Files
bee/audit/internal/collector/raid_controller_telemetry.go
2026-03-15 21:04:53 +03:00

335 lines
10 KiB
Go

package collector
import (
"bee/audit/internal/schema"
"encoding/json"
"log/slog"
"strconv"
"strings"
)
type raidControllerTelemetry struct {
BatteryChargePct *float64
BatteryHealthPct *float64
BatteryTemperatureC *float64
BatteryVoltageV *float64
BatteryReplaceRequired *bool
ErrorDescription *string
}
func enrichPCIeWithRAIDTelemetry(devs []schema.HardwarePCIeDevice) []schema.HardwarePCIeDevice {
byVendor := collectRAIDControllerTelemetry()
if len(byVendor) == 0 {
return devs
}
positions := map[int]int{}
for i := range devs {
if devs[i].VendorID == nil || !isLikelyRAIDController(devs[i]) {
continue
}
vendor := *devs[i].VendorID
list := byVendor[vendor]
if len(list) == 0 {
continue
}
index := positions[vendor]
if index >= len(list) {
continue
}
positions[vendor] = index + 1
applyRAIDControllerTelemetry(&devs[i], list[index])
}
return devs
}
func applyRAIDControllerTelemetry(dev *schema.HardwarePCIeDevice, tel raidControllerTelemetry) {
if tel.BatteryChargePct != nil {
dev.BatteryChargePct = tel.BatteryChargePct
}
if tel.BatteryHealthPct != nil {
dev.BatteryHealthPct = tel.BatteryHealthPct
}
if tel.BatteryTemperatureC != nil {
dev.BatteryTemperatureC = tel.BatteryTemperatureC
}
if tel.BatteryVoltageV != nil {
dev.BatteryVoltageV = tel.BatteryVoltageV
}
if tel.BatteryReplaceRequired != nil {
dev.BatteryReplaceRequired = tel.BatteryReplaceRequired
}
if tel.ErrorDescription != nil {
dev.ErrorDescription = tel.ErrorDescription
if dev.Status == nil || *dev.Status == statusOK {
status := statusWarning
dev.Status = &status
}
}
}
func collectRAIDControllerTelemetry() map[int][]raidControllerTelemetry {
out := map[int][]raidControllerTelemetry{}
if raw, err := raidToolQuery("storcli64", "/call", "show", "all", "J"); err == nil {
list := parseStorcliControllerTelemetry(raw)
if len(list) > 0 {
out[vendorBroadcomLSI] = append(out[vendorBroadcomLSI], list...)
slog.Info("raid: storcli controller telemetry", "count", len(list))
}
}
if raw, err := raidToolQuery("ssacli", "ctrl", "all", "show", "config", "detail"); err == nil {
list := parseSSACLIControllerTelemetry(string(raw))
if len(list) > 0 {
out[vendorHPE] = append(out[vendorHPE], list...)
slog.Info("raid: ssacli controller telemetry", "count", len(list))
}
}
if raw, err := raidToolQuery("arcconf", "getconfig", "1", "ad"); err == nil {
list := parseArcconfControllerTelemetry(string(raw))
if len(list) > 0 {
out[vendorAdaptec] = append(out[vendorAdaptec], list...)
slog.Info("raid: arcconf controller telemetry", "count", len(list))
}
}
return out
}
func parseStorcliControllerTelemetry(raw []byte) []raidControllerTelemetry {
var doc struct {
Controllers []struct {
ResponseData map[string]any `json:"Response Data"`
} `json:"Controllers"`
}
if err := json.Unmarshal(raw, &doc); err != nil {
slog.Warn("raid: parse storcli controller telemetry failed", "err", err)
return nil
}
var out []raidControllerTelemetry
for _, ctl := range doc.Controllers {
tel := raidControllerTelemetry{}
mergeStorcliBatteryMap(&tel, nestedStringMap(ctl.ResponseData["BBU_Info"]))
mergeStorcliBatteryMap(&tel, nestedStringMap(ctl.ResponseData["BBU_Info_Details"]))
mergeStorcliBatteryMap(&tel, nestedStringMap(ctl.ResponseData["CV_Info"]))
mergeStorcliBatteryMap(&tel, nestedStringMap(ctl.ResponseData["CV_Info_Details"]))
if hasRAIDControllerTelemetry(tel) {
out = append(out, tel)
}
}
return out
}
func nestedStringMap(raw any) map[string]string {
switch value := raw.(type) {
case map[string]any:
out := map[string]string{}
flattenStringMap("", value, out)
return out
case []any:
out := map[string]string{}
for _, item := range value {
if m, ok := item.(map[string]any); ok {
flattenStringMap("", m, out)
}
}
return out
default:
return nil
}
}
func flattenStringMap(prefix string, in map[string]any, out map[string]string) {
for key, raw := range in {
fullKey := strings.TrimSpace(strings.ToLower(strings.Trim(prefix+" "+key, " ")))
switch value := raw.(type) {
case map[string]any:
flattenStringMap(fullKey, value, out)
case []any:
for _, item := range value {
if m, ok := item.(map[string]any); ok {
flattenStringMap(fullKey, m, out)
}
}
case string:
out[fullKey] = value
case json.Number:
out[fullKey] = value.String()
case float64:
out[fullKey] = strconv.FormatFloat(value, 'f', -1, 64)
case bool:
if value {
out[fullKey] = "true"
} else {
out[fullKey] = "false"
}
}
}
}
func mergeStorcliBatteryMap(tel *raidControllerTelemetry, fields map[string]string) {
if len(fields) == 0 {
return
}
for key, raw := range fields {
lower := strings.ToLower(strings.TrimSpace(key))
switch {
case strings.Contains(lower, "relative state of charge"), strings.Contains(lower, "remaining capacity"), strings.Contains(lower, "charge"):
if tel.BatteryChargePct == nil {
tel.BatteryChargePct = parsePercentPtr(raw)
}
case strings.Contains(lower, "state of health"), strings.Contains(lower, "health"):
if tel.BatteryHealthPct == nil {
tel.BatteryHealthPct = parsePercentPtr(raw)
}
case strings.Contains(lower, "temperature"):
if tel.BatteryTemperatureC == nil {
tel.BatteryTemperatureC = parseFloatPtr(raw)
}
case strings.Contains(lower, "voltage"):
if tel.BatteryVoltageV == nil {
tel.BatteryVoltageV = parseFloatPtr(raw)
}
case strings.Contains(lower, "replace"), strings.Contains(lower, "replacement required"):
if tel.BatteryReplaceRequired == nil {
tel.BatteryReplaceRequired = parseReplaceRequired(raw)
}
case strings.Contains(lower, "learn cycle requested"), strings.Contains(lower, "battery state"), strings.Contains(lower, "capacitance state"):
if desc := batteryStateDescription(raw); desc != nil && tel.ErrorDescription == nil {
tel.ErrorDescription = desc
}
}
}
}
func parseSSACLIControllerTelemetry(raw string) []raidControllerTelemetry {
lines := strings.Split(raw, "\n")
var out []raidControllerTelemetry
var current *raidControllerTelemetry
flush := func() {
if current != nil && hasRAIDControllerTelemetry(*current) {
out = append(out, *current)
}
current = nil
}
for _, line := range lines {
trimmed := strings.TrimSpace(line)
if trimmed == "" {
continue
}
if strings.HasPrefix(strings.ToLower(trimmed), "smart array") || strings.HasPrefix(strings.ToLower(trimmed), "controller ") {
flush()
current = &raidControllerTelemetry{}
continue
}
if current == nil {
continue
}
if idx := strings.Index(trimmed, ":"); idx > 0 {
key := strings.ToLower(strings.TrimSpace(trimmed[:idx]))
val := strings.TrimSpace(trimmed[idx+1:])
switch {
case strings.Contains(key, "capacitor temperature"), strings.Contains(key, "battery temperature"):
current.BatteryTemperatureC = parseFloatPtr(val)
case strings.Contains(key, "capacitor voltage"), strings.Contains(key, "battery voltage"):
current.BatteryVoltageV = parseFloatPtr(val)
case strings.Contains(key, "capacitor charge"), strings.Contains(key, "battery charge"):
current.BatteryChargePct = parsePercentPtr(val)
case strings.Contains(key, "capacitor health"), strings.Contains(key, "battery health"):
current.BatteryHealthPct = parsePercentPtr(val)
case strings.Contains(key, "replace") || strings.Contains(key, "failed"):
if current.BatteryReplaceRequired == nil {
current.BatteryReplaceRequired = parseReplaceRequired(val)
}
if desc := batteryStateDescription(val); desc != nil && current.ErrorDescription == nil {
current.ErrorDescription = desc
}
}
}
}
flush()
return out
}
func parseArcconfControllerTelemetry(raw string) []raidControllerTelemetry {
lines := strings.Split(raw, "\n")
tel := raidControllerTelemetry{}
for _, line := range lines {
trimmed := strings.TrimSpace(line)
if idx := strings.Index(trimmed, ":"); idx > 0 {
key := strings.ToLower(strings.TrimSpace(trimmed[:idx]))
val := strings.TrimSpace(trimmed[idx+1:])
switch {
case strings.Contains(key, "battery temperature"), strings.Contains(key, "capacitor temperature"):
tel.BatteryTemperatureC = parseFloatPtr(val)
case strings.Contains(key, "battery voltage"), strings.Contains(key, "capacitor voltage"):
tel.BatteryVoltageV = parseFloatPtr(val)
case strings.Contains(key, "battery charge"), strings.Contains(key, "capacitor charge"):
tel.BatteryChargePct = parsePercentPtr(val)
case strings.Contains(key, "battery health"), strings.Contains(key, "capacitor health"):
tel.BatteryHealthPct = parsePercentPtr(val)
case strings.Contains(key, "replace"), strings.Contains(key, "failed"):
if tel.BatteryReplaceRequired == nil {
tel.BatteryReplaceRequired = parseReplaceRequired(val)
}
if desc := batteryStateDescription(val); desc != nil && tel.ErrorDescription == nil {
tel.ErrorDescription = desc
}
}
}
}
if hasRAIDControllerTelemetry(tel) {
return []raidControllerTelemetry{tel}
}
return nil
}
func hasRAIDControllerTelemetry(tel raidControllerTelemetry) bool {
return tel.BatteryChargePct != nil ||
tel.BatteryHealthPct != nil ||
tel.BatteryTemperatureC != nil ||
tel.BatteryVoltageV != nil ||
tel.BatteryReplaceRequired != nil ||
tel.ErrorDescription != nil
}
func parsePercentPtr(raw string) *float64 {
raw = strings.ReplaceAll(strings.TrimSpace(raw), "%", "")
return parseFloatPtr(raw)
}
func parseReplaceRequired(raw string) *bool {
lower := strings.ToLower(strings.TrimSpace(raw))
switch {
case lower == "":
return nil
case strings.Contains(lower, "replace"), strings.Contains(lower, "failed"), strings.Contains(lower, "yes"), strings.Contains(lower, "required"):
value := true
return &value
case strings.Contains(lower, "no"), strings.Contains(lower, "ok"), strings.Contains(lower, "good"), strings.Contains(lower, "optimal"):
value := false
return &value
default:
return nil
}
}
func batteryStateDescription(raw string) *string {
lower := strings.ToLower(strings.TrimSpace(raw))
if lower == "" {
return nil
}
switch {
case strings.Contains(lower, "failed"), strings.Contains(lower, "fault"), strings.Contains(lower, "replace"), strings.Contains(lower, "warning"), strings.Contains(lower, "degraded"):
return &raw
default:
return nil
}
}