drive-health-metrics/score.go

package main

import (
	"fmt"
	"strings"
)

// scoreDrive computes a drive's risk score, recommendation, and the reason
// string behind them. The scoring rules are deliberate:
//
//   - Only real, drive-attributable defects add meaningful score.
//   - Missing or unreadable data is never treated as a failure (no points).
//   - Wear and age are graded to nudge toward planned replacement.
//
// The score maps to a recommendation:
//
//	>= 100 -> REPLACE_NOW    (hard defect: drive is failing or failed)
//	>=  50 -> REPLACE_SOON   (serious wear or accumulating defects)
//	>=  20 -> MONITOR        (early warning signs)
//	<   20 -> OK
//
// A drive with no SMART data and no controller red flags scores NO_DATA, meaning
// re-collect rather than replace.
func scoreDrive(d *Drive) (int, string, string) {
	score := 0
	var reasons []string
	add := func(pts int, msg string) {
		score += pts
		reasons = append(reasons, msg)
	}
	min := func(a, b int) int {
		if a < b {
			return a
		}
		return b
	}

	// ---- Hard physical defects (drive-attributable) ----
	if realloc := iv(d.Reallocated); realloc > 0 {
		pts := min(40+realloc*5, 100)
		add(pts, fmt.Sprintf("reallocated=%d(+%d)", realloc, pts))
	}
	if pending := iv(d.Pending); pending > 0 {
		pts := min(50+pending*5, 100)
		add(pts, fmt.Sprintf("pending=%d(+%d)", pending, pts))
	}
	if uncorr := iv(d.Uncorrectable); uncorr > 0 {
		pts := min(60+uncorr*5, 100)
		add(pts, fmt.Sprintf("uncorrectable=%d(+%d)", uncorr, pts))
	}
	if reported := iv(d.ReportedUncorrect); reported > 0 {
		pts := min(reported*10, 60)
		add(pts, fmt.Sprintf("reported_uncorrect=%d(+%d)", reported, pts))
	}
	if e2e := iv(d.EndToEnd); e2e > 0 {
		pts := min(e2e*20, 80)
		add(pts, fmt.Sprintf("end_to_end_err=%d(+%d)", e2e, pts))
	}
	if badblk := iv(d.RuntimeBadBlocks); badblk > 0 {
		pts := min(badblk*5, 40)
		add(pts, fmt.Sprintf("runtime_bad_blocks=%d(+%d)", badblk, pts))
	}

	// ---- SMART self-assessment: only penalize an EXPLICIT failure ----
	if d.SmartHealth == "FAILED" {
		add(100, "SMART_health=FAILED(+100)")
	}

	// ---- SATA link quality (cabling/backplane, not the NAND) ----
	if crc := iv(d.UdmaCrc); crc > 0 {
		pts := min(crc*3, 25)
		add(pts, fmt.Sprintf("udma_crc=%d(+%d)", crc, pts))
	}

	// ---- RAID-controller signals ----
	if d.PredictiveFailureCtrl > 0 {
		add(70, "ctrl_predictive_failure(+70)")
	}
	if d.SmartAlertCtrl {
		add(50, "ctrl_smart_alert(+50)")
	}
	// Penalize a controller-reported state only when it is not a healthy one.
	// MegaCLI spells these out ("Online, Spun Up", "Hotspare") while storcli
	// abbreviates ("Onln", "GHS"/"DHS" for hot spares, "JBOD"); all are fine.
	fw := strings.ToLower(d.FwState)
	fwHealthy := fw == "" ||
		strings.Contains(fw, "online") || strings.Contains(fw, "onln") ||
		strings.Contains(fw, "hotspare") || strings.Contains(fw, "ghs") ||
		strings.Contains(fw, "dhs") || strings.Contains(fw, "jbod")
	if !fwHealthy {
		add(40, fmt.Sprintf("fw_state=%s(+40)", d.FwState))
	}

	// MegaCLI/storcli media errors: soft signal, graded gently and capped.
	me := d.MediaErrCtrl
	switch {
	case me >= 100:
		add(30, fmt.Sprintf("ctrl_media_errors=%d(+30)", me))
	case me >= 20:
		add(15, fmt.Sprintf("ctrl_media_errors=%d(+15)", me))
	case me > 0:
		add(5, fmt.Sprintf("ctrl_media_errors=%d(+5)", me))
	}

	// ---- NVMe critical warning bitmask (any bit set is a real alert) ----
	if d.NvmeCriticalWarning != nil && *d.NvmeCriticalWarning > 0 {
		add(60, fmt.Sprintf("nvme_critical_warning=0x%02x(+60)", *d.NvmeCriticalWarning))
	}
	// NVMe spare below threshold -> reserve exhaustion.
	if d.NvmeAvailSpare != nil && d.NvmeAvailSpareThresh != nil &&
		*d.NvmeAvailSpare <= *d.NvmeAvailSpareThresh {
		add(40, fmt.Sprintf("nvme_avail_spare<=thresh(%d<=%d)(+40)",
			*d.NvmeAvailSpare, *d.NvmeAvailSpareThresh))
	}

	// ---- Wear (graded; only meaningful with real SMART data) ----
	if d.WearPctConsumed != nil {
		wc := *d.WearPctConsumed
		switch {
		case wc >= 95:
			add(80, fmt.Sprintf("wear_consumed=%d%%(+80)", wc))
		case wc >= 90:
			add(55, fmt.Sprintf("wear_consumed=%d%%(+55)", wc))
		case wc >= 80:
			add(30, fmt.Sprintf("wear_consumed=%d%%(+30)", wc))
		case wc >= 70:
			add(15, fmt.Sprintf("wear_consumed=%d%%(+15)", wc))
		case wc >= 60:
			add(8, fmt.Sprintf("wear_consumed=%d%%(+8)", wc))
		}
	}

	// ---- Reserve-block exhaustion (Micron ID180 VALUE -> remaining %) ----
	if d.UnusedReservePct != nil && *d.UnusedReservePct <= 10 {
		add(30, fmt.Sprintf("reserve_blocks_low(val=%d)(+30)", *d.UnusedReservePct))
	}

	// ---- Age (gentle nudge only) ----
	hours := iv(d.PowerOnHours)
	switch {
	case hours >= 61320: // Older than 7 years.
		add(15, fmt.Sprintf("age=%dh(+15)", hours))
	case hours >= 52560: // Older than 6 years.
		add(8, fmt.Sprintf("age=%dh(+8)", hours))
	case hours >= 43800: // Older than 5 years.
		add(4, fmt.Sprintf("age=%dh(+4)", hours))
	}

	// ---- Decide recommendation ----
	// NO_DATA only when nothing observed the drive: no SMART, no controller error
	// signals, and no controller state. A controller-only drive (e.g. NVMe behind
	// a PERC) reports a FwState, so it is scored on controller evidence instead.
	if !d.HaveSmart && me == 0 && d.PredictiveFailureCtrl == 0 && !d.SmartAlertCtrl && d.FwState == "" {
		return 0, "NO_DATA", "smartctl returned no usable SMART data; re-collect"
	}

	var rec string
	switch {
	case score >= 100:
		rec = "REPLACE_NOW"
	case score >= 50:
		rec = "REPLACE_SOON"
	case score >= 20:
		rec = "MONITOR"
	default:
		rec = "OK"
	}

	if len(reasons) == 0 {
		return score, rec, "no defects detected"
	}
	return score, rec, strings.Join(reasons, "; ")
}