drive-health-metrics/collect.go

package main

import (
	"fmt"
	"sort"
	"strings"
	"time"
)

// collect discovers every drive, queries SMART, attaches controller data, and
// scores it.
func collect() ([]*Drive, int64) {
	host := hostname()
	if app != nil && app.config != nil && app.config.Hostname != "" {
		host = app.config.Hostname
	}
	collectedAt := time.Now().UTC().Format("2006-01-02T15:04:05Z")
	tsNs := time.Now().Unix() * 1e9

	st := newSmartTool()
	ctrl := controllerIndex()
	devices := st.scan()

	// Fallback: no scan-open results but we do have controller drives -> probe
	// a base device by megaraid index.
	if len(devices) == 0 && len(ctrl) > 0 {
		base := findBaseDev()
		ids := make([]string, 0, len(ctrl))
		for id := range ctrl {
			ids = append(ids, id)
		}
		sort.Strings(ids)
		for _, id := range ids {
			for _, tmpl := range megaraidDtypes {
				devices = append(devices, scanned{
					path: base, dtype: fmt.Sprintf(tmpl, id), megaraidN: id,
				})
			}
		}
	}

	var drives []*Drive
	matched := map[string]bool{} // Controller IDs covered by a smartctl device.
	for _, sc := range devices {
		if sc.megaraidN != "" {
			matched[sc.megaraidN] = true
		}
		d := &Drive{
			CollectedAt: collectedAt,
			Hostname:    host,
			DeviceID:    sc.megaraidN,
		}

		ok := st.querySmart(sc.path, sc.dtype, d)

		// Skip iSCSI LUNs and RAID virtual disks; they are not physical drives.
		if isPseudoDevice(d) {
			continue
		}

		// Attach controller-side data by megaraid index == DeviceID.
		if sc.megaraidN != "" {
			if cd, found := ctrl[sc.megaraidN]; found {
				applyController(d, cd)
			}
		}

		// Determine whether real SMART attribute data was obtained.
		d.HaveSmart = ok && d.Model != "" && (d.PowerOnHours != nil ||
			d.WearPctRemaining != nil ||
			d.SmartHealth == "PASSED" || d.SmartHealth == "FAILED" ||
			d.SmartHealth == "PASSED_BY_ATTR")

		finalizeDerived(d)
		drives = append(drives, d)
	}

	// Emit controller-only drives: physical drives the controller reports but
	// smartctl cannot reach (e.g. NVMe behind a PERC). Health comes entirely
	// from the controller (Status, predictive-failure, media/other counters).
	ids := make([]string, 0, len(ctrl))
	for id := range ctrl {
		ids = append(ids, id)
	}
	sort.Strings(ids)
	for _, id := range ids {
		if matched[id] {
			continue
		}
		d := &Drive{CollectedAt: collectedAt, Hostname: host, DeviceID: id}
		applyController(d, ctrl[id])
		if isPseudoDevice(d) {
			continue
		}
		d.HaveSmart = false
		finalizeDerived(d)
		drives = append(drives, d)
	}
	return drives, tsNs
}

// applyController fills controller-side fields and uses MegaCLI/storcli inquiry
// as an identity fallback when smartctl passthrough failed.
func applyController(d *Drive, cd ctrlDrive) {
	d.Enclosure = cd.Enclosure
	d.Slot = cd.Slot
	d.MediaErrCtrl = cd.MediaErr
	d.OtherErrCtrl = cd.OtherErr
	d.PredictiveFailureCtrl = cd.Predictive
	d.SmartAlertCtrl = cd.SmartAlert
	d.FwState = cd.FwState

	// Identity fallback for when smartctl could not read the drive. Prefer the
	// structured fields (perccli2); else split the legacy single-line Inquiry.
	if cd.Model != "" || cd.Serial != "" || cd.Firmware != "" {
		if d.Serial == "" {
			d.Serial = cd.Serial
		}
		if d.Model == "" {
			d.Model = cd.Model
		}
		if d.Firmware == "" {
			d.Firmware = cd.Firmware
		}
	} else if cd.Inquiry != "" {
		// Legacy MegaCLI "Inquiry Data" packs "<serial> <model...> <firmware>" on
		// one line, where the model itself can contain spaces and the token count
		// varies. Serial is always first and the firmware revision always last, so
		// anchor on those and treat everything between as the model.
		parts := strings.Fields(cd.Inquiry)
		if d.Serial == "" && len(parts) >= 1 {
			d.Serial = parts[0]
		}
		if d.Firmware == "" && len(parts) >= 2 {
			d.Firmware = parts[len(parts)-1]
		}
		if d.Model == "" && len(parts) >= 3 {
			d.Model = strings.Join(parts[1:len(parts)-1], " ")
		}
	}
	if d.Rotation == "" {
		d.Rotation = cd.Rotation
	}
	if d.TempC == nil {
		d.TempC = cd.TempC
	}
}

// finalizeDerived computes defect_total, power_on_years, and the risk score.
func finalizeDerived(d *Drive) {
	// Aggregate drive-attributable defect counters. nil only when NONE was
	// readable, so NO_DATA rows stay blank instead of showing a misleading 0.
	defectParts := []*int{
		d.Reallocated, d.Pending, d.Uncorrectable,
		d.ReportedUncorrect, d.RuntimeBadBlocks, d.EndToEnd,
	}
	anyKnown := false
	sum := 0
	for _, p := range defectParts {
		if p != nil {
			anyKnown = true
			sum += *p
		}
	}
	if anyKnown {
		d.DefectTotal = pInt(sum)
	}

	if d.PowerOnHours != nil && *d.PowerOnHours > 0 {
		y := float64(*d.PowerOnHours) / 8760.0
		d.PowerOnYears = pF(float64(int(y*100+0.5)) / 100) // Round to two decimals.
	}

	d.RiskScore, d.Recommendation, d.RiskReasons = scoreDrive(d)
}

// findBaseDev returns a real base block device to anchor the megaraid
// passthrough fallback probe, skipping loop and md devices and defaulting to
// /dev/sda when lsblk yields nothing usable.
func findBaseDev() string {
	out := run("lsblk", "-dno", "NAME")
	for _, ln := range strings.Split(out, "\n") {
		name := strings.TrimSpace(ln)
		if name != "" && !strings.Contains(name, "loop") && !strings.HasPrefix(name, "md") {
			return "/dev/" + name
		}
	}
	return "/dev/sda"
}