SMART: the attributes that actually predict a disk failure

The overall SMART verdict stays "OK" to the very end. The five counters that genuinely warn you, how to read them, and why a disk can die without crossing a single threshold.

A disk rarely starts dying abruptly. It degrades, and it signals that — just not where people look. The overall SMART verdict, the one everybody checks, stays green until a very advanced stage.

Why the overall verdict is nearly useless

That verdict says no attribute has crossed a manufacturer threshold. Those thresholds are calibrated for warranty purposes: conservative, designed to fire only on undeniable failure.

The consequence: a disk can accumulate bad sectors, pile up read errors, slow down noticeably, and keep reporting a healthy status. By the time it finally flips, your margin is already thin.

So that verdict is useful in one direction only: if it is bad, the disk is finished. If it is good, it tells you nothing.

The five counters that genuinely warn

Reallocated sectors. The drive found bad sectors and replaced them from the spare pool. Their existence is not alarming in itself — a new disk can have a few. Their progression is.

Pending sectors. This is the most important counter, and the least known. A sector failed to read but could not be replaced, because its contents could not be recovered. In other words: data may already be unreadable. A persistent non-zero value justifies an immediate backup.

Uncorrectable sectors. The drive gave up: the read failed permanently. At that stage the question is no longer whether to replace the disk, but how quickly you can finish copying what remains.

Raw read error rate. Interpret with care: some manufacturers publish high values by design, without indicating any problem. The variation matters, not the value.

Power cycles and temperature history. A disk that spent its life at 55 °C ages faster than one at 35 °C. This is not a sign of imminent failure, it is a risk factor to weigh.

Read a trend, not a value

This is the most common methodological mistake. "How many reallocated sectors are acceptable?" has no useful answer.

A disk with twelve reallocated sectors stable for two years hit a localised manufacturing defect, handled it, and has not moved since. A disk with three that appeared this week is actively degrading.

The second is far more worrying than the first, despite showing a value four times lower. And the progression is rarely linear: it accelerates.

In practice: sample the counters periodically and keep the history. Alert on change, not on a threshold. A counter that moves is a counter that is talking.

The SSD case

On an SSD, mechanical counters are meaningless. Three indicators replace them.

The wear indicator, expressing the fraction of planned write cycles already consumed. Its decline is normal and generally slow — most SSDs are replaced for obsolescence long before reaching their limit.

The total bytes written, compared against the manufacturer's rated endurance. It is the most factual measure of real wear.

The remaining spare blocks, the most revealing of the three. Its decline means the drive is dipping into its reserve to compensate for dead cells. Unlike a mechanical disk that slows and grinds, a failing SSD can go abruptly read-only, or become inaccessible outright. The warning window is short: this counter deserves an alert.

The surface test: what counters cannot see

SMART attributes only report errors that were encountered. A bad sector sitting in an area nothing has read for months stays completely invisible.

That is a serious blind spot, and it always surfaces at the worst moment: during an array rebuild, when the system finally reads every disk in full and discovers the defect that was lying in wait.

A surface test walks the whole disk and forces detection. A short version, a few minutes long, can run frequently; a long version, several hours, deserves a monthly pass. It runs during normal operation with limited impact if scheduled in quiet hours.

What to do when a counter moves

The response depends on the counter:

  • Non-zero pending or uncorrectable sectors: back up immediately, then replace. Do not launch a long test before securing the data.
  • Rising reallocated sectors: plan the replacement without panic, but first confirm your backups are current and restorable.
  • Declining SSD spare blocks: replace before reaching the critical zone, the warning window there is short.
  • Stable values, even non-zero: watch, change nothing. A disk with a past but no present is not a sick disk.

And remember none of this replaces a backup. SMART buys you time to react; it never guarantees it will warn you.

Frequently asked questions

What does the overall SMART verdict actually mean?
It means no attribute has crossed the critical threshold set by the manufacturer. Those thresholds are deliberately conservative: they are calibrated for warranty purposes, not to warn you early. A disk can accumulate bad sectors, pile up read errors and become genuinely unstable while still reporting a healthy verdict. It is a late indicator: by the time it flips, it is usually too late to act calmly.
What is the difference between reallocated and pending sectors?
A reallocated sector was identified as bad and replaced from the spare pool: the incident is handled. A pending sector failed to read but could not yet be replaced, because its contents could not be recovered. That is the more worrying of the two: it signals data that may already be unreadable. A non-zero value that persists justifies an immediate backup.
How many reallocated sectors are acceptable?
The question is wrong: what matters is the trend, not the number. A disk with twelve reallocated sectors, stable for two years, is more reassuring than a disk with three that appeared this week. A rising value indicates active surface degradation, and that progression is rarely linear — it accelerates. Watch the drift over time, not a threshold.
Which attributes matter on an SSD?
Mechanical counters are meaningless on an SSD. Look instead at the wear indicator, which expresses how much of the planned write endurance has been consumed, the total bytes written compared to rated endurance, and the remaining spare blocks. That last one is the most telling: its decline means the drive is dipping into its reserve to compensate for dead cells.
Is a surface test useful if everything is green?
Yes, and it is the only way to detect certain defects. SMART attributes only report errors that were encountered: a bad sector in an area nothing has read for months stays invisible. A surface test walks the entire disk and forces detection. That is exactly what stops you discovering the problem during an array rebuild — the worst possible moment.

Did you enjoy this article?

Comments

Morgann Riu

Cybersecurity and Linux administration expert. I help companies secure and optimize their critical infrastructures.

Back to the blog

Checklist Sécurité Linux

30 points essentiels pour sécuriser un serveur Linux. Recevez aussi les nouveaux tutoriels par email.

Pas de spam. Désabonnement en 1 clic.