
}

func TestEnrichPCIeWithNVIDIAData_eccWarningDoesNotDowngradeExistingWarning(t *testing.T) {
	vendorID := NvidiaVendorID
	bdf := "0000:65:00.0"
	status := statusWarning
	desc := "PCIe link speed degraded: running at Gen1, capable of Gen4"
	devices := []schema.HardwarePCIeDevice{
		{
			HardwareComponentStatus: schema.HardwareComponentStatus{Status: &status, ErrorDescription: &desc},
			VendorID:                &vendorID,
			BDF:                     &bdf,
		},
	}

	byBDF := map[string]nvidiaGPUInfo{
		"0000:65:00.0": {ECCUncorrected: ptrInt64(1)},
	}

	out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
	// ECCUncorrected only raises Warning (same severity as the pre-existing
	// PCIe-link Warning), so the merge must not lose the already-detected
	// (and equally severe) status/description.
	if out[0].Status == nil || *out[0].Status != statusWarning {
		t.Fatalf("status: got %v, want %v", out[0].Status, statusWarning)
	}
	if out[0].ErrorDescription == nil || *out[0].ErrorDescription != desc {
		t.Fatalf("error_description: got %v, want %q (equal-severity ECC finding must not clobber it)", out[0].ErrorDescription, desc)
	}
}

func TestEnrichPCIeWithNVIDIAData_resetRequiredEscalatesOverExistingWarning(t *testing.T) {
	vendorID := NvidiaVendorID
	bdf := "0000:65:00.0"
	status := statusWarning
	desc := "PCIe link speed degraded: running at Gen1, capable of Gen4"
	devices := []schema.HardwarePCIeDevice{
		{
			HardwareComponentStatus: schema.HardwareComponentStatus{Status: &status, ErrorDescription: &desc},
			VendorID:                &vendorID,
			BDF:                     &bdf,
		},
	}

	byBDF := map[string]nvidiaGPUInfo{
		"0000:65:00.0": {ResetRequired: ptrBool(true)},
	}

	out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
	if out[0].Status == nil || *out[0].Status != statusCritical {
		t.Fatalf("status: got %v, want %v (reset-required Critical must escalate over the PCIe-link Warning)", out[0].Status, statusCritical)
	}
	if out[0].ErrorDescription == nil || *out[0].ErrorDescription != "GPU requires a reset" {
		t.Fatalf("error_description: got %v", out[0].ErrorDescription)
	}
}

func TestEnrichPCIeWithNVIDIAData_regressionNoOverride(t *testing.T) {
	vendorID := NvidiaVendorID
	bdf := "0000:65:00.0"
	devices := []schema.HardwarePCIeDevice{
		{VendorID: &vendorID, BDF: &bdf},
	}

	byBDF := map[string]nvidiaGPUInfo{
		"0000:65:00.0": {},
	}

	out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
	if out[0].Status == nil || *out[0].Status != statusOK {
		t.Fatalf("status: got %v, want %v", out[0].Status, statusOK)
	}
	if out[0].ErrorDescription != nil {
		t.Fatalf("error_description: got %v, want nil", out[0].ErrorDescription)
	}
}

func ptrInt64(v int64) *int64     { return &v }
func ptrFloat(v float64) *float64 { return &v }
func ptrBool(v bool) *bool        { return &v }