Files
logpile/internal/parser/vendors/nvidia/parser_test.go

292 lines
8.4 KiB
Go

package nvidia
import (
"os"
"path/filepath"
"testing"
"time"
"git.mchus.pro/mchus/logpile/internal/parser"
)
func TestNVIDIAParser_RealArchive(t *testing.T) {
// Test with the real archive that was reported as problematic
archivePath := filepath.Join("../../../../example", "A514359X5A09844_logs-20260115-151707.tar")
// Check if file exists
if _, err := os.Stat(archivePath); os.IsNotExist(err) {
t.Skip("Test archive not found, skipping test")
}
// Extract files from archive
files, err := parser.ExtractArchive(archivePath)
if err != nil {
t.Fatalf("Failed to extract archive: %v", err)
}
// Check if inventory/output.log exists
hasInventoryLog := false
for _, f := range files {
if filepath.Base(f.Path) == "output.log" {
t.Logf("Found file: %s", f.Path)
}
if f.Path == "./inventory/output.log" || f.Path == "inventory/output.log" {
hasInventoryLog = true
t.Logf("Found inventory/output.log with %d bytes", len(f.Content))
}
}
if !hasInventoryLog {
t.Error("inventory/output.log not found in extracted files")
}
// Create parser and parse
p := &Parser{}
result, err := p.Parse(files)
if err != nil {
t.Fatalf("Failed to parse archive: %v", err)
}
// Verify basic system info
if result.Hardware.BoardInfo.Manufacturer == "" {
t.Error("Expected Manufacturer to be set")
}
if result.Hardware.BoardInfo.ProductName == "" {
t.Error("Expected ProductName to be set")
}
if result.Hardware.BoardInfo.SerialNumber == "" {
t.Error("Expected SerialNumber to be set")
}
t.Logf("System Info:")
t.Logf(" Manufacturer: %s", result.Hardware.BoardInfo.Manufacturer)
t.Logf(" Product: %s", result.Hardware.BoardInfo.ProductName)
t.Logf(" Serial: %s", result.Hardware.BoardInfo.SerialNumber)
// Verify GPUs were found
if len(result.Hardware.GPUs) == 0 {
t.Error("Expected to find GPUs")
}
t.Logf("\nFound %d GPUs:", len(result.Hardware.GPUs))
gpusWithSerials := 0
for _, gpu := range result.Hardware.GPUs {
t.Logf(" %s: %s (Firmware: %s, Serial: %s, BDF: %s)",
gpu.Slot, gpu.Model, gpu.Firmware, gpu.SerialNumber, gpu.BDF)
if gpu.SerialNumber != "" {
gpusWithSerials++
}
}
// Verify that GPU serial numbers were extracted
if gpusWithSerials == 0 {
t.Error("Expected at least some GPUs to have serial numbers")
}
t.Logf("\nGPUs with serial numbers: %d/%d", gpusWithSerials, len(result.Hardware.GPUs))
// Check events for SXM2 failures
t.Logf("\nTotal events: %d", len(result.Events))
// Look for the specific serial or SXM2
sxm2Events := 0
for _, event := range result.Events {
desc := event.Description + " " + event.RawData + " " + event.EventType
if contains(desc, "SXM2") || contains(desc, "1653925025827") {
t.Logf(" SXM2 Event: [%s] %s (Severity: %s)", event.EventType, event.Description, event.Severity)
sxm2Events++
}
}
if sxm2Events == 0 {
t.Error("Expected to find events for SXM2 (faulty GPU 1653925025827)")
}
t.Logf("\nSXM2 failure events: %d", sxm2Events)
}
func TestNVIDIAParser_GPUStatusFromSummary_RealArchive07900(t *testing.T) {
archivePath := filepath.Join("../../../../example", "A514359X5A07900_logs-20260122-074208.tar")
if _, err := os.Stat(archivePath); os.IsNotExist(err) {
t.Skip("Test archive not found, skipping test")
}
files, err := parser.ExtractArchive(archivePath)
if err != nil {
t.Fatalf("Failed to extract archive: %v", err)
}
p := &Parser{}
result, err := p.Parse(files)
if err != nil {
t.Fatalf("Failed to parse archive: %v", err)
}
if result.Hardware == nil || len(result.Hardware.GPUs) == 0 {
t.Fatalf("expected GPUs in parsed result")
}
statusBySerial := make(map[string]string, len(result.Hardware.GPUs))
for _, gpu := range result.Hardware.GPUs {
if gpu.SerialNumber != "" {
statusBySerial[gpu.SerialNumber] = gpu.Status
}
}
if got := statusBySerial["1653925025497"]; got != "FAIL" {
t.Fatalf("expected GPU serial 1653925025497 status FAIL, got %q", got)
}
for serial, st := range statusBySerial {
if serial == "1653925025497" {
continue
}
if st != "PASS" {
t.Fatalf("expected non-failing GPU serial %s status PASS, got %q", serial, st)
}
}
}
func TestNVIDIAParser_GPUErrorDetailsFromSummary_RealArchive07900(t *testing.T) {
archivePath := filepath.Join("../../../../example", "A514359X5A07900_logs-20260122-074208.tar")
if _, err := os.Stat(archivePath); os.IsNotExist(err) {
t.Skip("Test archive not found, skipping test")
}
files, err := parser.ExtractArchive(archivePath)
if err != nil {
t.Fatalf("Failed to extract archive: %v", err)
}
p := &Parser{}
result, err := p.Parse(files)
if err != nil {
t.Fatalf("Failed to parse archive: %v", err)
}
if result.Hardware == nil || len(result.Hardware.GPUs) == 0 {
t.Fatalf("expected GPUs in parsed result")
}
errBySerial := make(map[string]string, len(result.Hardware.GPUs))
for _, gpu := range result.Hardware.GPUs {
if gpu.SerialNumber != "" {
errBySerial[gpu.SerialNumber] = gpu.ErrorDescription
}
}
if got := errBySerial["1653925025497"]; got != "Row remapping failed" {
t.Fatalf("expected GPU serial 1653925025497 error Row remapping failed, got %q", got)
}
}
func TestNVIDIAParser_GPUModelFromSKU_RealArchive07900(t *testing.T) {
archivePath := filepath.Join("../../../../example", "A514359X5A07900_logs-20260122-074208.tar")
if _, err := os.Stat(archivePath); os.IsNotExist(err) {
t.Skip("Test archive not found, skipping test")
}
files, err := parser.ExtractArchive(archivePath)
if err != nil {
t.Fatalf("Failed to extract archive: %v", err)
}
p := &Parser{}
result, err := p.Parse(files)
if err != nil {
t.Fatalf("Failed to parse archive: %v", err)
}
if result.Hardware == nil || len(result.Hardware.GPUs) == 0 {
t.Fatalf("expected GPUs in parsed result")
}
found := false
for _, gpu := range result.Hardware.GPUs {
if gpu.Model == "692-2G520-0280-501" && gpu.Description == "hgx h200 8 gpu 141g aircooled" {
found = true
break
}
}
if !found {
t.Fatalf("expected at least one GPU with model 692-2G520-0280-501 and description hgx h200 8 gpu 141g aircooled")
}
}
func TestNVIDIAParser_ComponentCheckTimes_RealArchive07900(t *testing.T) {
archivePath := filepath.Join("../../../../example", "A514359X5A07900_logs-20260122-074208.tar")
if _, err := os.Stat(archivePath); os.IsNotExist(err) {
t.Skip("Test archive not found, skipping test")
}
files, err := parser.ExtractArchive(archivePath)
if err != nil {
t.Fatalf("Failed to extract archive: %v", err)
}
p := &Parser{}
result, err := p.Parse(files)
if err != nil {
t.Fatalf("Failed to parse archive: %v", err)
}
if result.Hardware == nil {
t.Fatalf("expected hardware in parsed result")
}
expectedGPU := time.Date(2026, 1, 22, 6, 45, 36, 0, time.UTC)
expectedNVSwitch := time.Date(2026, 1, 22, 6, 11, 32, 0, time.UTC)
if len(result.Hardware.GPUs) == 0 {
t.Fatalf("expected GPUs in parsed result")
}
for _, gpu := range result.Hardware.GPUs {
if !gpu.StatusCheckedAt.Equal(expectedGPU) {
t.Fatalf("expected GPU %s status_checked_at %s, got %s", gpu.Slot, expectedGPU.Format(time.RFC3339), gpu.StatusCheckedAt.Format(time.RFC3339))
}
if gpu.StatusAtCollect == nil || !gpu.StatusAtCollect.At.Equal(expectedGPU) {
t.Fatalf("expected GPU %s status_at_collection.at %s", gpu.Slot, expectedGPU.Format(time.RFC3339))
}
}
nvsCount := 0
for _, dev := range result.Hardware.PCIeDevices {
slot := normalizeNVSwitchSlot(dev.Slot)
if slot == "" {
continue
}
if dev.DeviceClass != "NVSwitch" && len(slot) < len("NVSWITCH") {
continue
}
if dev.DeviceClass != "NVSwitch" && slot[:len("NVSWITCH")] != "NVSWITCH" {
continue
}
nvsCount++
if !dev.StatusCheckedAt.Equal(expectedNVSwitch) {
t.Fatalf("expected NVSwitch %s status_checked_at %s, got %s", dev.Slot, expectedNVSwitch.Format(time.RFC3339), dev.StatusCheckedAt.Format(time.RFC3339))
}
if dev.StatusAtCollect == nil || !dev.StatusAtCollect.At.Equal(expectedNVSwitch) {
t.Fatalf("expected NVSwitch %s status_at_collection.at %s", dev.Slot, expectedNVSwitch.Format(time.RFC3339))
}
}
if nvsCount == 0 {
t.Fatalf("expected NVSwitch devices in parsed result")
}
}
func contains(s, substr string) bool {
return len(s) >= len(substr) && (s == substr || len(s) > len(substr) &&
(s[:len(substr)] == substr || s[len(s)-len(substr):] == substr ||
findSubstring(s, substr)))
}
func findSubstring(s, substr string) bool {
for i := 0; i <= len(s)-len(substr); i++ {
if s[i:i+len(substr)] == substr {
return true
}
}
return false
}