mchus/logpile
1
0
Fork 0
Code Issues 3 Pull Requests Actions Packages Projects Releases 13 Wiki Activity

nvidia: improve component mapping, firmware, statuses and check times

Browse Source
This commit is contained in:
Michael Chus
2026-02-16 23:17:13 +03:00
parent 514da76ddb
commit b33cca5fcc
19 changed files with 2051 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
internal/parser/vendors/nvidia/parser.go vendored
View File

@@ -14,7 +14,7 @@ import (
// parserVersion - version of this parser module
// IMPORTANT: Increment this version when making changes to parser logic!
const parserVersion = "1.2.4"
const parserVersion = "1.3.0"
func init() {
parser.Register(&Parser{})
@@ -70,7 +70,7 @@ func (p *Parser) Detect(files []parser.ExtractedFile) int {
if strings.HasSuffix(path, "output.log") {
// Check if it contains dmidecode output
if strings.Contains(string(f.Content), "dmidecode") ||
strings.Contains(string(f.Content), "System Information") {
strings.Contains(string(f.Content), "System Information") {
confidence += 10
}
}
@@ -106,6 +106,8 @@ func (p *Parser) Parse(files []parser.ExtractedFile) (*models.AnalysisResult, er
GPUs: make([]models.GPU, 0),
}
gpuStatuses := make(map[string]string)
gpuFailureDetails := make(map[string]string)
nvswitchStatuses := make(map[string]string)
// Parse output.log first (contains dmidecode system info)
// Find the output.log file that contains dmidecode output
@@ -134,9 +136,26 @@ func (p *Parser) Parse(files []parser.ExtractedFile) (*models.AnalysisResult, er
}
}
// Parse inventory/inventory.log to enrich PCI BDF mapping for components.
inventoryInfoLog := findInventoryInfoLog(files)
if inventoryInfoLog != nil {
if err := ApplyInventoryPCIIDs(inventoryInfoLog.Content, result); err != nil {
_ = err
}
}
// Enhance GPU model names using SKU mapping from testspec + inventory summary.
ApplyGPUModelsFromSKU(files, result)
// Parse inventory/nvflash_verbose.log and apply firmware versions by BDF + IDs.
// This runs after GPU model/part-number enrichment so firmware tab uses final model labels.
nvflashVerbose := findNVFlashVerboseLog(files)
if nvflashVerbose != nil {
if err := ParseNVFlashVerboseLog(nvflashVerbose.Content, result); err != nil {
_ = err
}
}
// Parse summary.json (test results summary)
if f := parser.FindFileByName(files, "summary.json"); f != nil {
events := ParseSummaryJSON(f.Content)
@@ -144,6 +163,14 @@ func (p *Parser) Parse(files []parser.ExtractedFile) (*models.AnalysisResult, er
for componentID, status := range CollectGPUStatusesFromSummaryJSON(f.Content) {
gpuStatuses[componentID] = mergeGPUStatus(gpuStatuses[componentID], status)
}
for slot, status := range CollectNVSwitchStatusesFromSummaryJSON(f.Content) {
nvswitchStatuses[slot] = mergeGPUStatus(nvswitchStatuses[slot], status)
}
for componentID, detail := range CollectGPUFailureDetailsFromSummaryJSON(f.Content) {
if _, exists := gpuFailureDetails[componentID]; !exists && strings.TrimSpace(detail) != "" {
gpuFailureDetails[componentID] = strings.TrimSpace(detail)
}
}
}
// Parse summary.csv (alternative format)
@@ -153,10 +180,21 @@ func (p *Parser) Parse(files []parser.ExtractedFile) (*models.AnalysisResult, er
for componentID, status := range CollectGPUStatusesFromSummaryCSV(f.Content) {
gpuStatuses[componentID] = mergeGPUStatus(gpuStatuses[componentID], status)
}
for slot, status := range CollectNVSwitchStatusesFromSummaryCSV(f.Content) {
nvswitchStatuses[slot] = mergeGPUStatus(nvswitchStatuses[slot], status)
}
for componentID, detail := range CollectGPUFailureDetailsFromSummaryCSV(f.Content) {
if _, exists := gpuFailureDetails[componentID]; !exists && strings.TrimSpace(detail) != "" {
gpuFailureDetails[componentID] = strings.TrimSpace(detail)
}
}
}
// Apply per-GPU PASS/FAIL status derived from summary files.
ApplyGPUStatuses(result, gpuStatuses)
ApplyGPUFailureDetails(result, gpuFailureDetails)
ApplyNVSwitchStatuses(result, nvswitchStatuses)
ApplyGPUAndNVSwitchCheckTimes(result, CollectGPUAndNVSwitchCheckTimes(files))
// Parse GPU field diagnostics logs
gpuFieldiagFiles := parser.FindFileByPattern(files, "gpu_fieldiag/", ".log")
@@ -180,7 +218,7 @@ func findDmidecodeOutputLog(files []parser.ExtractedFile) *parser.ExtractedFile
// Check if it contains dmidecode output
content := string(f.Content)
if strings.Contains(content, "dmidecode") &&
strings.Contains(content, "System Information") {
strings.Contains(content, "System Information") {
return &f
}
}