reanimator/bee
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity

Sync hardware ingest contract v2.7

Browse Source
This commit is contained in:
Mikhail Chusavitin
2026-03-15 23:03:38 +03:00
parent 72cf482ad3
commit 78c6dfc0ef
12 changed files with 200 additions and 176 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
audit/internal/collector/collector.go
View File

@@ -24,7 +24,7 @@ func Run(_ runtimeenv.Mode) schema.HardwareIngestRequest {
snap.Board = board
snap.Firmware = append(snap.Firmware, biosFW...)
snap.CPUs = collectCPUs(snap.Board.SerialNumber)
snap.CPUs = collectCPUs()
snap.Memory = collectMemory()
sensorDoc, err := readSensorsJSONDoc()
@@ -35,7 +35,7 @@ func Run(_ runtimeenv.Mode) schema.HardwareIngestRequest {
snap.Memory = enrichMemoryWithTelemetry(snap.Memory, sensorDoc)
snap.Storage = collectStorage()
snap.PCIeDevices = collectPCIe()
snap.PCIeDevices = enrichPCIeWithNVIDIA(snap.PCIeDevices, snap.Board.SerialNumber)
snap.PCIeDevices = enrichPCIeWithNVIDIA(snap.PCIeDevices)
snap.PCIeDevices = enrichPCIeWithMellanox(snap.PCIeDevices)
snap.PCIeDevices = enrichPCIeWithNICTelemetry(snap.PCIeDevices)
snap.PCIeDevices = enrichPCIeWithRAIDTelemetry(snap.PCIeDevices)

16
audit/internal/collector/cpu.go
View File

@@ -3,7 +3,6 @@ package collector
import (
"bee/audit/internal/schema"
"bufio"
"fmt"
"log/slog"
"os"
"path/filepath"
@@ -12,14 +11,14 @@ import (
)
// collectCPUs runs dmidecode -t 4 and enriches CPUs with microcode from sysfs.
func collectCPUs(boardSerial string) []schema.HardwareCPU {
func collectCPUs() []schema.HardwareCPU {
out, err := runDmidecode("4")
if err != nil {
slog.Warn("cpu: dmidecode type 4 failed", "err", err)
return nil
}
cpus := parseCPUs(out, boardSerial)
cpus := parseCPUs(out)
if mc := readMicrocode(); mc != "" {
for i := range cpus {
cpus[i].Firmware = &mc
@@ -31,12 +30,12 @@ func collectCPUs(boardSerial string) []schema.HardwareCPU {
}
// parseCPUs splits dmidecode output into per-processor sections and parses each.
func parseCPUs(output, boardSerial string) []schema.HardwareCPU {
func parseCPUs(output string) []schema.HardwareCPU {
sections := splitDMISections(output, "Processor Information")
cpus := make([]schema.HardwareCPU, 0, len(sections))
for _, section := range sections {
cpu, ok := parseCPUSection(section, boardSerial)
cpu, ok := parseCPUSection(section)
if !ok {
continue
}
@@ -47,7 +46,7 @@ func parseCPUs(output, boardSerial string) []schema.HardwareCPU {
// parseCPUSection parses one "Processor Information" block into a HardwareCPU.
// Returns false if the socket is unpopulated.
func parseCPUSection(fields map[string]string, boardSerial string) (schema.HardwareCPU, bool) {
func parseCPUSection(fields map[string]string) (schema.HardwareCPU, bool) {
status := parseCPUStatus(fields["Status"])
if status == statusEmpty {
return schema.HardwareCPU{}, false
@@ -70,11 +69,6 @@ func parseCPUSection(fields map[string]string, boardSerial string) (schema.Hardw
}
if v := cleanDMIValue(fields["Serial Number"]); v != "" {
cpu.SerialNumber = &v
} else if boardSerial != "" && cpu.Socket != nil {
// Intel Xeon never exposes serial via DMI — generate stable fallback
// matching core's generateCPUVendorSerial() logic
fb := fmt.Sprintf("%s-CPU-%d", boardSerial, *cpu.Socket)
cpu.SerialNumber = &fb
}
if v := parseMHz(fields["Max Speed"]); v > 0 {

15
audit/internal/collector/cpu_test.go
View File

@@ -8,7 +8,7 @@ import (
func TestParseCPUs_dual_socket(t *testing.T) {
out := mustReadFile(t, "testdata/dmidecode_type4.txt")
cpus := parseCPUs(out, "CAR315KA0803B90")
cpus := parseCPUs(out)
if len(cpus) != 2 {
t.Fatalf("expected 2 CPUs, got %d", len(cpus))
@@ -39,23 +39,22 @@ func TestParseCPUs_dual_socket(t *testing.T) {
if cpu0.Status == nil || *cpu0.Status != "OK" {
t.Errorf("cpu0 status: got %v, want OK", cpu0.Status)
}
// Intel Xeon serial not available → fallback
if cpu0.SerialNumber == nil || *cpu0.SerialNumber != "CAR315KA0803B90-CPU-0" {
t.Errorf("cpu0 serial fallback: got %v, want CAR315KA0803B90-CPU-0", cpu0.SerialNumber)
if cpu0.SerialNumber != nil {
t.Errorf("cpu0 serial should stay nil without source data, got %v", cpu0.SerialNumber)
}
cpu1 := cpus[1]
if cpu1.Socket == nil || *cpu1.Socket != 1 {
t.Errorf("cpu1 socket: got %v, want 1", cpu1.Socket)
}
if cpu1.SerialNumber == nil || *cpu1.SerialNumber != "CAR315KA0803B90-CPU-1" {
t.Errorf("cpu1 serial fallback: got %v, want CAR315KA0803B90-CPU-1", cpu1.SerialNumber)
if cpu1.SerialNumber != nil {
t.Errorf("cpu1 serial should stay nil without source data, got %v", cpu1.SerialNumber)
}
}
func TestParseCPUs_unpopulated_skipped(t *testing.T) {
out := mustReadFile(t, "testdata/dmidecode_type4_disabled.txt")
cpus := parseCPUs(out, "BOARD-001")
cpus := parseCPUs(out)
if len(cpus) != 1 {
t.Fatalf("expected 1 CPU (unpopulated skipped), got %d", len(cpus))
@@ -87,7 +86,7 @@ func TestCollectCPUsSetsFirmwareFromMicrocode(t *testing.T) {
}
t.Cleanup(func() { execDmidecode = origRun })
cpus := collectCPUs("CAR315KA0803B90")
cpus := collectCPUs()
if len(cpus) != 2 {
t.Fatalf("expected 2 CPUs, got %d", len(cpus))
}

104
audit/internal/collector/finalize.go
View File

@@ -1,9 +1,6 @@
package collector
import (
"bee/audit/internal/schema"
"fmt"
)
import "bee/audit/internal/schema"
func finalizeSnapshot(snap *schema.HardwareSnapshot, collectedAt string) {
snap.Memory = filterMemory(snap.Memory)
@@ -11,7 +8,6 @@ func finalizeSnapshot(snap *schema.HardwareSnapshot, collectedAt string) {
snap.PowerSupplies = filterPSUs(snap.PowerSupplies)
setComponentStatusMetadata(snap, collectedAt)
deduplicateComponentSerials(snap)
}
func filterMemory(dimms []schema.HardwareMemory) []schema.HardwareMemory {
@@ -79,101 +75,3 @@ func setStatusCheckedAt(status *schema.HardwareComponentStatus, collectedAt stri
status.StatusCheckedAt = &collectedAt
}
}
func deduplicateComponentSerials(snap *schema.HardwareSnapshot) {
deduplicateCPUSerials(snap.CPUs)
deduplicateMemorySerials(snap.Memory)
deduplicateStorageSerials(snap.Storage)
deduplicatePCIeSerials(snap.PCIeDevices)
deduplicatePSUSerials(snap.PowerSupplies)
}
func deduplicateCPUSerials(items []schema.HardwareCPU) {
seen := map[string]int{}
seq := 1
for i := range items {
if items[i].SerialNumber == nil || *items[i].SerialNumber == "" {
continue
}
model := derefString(items[i].Model)
key := model + "\x00" + *items[i].SerialNumber
seen[key]++
if seen[key] > 1 {
repl := fmt.Sprintf("NO_SN-%08d", seq)
seq++
items[i].SerialNumber = &repl
}
}
}
func deduplicateMemorySerials(items []schema.HardwareMemory) {
seen := map[string]int{}
seq := 1
for i := range items {
if items[i].SerialNumber == nil || *items[i].SerialNumber == "" {
continue
}
model := derefString(items[i].PartNumber)
key := model + "\x00" + *items[i].SerialNumber
seen[key]++
if seen[key] > 1 {
repl := fmt.Sprintf("NO_SN-%08d", seq)
seq++
items[i].SerialNumber = &repl
}
}
}
func deduplicateStorageSerials(items []schema.HardwareStorage) {
seen := map[string]int{}
seq := 1
for i := range items {
if items[i].SerialNumber == nil || *items[i].SerialNumber == "" {
continue
}
model := derefString(items[i].Model)
key := model + "\x00" + *items[i].SerialNumber
seen[key]++
if seen[key] > 1 {
repl := fmt.Sprintf("NO_SN-%08d", seq)
seq++
items[i].SerialNumber = &repl
}
}
}
func deduplicatePCIeSerials(items []schema.HardwarePCIeDevice) {
seen := map[string]int{}
seq := 1
for i := range items {
if items[i].SerialNumber == nil || *items[i].SerialNumber == "" {
continue
}
model := derefString(items[i].Model)
key := model + "\x00" + *items[i].SerialNumber
seen[key]++
if seen[key] > 1 {
repl := fmt.Sprintf("NO_SN-%08d", seq)
seq++
items[i].SerialNumber = &repl
}
}
}
func deduplicatePSUSerials(items []schema.HardwarePowerSupply) {
seen := map[string]int{}
seq := 1
for i := range items {
if items[i].SerialNumber == nil || *items[i].SerialNumber == "" {
continue
}
model := derefString(items[i].Model)
key := model + "\x00" + *items[i].SerialNumber
seen[key]++
if seen[key] > 1 {
repl := fmt.Sprintf("NO_SN-%08d", seq)
seq++
items[i].SerialNumber = &repl
}
}
}

6
audit/internal/collector/finalize_test.go
View File

@@ -39,7 +39,7 @@ func TestFinalizeSnapshotFiltersComponentsWithoutRequiredSerials(t *testing.T) {
}
}
func TestFinalizeSnapshotDeduplicatesSerials(t *testing.T) {
func TestFinalizeSnapshotPreservesDuplicateSerials(t *testing.T) {
collectedAt := "2026-03-15T12:00:00Z"
status := statusOK
model := "Device"
@@ -57,7 +57,7 @@ func TestFinalizeSnapshotDeduplicatesSerials(t *testing.T) {
if got := *snap.Storage[0].SerialNumber; got != serial {
t.Fatalf("first serial changed: %q", got)
}
if got := *snap.Storage[1].SerialNumber; got != "NO_SN-00000001" {
t.Fatalf("duplicate serial mismatch: %q", got)
if got := *snap.Storage[1].SerialNumber; got != serial {
t.Fatalf("duplicate serial should stay unchanged: %q", got)
}
}

34
audit/internal/collector/nvidia.go
View File

@@ -24,18 +24,17 @@ type nvidiaGPUInfo struct {
}
// enrichPCIeWithNVIDIA enriches NVIDIA PCIe devices with data from nvidia-smi.
// If the driver/tool is unavailable, NVIDIA devices get Unknown status and
// a stable serial fallback based on board serial + slot.
func enrichPCIeWithNVIDIA(devs []schema.HardwarePCIeDevice, boardSerial string) []schema.HardwarePCIeDevice {
// If the driver/tool is unavailable, NVIDIA devices get Unknown status.
func enrichPCIeWithNVIDIA(devs []schema.HardwarePCIeDevice) []schema.HardwarePCIeDevice {
if !hasNVIDIADevices(devs) {
return devs
}
gpuByBDF, err := queryNVIDIAGPUs()
if err != nil {
slog.Info("nvidia: enrichment skipped", "err", err)
return enrichPCIeWithNVIDIAData(devs, nil, boardSerial, false)
return enrichPCIeWithNVIDIAData(devs, nil, false)
}
return enrichPCIeWithNVIDIAData(devs, gpuByBDF, boardSerial, true)
return enrichPCIeWithNVIDIAData(devs, gpuByBDF, true)
}
func hasNVIDIADevices(devs []schema.HardwarePCIeDevice) bool {
@@ -47,7 +46,7 @@ func hasNVIDIADevices(devs []schema.HardwarePCIeDevice) bool {
return false
}
func enrichPCIeWithNVIDIAData(devs []schema.HardwarePCIeDevice, gpuByBDF map[string]nvidiaGPUInfo, boardSerial string, driverLoaded bool) []schema.HardwarePCIeDevice {
func enrichPCIeWithNVIDIAData(devs []schema.HardwarePCIeDevice, gpuByBDF map[string]nvidiaGPUInfo, driverLoaded bool) []schema.HardwarePCIeDevice {
enriched := 0
for i := range devs {
if !isNVIDIADevice(devs[i]) {
@@ -55,7 +54,7 @@ func enrichPCIeWithNVIDIAData(devs []schema.HardwarePCIeDevice, gpuByBDF map[str
}
if !driverLoaded {
setPCIeFallback(&devs[i], boardSerial)
setPCIeFallback(&devs[i])
continue
}
@@ -65,14 +64,12 @@ func enrichPCIeWithNVIDIAData(devs []schema.HardwarePCIeDevice, gpuByBDF map[str
}
info, ok := gpuByBDF[bdf]
if !ok {
setPCIeFallback(&devs[i], boardSerial)
setPCIeFallback(&devs[i])
continue
}
if v := strings.TrimSpace(info.Serial); v != "" {
devs[i].SerialNumber = &v
} else {
setPCIeFallbackSerial(&devs[i], boardSerial)
}
if v := strings.TrimSpace(info.VBIOS); v != "" {
devs[i].Firmware = &v
@@ -213,26 +210,11 @@ func isNVIDIADevice(dev schema.HardwarePCIeDevice) bool {
return false
}
func setPCIeFallback(dev *schema.HardwarePCIeDevice, boardSerial string) {
setPCIeFallbackSerial(dev, boardSerial)
func setPCIeFallback(dev *schema.HardwarePCIeDevice) {
status := statusUnknown
dev.Status = &status
}
func setPCIeFallbackSerial(dev *schema.HardwarePCIeDevice, boardSerial string) {
if strings.TrimSpace(boardSerial) == "" || dev.SerialNumber != nil {
return
}
slot := "unknown"
if dev.BDF != nil && strings.TrimSpace(*dev.BDF) != "" {
slot = strings.TrimSpace(*dev.BDF)
} else if dev.Slot != nil && strings.TrimSpace(*dev.Slot) != "" {
slot = strings.TrimSpace(*dev.Slot)
}
fb := fmt.Sprintf("%s-PCIE-%s", boardSerial, slot)
dev.SerialNumber = &fb
}
func injectNVIDIATelemetry(dev *schema.HardwarePCIeDevice, info nvidiaGPUInfo) {
if info.TemperatureC != nil {
dev.TemperatureC = info.TemperatureC

8
audit/internal/collector/nvidia_test.go
View File

@@ -73,7 +73,7 @@ func TestEnrichPCIeWithNVIDIAData_driverLoaded(t *testing.T) {
},
}
out := enrichPCIeWithNVIDIAData(devices, byBDF, "BOARD-001", true)
out := enrichPCIeWithNVIDIAData(devices, byBDF, true)
if out[0].SerialNumber == nil || *out[0].SerialNumber != "GPU-ABC" {
t.Fatalf("serial: got %v", out[0].SerialNumber)
}
@@ -103,9 +103,9 @@ func TestEnrichPCIeWithNVIDIAData_driverMissingFallback(t *testing.T) {
},
}
out := enrichPCIeWithNVIDIAData(devices, nil, "BOARD-123", false)
if out[0].SerialNumber == nil || *out[0].SerialNumber != "BOARD-123-PCIE-0000:17:00.0" {
t.Fatalf("fallback serial: got %v", out[0].SerialNumber)
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)

1
audit/internal/collector/pcie.go
View File

@@ -84,6 +84,7 @@ func parseLspciDevice(fields map[string]string) schema.HardwarePCIeDevice {
// Slot is the BDF: "0000:00:02.0"
if bdf := fields["Slot"]; bdf != "" {
dev.Slot = &bdf
dev.BDF = &bdf
// parse vendor_id and device_id from sysfs
vendorID, deviceID := readPCIIDs(bdf)

29
audit/internal/collector/pcie_filter_test.go
View File

@@ -1,6 +1,10 @@
package collector
import "testing"
import (
"encoding/json"
"strings"
"testing"
)
func TestShouldIncludePCIeDevice(t *testing.T) {
tests := []struct {
@@ -38,6 +42,29 @@ func TestParseLspci_filtersExcludedClasses(t *testing.T) {
if devs[0].DeviceClass == nil || *devs[0].DeviceClass != "VideoController" {
t.Fatalf("unexpected remaining class: %v", devs[0].DeviceClass)
}
if devs[0].Slot == nil || *devs[0].Slot != "0000:65:00.0" {
t.Fatalf("slot: got %v", devs[0].Slot)
}
if devs[0].BDF == nil || *devs[0].BDF != "0000:65:00.0" {
t.Fatalf("bdf: got %v", devs[0].BDF)
}
}
func TestPCIeJSONUsesSlotNotBDF(t *testing.T) {
input := "Slot:\t0000:65:00.0\nClass:\tVGA compatible controller\nVendor:\tNVIDIA Corporation\nDevice:\tH100\n\n"
devs := parseLspci(input)
data, err := json.Marshal(devs[0])
if err != nil {
t.Fatalf("marshal: %v", err)
}
text := string(data)
if !strings.Contains(text, `"slot":"0000:65:00.0"`) {
t.Fatalf("json missing slot: %s", text)
}
if strings.Contains(text, `"bdf"`) {
t.Fatalf("json should not emit bdf: %s", text)
}
}
func TestNormalizePCILinkSpeed(t *testing.T) {

26
audit/internal/schema/hardware.go
View File

@@ -22,6 +22,7 @@ type HardwareSnapshot struct {
PCIeDevices []HardwarePCIeDevice `json:"pcie_devices,omitempty"`
PowerSupplies []HardwarePowerSupply `json:"power_supplies,omitempty"`
Sensors *HardwareSensors `json:"sensors,omitempty"`
EventLogs []HardwareEventLog `json:"event_logs,omitempty"`
}
type HardwareHealthSummary struct {
@@ -148,7 +149,7 @@ type HardwarePCIeDevice struct {
SFPRXPowerDBM *float64 `json:"sfp_rx_power_dbm,omitempty"`
SFPVoltageV *float64 `json:"sfp_voltage_v,omitempty"`
SFPBiasMA *float64 `json:"sfp_bias_ma,omitempty"`
BDF *string `json:"bdf,omitempty"`
BDF *string `json:"-"`
DeviceClass *string `json:"device_class,omitempty"`
Manufacturer *string `json:"manufacturer,omitempty"`
Model *string `json:"model,omitempty"`
@@ -183,11 +184,12 @@ type HardwarePowerSupply struct {
}
type HardwareComponentStatus struct {
Status *string `json:"status,omitempty"`
StatusCheckedAt *string `json:"status_checked_at,omitempty"`
StatusChangedAt *string `json:"status_changed_at,omitempty"`
StatusHistory []HardwareStatusHistory `json:"status_history,omitempty"`
ErrorDescription *string `json:"error_description,omitempty"`
Status *string `json:"status,omitempty"`
StatusCheckedAt *string `json:"status_checked_at,omitempty"`
StatusChangedAt *string `json:"status_changed_at,omitempty"`
StatusHistory []HardwareStatusHistory `json:"status_history,omitempty"`
ErrorDescription *string `json:"error_description,omitempty"`
ManufacturedYearWeek *string `json:"manufactured_year_week,omitempty"`
}
type HardwareStatusHistory struct {
@@ -235,3 +237,15 @@ type HardwareOtherSensor struct {
Unit *string `json:"unit,omitempty"`
Status *string `json:"status,omitempty"`
}
type HardwareEventLog struct {
Source string `json:"source"`
EventTime *string `json:"event_time,omitempty"`
Severity *string `json:"severity,omitempty"`
MessageID *string `json:"message_id,omitempty"`
Message string `json:"message"`
ComponentRef *string `json:"component_ref,omitempty"`
Fingerprint *string `json:"fingerprint,omitempty"`
IsActive *bool `json:"is_active,omitempty"`
RawPayload map[string]any `json:"raw_payload,omitempty"`
}

46
audit/internal/schema/hardware_test.go Normal file
View File

@@ -0,0 +1,46 @@
package schema
import (
"encoding/json"
"strings"
"testing"
)
func TestHardwareSnapshotMarshalsNewContractFields(t *testing.T) {
week := "2024-W07"
eventTime := "2026-03-15T14:03:11Z"
message := "Correctable ECC error threshold exceeded"
payload := HardwareIngestRequest{
CollectedAt: "2026-03-15T15:00:00Z",
Hardware: HardwareSnapshot{
Board: HardwareBoard{SerialNumber: "SRV-001"},
CPUs: []HardwareCPU{
{
HardwareComponentStatus: HardwareComponentStatus{
ManufacturedYearWeek: &week,
},
},
},
EventLogs: []HardwareEventLog{
{
Source: "bmc",
EventTime: &eventTime,
Message: message,
},
},
},
}
data, err := json.Marshal(payload)
if err != nil {
t.Fatalf("marshal: %v", err)
}
text := string(data)
if !strings.Contains(text, `"manufactured_year_week":"2024-W07"`) {
t.Fatalf("missing manufactured_year_week: %s", text)
}
if !strings.Contains(text, `"event_logs":[{"source":"bmc","event_time":"2026-03-15T14:03:11Z","message":"Correctable ECC error threshold exceeded"}]`) {
t.Fatalf("missing event_logs payload: %s", text)
}
}