Align hardware export with ingest contract

audit/internal/collector/memory_telemetry.go

@@ -0,0 +1,203 @@
+package collector
+
+import (
+	"bee/audit/internal/schema"
+	"os"
+	"path/filepath"
+	"sort"
+	"strconv"
+	"strings"
+)
+
+var edacBaseDir = "/sys/devices/system/edac/mc"
+
+type edacDIMMStats struct {
+	Label   string
+	CECount *int64
+	UECount *int64
+}
+
+func enrichMemoryWithTelemetry(dimms []schema.HardwareMemory, doc sensorsDoc) []schema.HardwareMemory {
+	if len(dimms) == 0 {
+		return dimms
+	}
+
+	tempByLabel := memoryTempsFromSensors(doc)
+	stats := readEDACStats()
+
+	for i := range dimms {
+		labelKeys := dimmMatchKeys(dimms[i].Slot, dimms[i].Location)
+
+		for _, key := range labelKeys {
+			if temp, ok := tempByLabel[key]; ok {
+				dimms[i].TemperatureC = &temp
+				break
+			}
+		}
+
+		for _, key := range labelKeys {
+			if stat, ok := stats[key]; ok {
+				if stat.CECount != nil {
+					dimms[i].CorrectableECCErrorCount = stat.CECount
+				}
+				if stat.UECount != nil {
+					dimms[i].UncorrectableECCErrorCount = stat.UECount
+				}
+				if stat.UECount != nil && *stat.UECount > 0 {
+					dimms[i].DataLossDetected = boolPtr(true)
+					status := statusCritical
+					dimms[i].Status = &status
+					if dimms[i].ErrorDescription == nil {
+						dimms[i].ErrorDescription = stringPtr("EDAC reports uncorrectable ECC errors")
+					}
+				} else if stat.CECount != nil && *stat.CECount > 0 && (dimms[i].Status == nil || *dimms[i].Status == statusOK) {
+					status := statusWarning
+					dimms[i].Status = &status
+					if dimms[i].ErrorDescription == nil {
+						dimms[i].ErrorDescription = stringPtr("EDAC reports correctable ECC errors")
+					}
+				}
+				break
+			}
+		}
+	}
+
+	return dimms
+}
+
+func memoryTempsFromSensors(doc sensorsDoc) map[string]float64 {
+	out := map[string]float64{}
+	if len(doc) == 0 {
+		return out
+	}
+	for chip, features := range doc {
+		for featureName, raw := range features {
+			feature, ok := raw.(map[string]any)
+			if !ok || classifySensorFeature(feature) != "temp" {
+				continue
+			}
+			if !isLikelyMemoryTemp(chip, featureName) {
+				continue
+			}
+			temp, ok := firstFeatureFloat(feature, "_input")
+			if !ok {
+				continue
+			}
+			key := canonicalLabel(featureName)
+			if key == "" {
+				continue
+			}
+			if _, exists := out[key]; !exists {
+				out[key] = temp
+			}
+		}
+	}
+	return out
+}
+
+func readEDACStats() map[string]edacDIMMStats {
+	out := map[string]edacDIMMStats{}
+	mcDirs, err := filepath.Glob(filepath.Join(edacBaseDir, "mc*"))
+	if err != nil {
+		return out
+	}
+	sort.Strings(mcDirs)
+	for _, mcDir := range mcDirs {
+		dimmDirs, err := filepath.Glob(filepath.Join(mcDir, "dimm*"))
+		if err != nil {
+			continue
+		}
+		sort.Strings(dimmDirs)
+		for _, dimmDir := range dimmDirs {
+			stat, ok := readEDACDIMMStats(dimmDir)
+			if !ok {
+				continue
+			}
+			key := canonicalLabel(stat.Label)
+			if key == "" {
+				continue
+			}
+			out[key] = stat
+		}
+	}
+	return out
+}
+
+func readEDACDIMMStats(dimmDir string) (edacDIMMStats, bool) {
+	labelBytes, err := os.ReadFile(filepath.Join(dimmDir, "dimm_label"))
+	if err != nil {
+		labelBytes, err = os.ReadFile(filepath.Join(dimmDir, "label"))
+		if err != nil {
+			return edacDIMMStats{}, false
+		}
+	}
+	label := strings.TrimSpace(string(labelBytes))
+	if label == "" {
+		return edacDIMMStats{}, false
+	}
+
+	stat := edacDIMMStats{Label: label}
+	if value, ok := readEDACCount(dimmDir, []string{"dimm_ce_count", "ce_count"}); ok {
+		stat.CECount = &value
+	}
+	if value, ok := readEDACCount(dimmDir, []string{"dimm_ue_count", "ue_count"}); ok {
+		stat.UECount = &value
+	}
+	return stat, true
+}
+
+func readEDACCount(dir string, names []string) (int64, bool) {
+	for _, name := range names {
+		raw, err := os.ReadFile(filepath.Join(dir, name))
+		if err != nil {
+			continue
+		}
+		value, err := strconv.ParseInt(strings.TrimSpace(string(raw)), 10, 64)
+		if err == nil && value >= 0 {
+			return value, true
+		}
+	}
+	return 0, false
+}
+
+func dimmMatchKeys(slot, location *string) []string {
+	var out []string
+	add := func(value *string) {
+		key := canonicalLabel(derefString(value))
+		if key == "" {
+			return
+		}
+		for _, existing := range out {
+			if existing == key {
+				return
+			}
+		}
+		out = append(out, key)
+	}
+	add(slot)
+	add(location)
+	return out
+}
+
+func canonicalLabel(value string) string {
+	value = strings.ToUpper(strings.TrimSpace(value))
+	if value == "" {
+		return ""
+	}
+	var b strings.Builder
+	for _, r := range value {
+		if (r >= 'A' && r <= 'Z') || (r >= '0' && r <= '9') {
+			b.WriteRune(r)
+		}
+	}
+	return b.String()
+}
+
+func isLikelyMemoryTemp(chip, feature string) bool {
+	value := strings.ToLower(chip + " " + feature)
+	return strings.Contains(value, "dimm") || strings.Contains(value, "sodimm")
+}
+
+func boolPtr(value bool) *bool {
+	return &value
+}