SMART Warnings on a Home NAS: Which Ones to Watch and Which Mean Replace the Drive Now

Sooner or later, every home NAS shows a disk warning. The dashboard turns amber, an email arrives with a line like “Current_Pending_Sector: 8”, and you have to decide on the spot: is this drive dying, or is this noise? Replace too early and you waste money on healthy drives. Replace too late and a second failure during a rebuild can take the whole array with it.

SMART (Self-Monitoring, Analysis and Reporting Technology) is the data drives report about their own health. Some of it predicts failure well. Much of it is confusing, vendor-specific, or harmless. This guide sorts the warnings you will actually see into three groups: replace now, watch closely, and usually ignore.

What SMART can and cannot tell you

Each hard drive tracks a list of attributes: counts of remapped sectors, read errors, temperatures, hours powered on, and more. Each attribute has a raw value, a normalised value and a threshold set by the manufacturer. A drive “fails SMART” only when a normalised value crosses its threshold, which often happens late, sometimes after the drive has already caused problems.

Large-scale drive statistics published by cloud backup providers have shown a consistent pattern: a handful of attributes are strongly linked to failure, yet a meaningful share of drives fail with no SMART warning at all. SMART is an early-warning system, not a guarantee. That is why redundancy and backups matter more than any single warning, a point explored in Scrutiny vs a green SMART dashboard.

Replace now: the warnings that predict failure

These attributes count real damage to the drive’s surface or its ability to read data. Any non-zero raw value deserves attention; a rising value means the drive is getting worse.

Pending sectors deserve a special note. Sometimes they clear: when the sector is written again, the drive either succeeds or remaps it, and the pending count drops. A one-time pending sector that clears after a scrub, with no growth in reallocated sectors, may not need immediate replacement. Pending sectors that keep appearing mean the surface is degrading.

Watch closely: warnings that need context

These do not always mean the drive is failing, but they deserve investigation:

• 199 UDMA CRC Error Count. Errors in data transfer between the drive and the controller. This almost always points to a bad SATA cable, a loose connection or a faulty backplane slot, not the drive itself. Reseat or replace the cable; if the count stops rising, the drive is fine. The count never resets, so watch for increases, not the absolute number.
• 194 or 190 Temperature. Hard drives are happiest roughly between 25 and 45°C. Sustained temperatures in the 50s shorten life. Improve airflow before blaming the drive.
• 10 Spin Retry Count. The drive needed more than one attempt to spin up. Can indicate a power supply problem or a failing motor. Check power, then watch.
• Self-test failures. A failed short or extended self-test is a strong sign of trouble, even if attributes look normal. Re-run the test; a repeat failure means replace.

Usually ignore: alarming numbers that are normal

Some attributes generate scary-looking raw values that mean nothing on their own:

• 1 Raw Read Error Rate and 7 Seek Error Rate on Seagate drives. Seagate encodes these raw values in a way that produces enormous numbers on perfectly healthy drives. Look at the normalised value, not the raw one.
• 9 Power-On Hours. Age, not health. Older drives fail more often on average, but a drive with many hours and clean failure attributes is not a reason to panic.
• 193 Load Cycle Count. How often the heads park. Some drives park aggressively when idle, producing high counts. Only a concern if it climbs extremely fast; NAS-rated drives usually handle this sensibly.
• 12 Power Cycle Count and 4 Start/Stop Count. Usage history, not damage.

Monitoring tools that flag every attribute crossing an arbitrary threshold produce many false alarms. Tools that understand vendor-specific encodings, or let you choose which attributes to alert on, are much easier to live with.

SSDs and NVMe drives are different

Solid-state drives wear out by writing, not by mechanical damage, so their health data looks different. For NVMe drives, the important fields are:

• Percentage Used. An estimate of how much of the drive’s rated write endurance is consumed. Plan replacement as it approaches 100%, though many drives keep working beyond it.
• Available Spare. Remaining spare blocks; a falling value below its threshold is a warning.
• Media and Data Integrity Errors. Should be zero. Non-zero means data could not be read reliably.
• Critical Warning. A flag field; any set bit deserves attention.

SATA SSDs report similar information through vendor-specific attributes such as wear levelling count or remaining life. SSDs can also fail suddenly, often controller failures rather than worn-out cells, with no warning at all, which again makes backups essential.

Errors from the file system, not SMART

ZFS and Btrfs check data with checksums every time it is read and during scrubs. A scrub that reports checksum errors on a disk tells you something SMART may not: data came back wrong. A single event after a power cut or a cable problem is worth investigating; repeated checksum errors on one disk point to a failing drive or a bad connection.

Combine both sources. A disk with growing pending sectors and checksum errors in scrubs is failing. A disk with checksum errors but clean SMART data, especially alongside CRC errors, points to cables, the controller or memory.

Setting up monitoring that tells you in time

SMART data is only useful if someone reads it. A sensible setup for a home NAS:

1. Schedule self-tests. A short test weekly and an extended test monthly, staggered across disks. Most NAS systems have a scheduler; on Linux, smartd can run them.
2. Alert on the replace-now attributes increasing, and on failed self-tests, by email or push notification.
3. Track history. A dashboard such as Scrutiny keeps attribute history, so you can see a count rising over weeks rather than a single snapshot.
4. Run regular scrubs on ZFS or Btrfs, monthly for most home pools, and alert on errors.
5. Test the alert path once, so you know the warning will actually reach your phone.

Replacing a drive without drama

When a drive needs replacing, do it calmly:

• Check your backup first. A rebuild stresses every remaining drive. Make sure a recent backup exists before you start.
• Replace with the drive still in place if possible. ZFS and some NAS systems can add the new disk and rebuild while the old one still contributes, which is safer than pulling it first.
• Use a CMR drive of equal or larger size. Avoid SMR drives in RAID arrays; rebuilds on them can be extremely slow. Drive choices are compared in WD Red Plus vs Seagate IronWolf.
• Watch the other drives during the rebuild. If drives were bought together, they may age together.
• Check warranty. Reallocated or pending sectors usually qualify for warranty replacement; keep a spare so you are not waiting on shipping.

Reading SMART yourself with smartctl

NAS dashboards summarise SMART data, sometimes too much. When a warning appears, it helps to look at the raw report. On Linux-based systems, including TrueNAS, Unraid, OpenMediaVault and most NAS units with SSH access, the smartmontools package provides the smartctl command:

• smartctl -a /dev/sdX shows all attributes, the self-test log and the error log for a SATA drive.
• smartctl -a /dev/nvme0 shows the health log for an NVMe drive.
• smartctl -t short /dev/sdX starts a short self-test; use -t long for an extended one, which can take many hours on large drives.
• smartctl -l selftest /dev/sdX shows the results of past tests.

Look at three things in the output. First, the attributes in the replace-now group and their raw values. Second, the self-test log: any “Completed: read failure” entry is serious. Third, the drive’s error log, which lists recent errors with the operations that caused them.

Save a copy of the output when you first install each drive. Comparing today’s report with the day-one report is the simplest way to see what has changed, and it is useful evidence for a warranty claim.

Three questions for every warning

Replace drives with rising reallocated, pending or uncorrectable sector counts, or failed self-tests. Investigate CRC errors and high temperatures, which usually point to cables and airflow. Ignore scary raw numbers on vendor-specific attributes and power-on hours on their own. And remember that SMART misses some failures entirely, so redundancy and a tested backup are the real protection.