fix(parser): support Inspur HGX dump_<serial>_<timestamp>/ onekeylog layout
Fixes #20. This onekeylog variant has no devicefrusdr.log at all: FRU/sensors come from raw ipmitool text output, PCIe/GPU presence has a dedicated structural snapshot, SEL lives at a different path, and BMC component failures are logged separately from SEL/IDL. - Fall back to component/fru.txt (same FRU block format as devicefrusdr.log) and component/sensor.txt / sdr.txt (ipmitool sensor list / sdr elist) when devicefrusdr.log is absent. - Parse log/bmc/diagnose/OtrdDiagnoseComponent.json's PCIe Device Info array for GPU/PCIe presence and link state, independent of SEL/IDL alarm history; flag devices running below their negotiated max link speed/width as degraded with a Warning event. - Fall back to log/sel.csv (same format as selelist.csv) when selelist.csv is absent. - Parse log/bmc/commer-comp/{commerslot,commerhmc,commerswvr,commerswcpld} logs (including rotated *.tar.gz.N parts) into failure events, filtering known-noisy lines. - Collapse SEL events duplicated across sources by (timestamp, event_type, description). - Surface a CollectionError when FRU/sensors are still empty after all fallbacks, instead of silently returning an empty inventory. - Fix ParseFRU: a later placeholder "Product Serial : 0" / "Product Part Number : NULL" line in the same FRU block (e.g. SCM_FRU) was overwriting an already-parsed real Board Serial/Part Number. Verified against dump_23DB01633_20260727-1359.tar.gz (HGX B200, KR9288-X3): fru 0→21, sensors 0→303, 8 GPUs present at Gen5 x16 in slots 100-107. Deferred (not covered by this change): BIOS-change-settings context and BIOS POST codes from the same layout — see bible-local/10-decisions.md ADL-048. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
+26
@@ -176,6 +176,32 @@ func determineSELSeverity(sensorStr, eventDesc, status string) models.Severity {
|
||||
return models.SeverityInfo
|
||||
}
|
||||
|
||||
// dedupSELEvents collapses events reported more than once with the same
|
||||
// (timestamp, event_type, description) triple. Unlike ParseIDLLog's
|
||||
// dedup (which must keep recurring alarms with distinct timestamps), this
|
||||
// only removes true duplicates: the same SEL entry surfacing through more
|
||||
// than one source file for the exact same moment.
|
||||
func dedupSELEvents(events []models.Event) []models.Event {
|
||||
if len(events) == 0 {
|
||||
return events
|
||||
}
|
||||
seen := make(map[string]struct{}, len(events))
|
||||
out := make([]models.Event, 0, len(events))
|
||||
for _, e := range events {
|
||||
if e.Source != "SEL" {
|
||||
out = append(out, e)
|
||||
continue
|
||||
}
|
||||
key := e.Timestamp.String() + "|" + e.EventType + "|" + e.Description
|
||||
if _, ok := seen[key]; ok {
|
||||
continue
|
||||
}
|
||||
seen[key] = struct{}{}
|
||||
out = append(out, e)
|
||||
}
|
||||
return out
|
||||
}
|
||||
|
||||
// buildSELDescription builds human-readable description
|
||||
func buildSELDescription(eventDesc, status string) string {
|
||||
if status == "Asserted" || status == "Deasserted" {
|
||||
|
||||
Reference in New Issue
Block a user