Add SMART self-test metrics

.gitignore

@@ -8,3 +8,4 @@ debug/
 Manifest
 smartctl_exporter
 *.exe
+.aider*

go.mod

@@ -5,8 +5,10 @@ go 1.22
 require (
 	github.com/alecthomas/kingpin/v2 v2.4.0
 	github.com/prometheus/client_golang v1.20.5
+	github.com/prometheus/client_model v0.6.1
 	github.com/prometheus/common v0.61.0
 	github.com/prometheus/exporter-toolkit v0.13.2
+	github.com/stretchr/testify v1.10.0
 	github.com/tidwall/gjson v1.18.0
 )
 
@@ -15,13 +17,14 @@ require (
 	github.com/beorn7/perks v1.0.1 // indirect
 	github.com/cespare/xxhash/v2 v2.3.0 // indirect
 	github.com/coreos/go-systemd/v22 v22.5.0 // indirect
+	github.com/davecgh/go-spew v1.1.1 // indirect
 	github.com/jpillora/backoff v1.0.0 // indirect
 	github.com/klauspost/compress v1.17.9 // indirect
 	github.com/mdlayher/socket v0.4.1 // indirect
 	github.com/mdlayher/vsock v1.2.1 // indirect
 	github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822 // indirect
 	github.com/mwitkow/go-conntrack v0.0.0-20190716064945-2f068394615f // indirect
-	github.com/prometheus/client_model v0.6.1 // indirect
+	github.com/pmezard/go-difflib v1.0.0 // indirect
 	github.com/prometheus/procfs v0.15.1 // indirect
 	github.com/tidwall/match v1.1.1 // indirect
 	github.com/tidwall/pretty v1.2.0 // indirect
@@ -34,4 +37,5 @@ require (
 	golang.org/x/text v0.21.0 // indirect
 	google.golang.org/protobuf v1.35.2 // indirect
 	gopkg.in/yaml.v2 v2.4.0 // indirect
+	gopkg.in/yaml.v3 v3.0.1 // indirect
 )

metrics.go

@@ -354,4 +354,27 @@ var (
 		},
 		nil,
 	)
+	metricDeviceLastSelfTest = prometheus.NewDesc(
+		"smartctl_device_last_self_test",
+		"Device last SMART self test status",
+		[]string{
+			"device",
+			"type",
+			"lifetime_hours",
+		},
+		nil,
+	)
+
+	metricDeviceLastSelfTestInfo = prometheus.NewDesc(
+		"smartctl_device_last_self_test_info",
+		"Device last SMART self test info",
+		[]string{
+			"device",
+			"type",
+			"lifetime_hours",
+			"status",
+			"description",
+		},
+		nil,
+	)
 )

readjson.go

@@ -63,7 +63,7 @@ func readFakeSMARTctl(logger *slog.Logger, device Device) gjson.Result {
 // Get json from smartctl and parse it
 func readSMARTctl(logger *slog.Logger, device Device) (gjson.Result, bool) {
 	start := time.Now()
-	var smartctlArgs = []string{"--json", "--info", "--health", "--attributes", "--tolerance=verypermissive", "--nocheck=standby", "--format=brief", "--log=error", "--device=" + device.Type, device.Name}
+	var smartctlArgs = []string{"--json", "--info", "--health", "--attributes", "--tolerance=verypermissive", "--nocheck=standby", "--format=brief", "--log=error", "--log=selftest", "--device=" + device.Type, device.Name}
 
 	logger.Debug("Calling smartctl with args", "args", strings.Join(smartctlArgs, " "))
 	out, err := exec.Command(*smartctlPath, smartctlArgs...).Output()

smartctl.go

@@ -541,6 +541,48 @@ func (smart *SMARTctl) mineDeviceErrorLog() {
 }
 
 func (smart *SMARTctl) mineDeviceSelfTestLog() {
+	// For SCSI devices
+	if smart.device.interface_ == "scsi" {
+		var lastTest gjson.Result
+		var maxHours int64
+
+		// Find the most recent test by power_on_time.hours
+		smart.json.ForEach(func(key, entry gjson.Result) bool {
+			if strings.HasPrefix(key.String(), "scsi_self_test_") {
+				hours := entry.Get("power_on_time.hours").Int()
+				if hours > maxHours {
+					maxHours = hours
+					lastTest = entry
+				}
+			}
+			return true
+		})
+
+		if lastTest.Exists() {
+			smart.ch <- prometheus.MustNewConstMetric(
+				metricDeviceLastSelfTest,
+				prometheus.GaugeValue,
+				lastTest.Get("result.value").Float(),
+				smart.device.device,
+				lastTest.Get("code.string").String(),
+				fmt.Sprintf("%d", maxHours),
+			)
+
+			smart.ch <- prometheus.MustNewConstMetric(
+				metricDeviceLastSelfTestInfo,
+				prometheus.GaugeValue,
+				1.0,
+				smart.device.device,
+				lastTest.Get("code.string").String(),
+				fmt.Sprintf("%d", maxHours),
+				fmt.Sprintf("%d", lastTest.Get("result.value").Int()),
+				lastTest.Get("result.string").String(),
+			)
+		}
+		return
+	}
+
+	// Original ATA/SAS handling
 	for logType, status := range smart.json.Get("ata_smart_self_test_log").Map() {
 		smart.ch <- prometheus.MustNewConstMetric(
 			metricDeviceSelfTestLogCount,
@@ -556,6 +598,31 @@ func (smart *SMARTctl) mineDeviceSelfTestLog() {
 			smart.device.device,
 			logType,
 		)
+
+		// Get the first (most recent) test from the table if it exists
+		table := status.Get("table")
+		if table.Exists() && len(table.Array()) > 0 {
+			lastTest := table.Array()[0]
+			smart.ch <- prometheus.MustNewConstMetric(
+				metricDeviceLastSelfTest,
+				prometheus.GaugeValue,
+				lastTest.Get("status.value").Float(),
+				smart.device.device,
+				lastTest.Get("type.string").String(),
+				fmt.Sprintf("%d", lastTest.Get("lifetime_hours").Int()),
+			)
+
+			smart.ch <- prometheus.MustNewConstMetric(
+				metricDeviceLastSelfTestInfo,
+				prometheus.GaugeValue,
+				1,
+				smart.device.device,
+				lastTest.Get("type.string").String(),
+				fmt.Sprintf("%d", lastTest.Get("lifetime_hours").Int()),
+				fmt.Sprintf("%d", lastTest.Get("status.value").Int()),
+				lastTest.Get("status.string").String(),
+			)
+		}
 	}
 }