Align hardware export with ingest contract
This commit is contained in:
Mikhail Chusavitin
+137
-102
@@ -26,13 +26,13 @@ func collectStorage() []schema.HardwareStorage {
|
||||
|
||||
// lsblkDevice is a minimal lsblk JSON record.
|
||||
type lsblkDevice struct {
|
||||
Name string `json:"name"`
|
||||
Type string `json:"type"`
|
||||
Size string `json:"size"`
|
||||
Serial string `json:"serial"`
|
||||
Model string `json:"model"`
|
||||
Tran string `json:"tran"`
|
||||
Hctl string `json:"hctl"`
|
||||
Name string `json:"name"`
|
||||
Type string `json:"type"`
|
||||
Size string `json:"size"`
|
||||
Serial string `json:"serial"`
|
||||
Model string `json:"model"`
|
||||
Tran string `json:"tran"`
|
||||
Hctl string `json:"hctl"`
|
||||
}
|
||||
|
||||
type lsblkRoot struct {
|
||||
@@ -67,7 +67,10 @@ type smartctlInfo struct {
|
||||
SerialNumber string `json:"serial_number"`
|
||||
FirmwareVer string `json:"firmware_version"`
|
||||
RotationRate int `json:"rotation_rate"`
|
||||
SmartStatus struct {
|
||||
Temperature struct {
|
||||
Current int `json:"current"`
|
||||
} `json:"temperature"`
|
||||
SmartStatus struct {
|
||||
Passed bool `json:"passed"`
|
||||
} `json:"smart_status"`
|
||||
UserCapacity struct {
|
||||
@@ -75,9 +78,11 @@ type smartctlInfo struct {
|
||||
} `json:"user_capacity"`
|
||||
AtaSmartAttributes struct {
|
||||
Table []struct {
|
||||
ID int `json:"id"`
|
||||
Name string `json:"name"`
|
||||
Raw struct{ Value int64 `json:"value"` } `json:"raw"`
|
||||
ID int `json:"id"`
|
||||
Name string `json:"name"`
|
||||
Raw struct {
|
||||
Value int64 `json:"value"`
|
||||
} `json:"raw"`
|
||||
} `json:"table"`
|
||||
} `json:"ata_smart_attributes"`
|
||||
PowerOnTime struct {
|
||||
@@ -130,7 +135,7 @@ func enrichWithSmartctl(dev lsblkDevice) schema.HardwareStorage {
|
||||
return s
|
||||
}
|
||||
|
||||
var info smartctlInfo
|
||||
var info smartctlInfo
|
||||
if err := json.Unmarshal(out, &info); err == nil {
|
||||
if v := cleanDMIValue(info.ModelName); v != "" {
|
||||
s.Model = &v
|
||||
@@ -152,14 +157,19 @@ func enrichWithSmartctl(dev lsblkDevice) schema.HardwareStorage {
|
||||
} else if info.RotationRate > 0 {
|
||||
devType = "HDD"
|
||||
}
|
||||
s.Type = &devType
|
||||
|
||||
// telemetry
|
||||
tel := map[string]any{}
|
||||
if info.Temperature.Current > 0 {
|
||||
t := float64(info.Temperature.Current)
|
||||
s.TemperatureC = &t
|
||||
}
|
||||
if info.PowerOnTime.Hours > 0 {
|
||||
tel["power_on_hours"] = info.PowerOnTime.Hours
|
||||
v := int64(info.PowerOnTime.Hours)
|
||||
s.PowerOnHours = &v
|
||||
}
|
||||
if info.PowerCycleCount > 0 {
|
||||
tel["power_cycles"] = info.PowerCycleCount
|
||||
v := int64(info.PowerCycleCount)
|
||||
s.PowerCycles = &v
|
||||
}
|
||||
reallocated := int64(0)
|
||||
pending := int64(0)
|
||||
@@ -169,77 +179,79 @@ func enrichWithSmartctl(dev lsblkDevice) schema.HardwareStorage {
|
||||
switch attr.ID {
|
||||
case 5:
|
||||
reallocated = attr.Raw.Value
|
||||
tel["reallocated_sectors"] = attr.Raw.Value
|
||||
s.ReallocatedSectors = &reallocated
|
||||
case 177:
|
||||
tel["wear_leveling_pct"] = attr.Raw.Value
|
||||
value := float64(attr.Raw.Value)
|
||||
s.LifeUsedPct = &value
|
||||
case 231:
|
||||
lifeRemaining = attr.Raw.Value
|
||||
tel["life_remaining_pct"] = attr.Raw.Value
|
||||
value := float64(attr.Raw.Value)
|
||||
s.LifeRemainingPct = &value
|
||||
case 241:
|
||||
tel["total_lba_written"] = attr.Raw.Value
|
||||
value := attr.Raw.Value
|
||||
s.WrittenBytes = &value
|
||||
case 197:
|
||||
pending = attr.Raw.Value
|
||||
tel["current_pending_sectors"] = attr.Raw.Value
|
||||
s.CurrentPendingSectors = &pending
|
||||
case 198:
|
||||
uncorrectable = attr.Raw.Value
|
||||
tel["offline_uncorrectable"] = attr.Raw.Value
|
||||
s.OfflineUncorrectable = &uncorrectable
|
||||
}
|
||||
}
|
||||
if len(tel) > 0 {
|
||||
s.Telemetry = tel
|
||||
}
|
||||
|
||||
status := storageHealthStatus{
|
||||
overallPassed: info.SmartStatus.Passed,
|
||||
hasOverall: true,
|
||||
reallocatedSectors: reallocated,
|
||||
pendingSectors: pending,
|
||||
overallPassed: info.SmartStatus.Passed,
|
||||
hasOverall: true,
|
||||
reallocatedSectors: reallocated,
|
||||
pendingSectors: pending,
|
||||
offlineUncorrectable: uncorrectable,
|
||||
lifeRemainingPct: lifeRemaining,
|
||||
lifeRemainingPct: lifeRemaining,
|
||||
}
|
||||
setStorageHealthStatus(&s, status)
|
||||
return s
|
||||
}
|
||||
|
||||
s.Type = &devType
|
||||
status := "UNKNOWN"
|
||||
status := statusUnknown
|
||||
s.Status = &status
|
||||
return s
|
||||
}
|
||||
|
||||
// nvmeSmartLog is the subset of `nvme smart-log -o json` output we care about.
|
||||
type nvmeSmartLog struct {
|
||||
CriticalWarning int `json:"critical_warning"`
|
||||
PercentageUsed int `json:"percentage_used"`
|
||||
AvailableSpare int `json:"available_spare"`
|
||||
SpareThreshold int `json:"spare_thresh"`
|
||||
PowerOnHours int64 `json:"power_on_hours"`
|
||||
PowerCycles int64 `json:"power_cycles"`
|
||||
UnsafeShutdowns int64 `json:"unsafe_shutdowns"`
|
||||
DataUnitsWritten int64 `json:"data_units_written"`
|
||||
ControllerBusy int64 `json:"controller_busy_time"`
|
||||
MediaErrors int64 `json:"media_errors"`
|
||||
NumErrLogEntries int64 `json:"num_err_log_entries"`
|
||||
CriticalWarning int `json:"critical_warning"`
|
||||
PercentageUsed int `json:"percentage_used"`
|
||||
AvailableSpare int `json:"available_spare"`
|
||||
SpareThreshold int `json:"spare_thresh"`
|
||||
Temperature int64 `json:"temperature"`
|
||||
PowerOnHours int64 `json:"power_on_hours"`
|
||||
PowerCycles int64 `json:"power_cycles"`
|
||||
UnsafeShutdowns int64 `json:"unsafe_shutdowns"`
|
||||
DataUnitsRead int64 `json:"data_units_read"`
|
||||
DataUnitsWritten int64 `json:"data_units_written"`
|
||||
ControllerBusy int64 `json:"controller_busy_time"`
|
||||
MediaErrors int64 `json:"media_errors"`
|
||||
NumErrLogEntries int64 `json:"num_err_log_entries"`
|
||||
}
|
||||
|
||||
// nvmeIDCtrl is the subset of `nvme id-ctrl -o json` output.
|
||||
type nvmeIDCtrl struct {
|
||||
ModelNumber string `json:"mn"`
|
||||
SerialNumber string `json:"sn"`
|
||||
FirmwareRev string `json:"fr"`
|
||||
TotalCapacity int64 `json:"tnvmcap"`
|
||||
ModelNumber string `json:"mn"`
|
||||
SerialNumber string `json:"sn"`
|
||||
FirmwareRev string `json:"fr"`
|
||||
TotalCapacity int64 `json:"tnvmcap"`
|
||||
}
|
||||
|
||||
func enrichWithNVMe(dev lsblkDevice) schema.HardwareStorage {
|
||||
present := true
|
||||
devType := "NVMe"
|
||||
iface := "NVMe"
|
||||
status := "OK"
|
||||
status := statusOK
|
||||
s := schema.HardwareStorage{
|
||||
Present: &present,
|
||||
Type: &devType,
|
||||
Interface: &iface,
|
||||
Status: &status,
|
||||
HardwareComponentStatus: schema.HardwareComponentStatus{Status: &status},
|
||||
Present: &present,
|
||||
Type: &devType,
|
||||
Interface: &iface,
|
||||
}
|
||||
|
||||
devPath := "/dev/" + dev.Name
|
||||
@@ -268,100 +280,123 @@ func enrichWithNVMe(dev lsblkDevice) schema.HardwareStorage {
|
||||
if out, err := exec.Command("nvme", "smart-log", devPath, "-o", "json").Output(); err == nil {
|
||||
var log nvmeSmartLog
|
||||
if json.Unmarshal(out, &log) == nil {
|
||||
tel := map[string]any{}
|
||||
if log.CriticalWarning > 0 {
|
||||
tel["critical_warning"] = log.CriticalWarning
|
||||
}
|
||||
if log.PowerOnHours > 0 {
|
||||
tel["power_on_hours"] = log.PowerOnHours
|
||||
s.PowerOnHours = &log.PowerOnHours
|
||||
}
|
||||
if log.PowerCycles > 0 {
|
||||
tel["power_cycles"] = log.PowerCycles
|
||||
s.PowerCycles = &log.PowerCycles
|
||||
}
|
||||
if log.UnsafeShutdowns > 0 {
|
||||
tel["unsafe_shutdowns"] = log.UnsafeShutdowns
|
||||
s.UnsafeShutdowns = &log.UnsafeShutdowns
|
||||
}
|
||||
if log.PercentageUsed > 0 {
|
||||
tel["percentage_used"] = log.PercentageUsed
|
||||
v := float64(log.PercentageUsed)
|
||||
s.LifeUsedPct = &v
|
||||
remaining := 100 - v
|
||||
s.LifeRemainingPct = &remaining
|
||||
}
|
||||
if log.DataUnitsWritten > 0 {
|
||||
tel["data_units_written"] = log.DataUnitsWritten
|
||||
v := nvmeDataUnitsToBytes(log.DataUnitsWritten)
|
||||
s.WrittenBytes = &v
|
||||
}
|
||||
if log.ControllerBusy > 0 {
|
||||
tel["controller_busy_time"] = log.ControllerBusy
|
||||
if log.DataUnitsRead > 0 {
|
||||
v := nvmeDataUnitsToBytes(log.DataUnitsRead)
|
||||
s.ReadBytes = &v
|
||||
}
|
||||
if log.AvailableSpare > 0 {
|
||||
tel["available_spare_pct"] = log.AvailableSpare
|
||||
}
|
||||
if log.SpareThreshold > 0 {
|
||||
tel["available_spare_threshold_pct"] = log.SpareThreshold
|
||||
v := float64(log.AvailableSpare)
|
||||
s.AvailableSparePct = &v
|
||||
}
|
||||
if log.MediaErrors > 0 {
|
||||
tel["media_errors"] = log.MediaErrors
|
||||
s.MediaErrors = &log.MediaErrors
|
||||
}
|
||||
if log.NumErrLogEntries > 0 {
|
||||
tel["error_log_entries"] = log.NumErrLogEntries
|
||||
s.ErrorLogEntries = &log.NumErrLogEntries
|
||||
}
|
||||
if len(tel) > 0 {
|
||||
s.Telemetry = tel
|
||||
if log.Temperature > 0 {
|
||||
v := float64(log.Temperature - 273)
|
||||
s.TemperatureC = &v
|
||||
}
|
||||
setStorageHealthStatus(&s, storageHealthStatus{
|
||||
criticalWarning: log.CriticalWarning,
|
||||
percentageUsed: int64(log.PercentageUsed),
|
||||
availableSpare: int64(log.AvailableSpare),
|
||||
spareThreshold: int64(log.SpareThreshold),
|
||||
unsafeShutdowns: log.UnsafeShutdowns,
|
||||
mediaErrors: log.MediaErrors,
|
||||
errorLogEntries: log.NumErrLogEntries,
|
||||
criticalWarning: log.CriticalWarning,
|
||||
percentageUsed: int64(log.PercentageUsed),
|
||||
availableSpare: int64(log.AvailableSpare),
|
||||
spareThreshold: int64(log.SpareThreshold),
|
||||
unsafeShutdowns: log.UnsafeShutdowns,
|
||||
mediaErrors: log.MediaErrors,
|
||||
errorLogEntries: log.NumErrLogEntries,
|
||||
})
|
||||
return s
|
||||
}
|
||||
}
|
||||
|
||||
status = "UNKNOWN"
|
||||
status = statusUnknown
|
||||
s.Status = &status
|
||||
return s
|
||||
}
|
||||
|
||||
func nvmeDataUnitsToBytes(units int64) int64 {
|
||||
if units <= 0 {
|
||||
return 0
|
||||
}
|
||||
return units * 512000
|
||||
}
|
||||
|
||||
type storageHealthStatus struct {
|
||||
hasOverall bool
|
||||
overallPassed bool
|
||||
reallocatedSectors int64
|
||||
pendingSectors int64
|
||||
offlineUncorrectable int64
|
||||
lifeRemainingPct int64
|
||||
criticalWarning int
|
||||
percentageUsed int64
|
||||
availableSpare int64
|
||||
spareThreshold int64
|
||||
unsafeShutdowns int64
|
||||
mediaErrors int64
|
||||
errorLogEntries int64
|
||||
hasOverall bool
|
||||
overallPassed bool
|
||||
reallocatedSectors int64
|
||||
pendingSectors int64
|
||||
offlineUncorrectable int64
|
||||
lifeRemainingPct int64
|
||||
criticalWarning int
|
||||
percentageUsed int64
|
||||
availableSpare int64
|
||||
spareThreshold int64
|
||||
unsafeShutdowns int64
|
||||
mediaErrors int64
|
||||
errorLogEntries int64
|
||||
}
|
||||
|
||||
func setStorageHealthStatus(s *schema.HardwareStorage, health storageHealthStatus) {
|
||||
status := "OK"
|
||||
status := statusOK
|
||||
var description *string
|
||||
switch {
|
||||
case health.hasOverall && !health.overallPassed:
|
||||
status = "FAILED"
|
||||
status = statusCritical
|
||||
description = stringPtr("SMART overall self-assessment failed")
|
||||
case health.criticalWarning > 0:
|
||||
status = "FAILED"
|
||||
status = statusCritical
|
||||
description = stringPtr("NVMe critical warning is set")
|
||||
case health.pendingSectors > 0 || health.offlineUncorrectable > 0:
|
||||
status = "FAILED"
|
||||
status = statusCritical
|
||||
description = stringPtr("Pending or offline uncorrectable sectors detected")
|
||||
case health.mediaErrors > 0:
|
||||
status = "WARNING"
|
||||
status = statusWarning
|
||||
description = stringPtr("Media errors reported")
|
||||
case health.reallocatedSectors > 0:
|
||||
status = "WARNING"
|
||||
status = statusWarning
|
||||
description = stringPtr("Reallocated sectors detected")
|
||||
case health.errorLogEntries > 0:
|
||||
status = "WARNING"
|
||||
status = statusWarning
|
||||
description = stringPtr("Device error log contains entries")
|
||||
case health.lifeRemainingPct > 0 && health.lifeRemainingPct <= 10:
|
||||
status = "WARNING"
|
||||
status = statusWarning
|
||||
description = stringPtr("Life remaining is low")
|
||||
case health.percentageUsed >= 95:
|
||||
status = "WARNING"
|
||||
status = statusWarning
|
||||
description = stringPtr("Drive wear level is high")
|
||||
case health.availableSpare > 0 && health.spareThreshold > 0 && health.availableSpare <= health.spareThreshold:
|
||||
status = "WARNING"
|
||||
status = statusWarning
|
||||
description = stringPtr("Available spare is at or below threshold")
|
||||
case health.unsafeShutdowns > 100:
|
||||
status = "WARNING"
|
||||
status = statusWarning
|
||||
description = stringPtr("Unsafe shutdown count is high")
|
||||
}
|
||||
s.Status = &status
|
||||
s.ErrorDescription = description
|
||||
}
|
||||
|
||||
func stringPtr(value string) *string {
|
||||
return &value
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user