My Synology DS920+ started beeping at 2am on a Tuesday, and the notification email just said "Volume 1 has degraded." No drive number, no error code, nothing useful. That vague warning is what sends most people straight to Amazon to buy a replacement they may not even need.
Slow down. A drive that trips a warning is not always a dead drive.
I have pulled disks that turned out to be perfectly healthy, where the real problem was a loose SATA connection or a flaky backplane slot. Testing properly takes a few hours of the drive's time and about ten minutes of yours, and it can save you sixty to two hundred dollars per disk.
Read the SMART data before you touch anything
Every modern hard drive tracks its own health using SMART, which stands for Self-Monitoring, Analysis and Reporting Technology. Your NAS reads these values constantly, and they are the fastest way to separate a real problem from a false alarm.
On a Synology, open Storage Manager, click the drive, and open the Health Info tab. On QNAP it lives under Storage & Snapshots. On TrueNAS you run smartctl -a /dev/sda from a shell.
Four attributes matter more than the rest. Watch these closely.
- Reallocated Sectors Count (ID 5): bad sectors the drive has already swapped out for spares. Anything above zero is a yellow flag, and a climbing number is a red one.
- Current Pending Sectors (ID 197): sectors the drive suspects are bad but has not remapped yet. This is the single most predictive value I watch.
- Uncorrectable Sectors (ID 198): sectors that failed a read and could not be recovered.
- UDMA CRC Error Count (ID 199): communication errors between the drive and the controller. High numbers here often mean a bad cable, not a bad drive.
Run an extended SMART test
The quick SMART status you see in the dashboard only reports what the drive already knows. To actually stress the platters, you need the extended (long) self-test, which reads the entire surface of the disk.
On Synology and QNAP, look for the "S.M.A.R.T. Test" button and choose Extended Test, not Quick. On TrueNAS or any Linux box, run smartctl -t long /dev/sda.
Expect this to take a while. A 4TB drive runs roughly 6 to 8 hours, and a 16TB drive can push past 24. The good news is the test runs in the background, so your NAS stays online the whole time.
When it finishes, check the result with smartctl -l selftest /dev/sda. You want to see "Completed without error." A line reading "Completed: read failure" with an LBA number is the drive telling you exactly where it choked.
One detail trips people up here. If the extended test aborts early with "Interrupted (host reset)," that usually means the NAS put the drive to sleep or the array touched it mid-test. Disable disk hibernation for the duration and run it again before you draw any conclusion.
Confirm with badblocks or a manufacturer tool
SMART is honest but conservative. Some drives report "passed" while still throwing read errors during real use, so a second opinion is worth the time.
The badblocks approach
On a Linux-based NAS or a drive pulled into a desktop, badblocks reads every sector and reports failures. The safe, read-only version is:
badblocks -sv /dev/sda
The -s flag shows progress and -v makes it verbose. This can run for many hours on a large disk, but it catches weak sectors that a single SMART pass sometimes misses. If you want a more aggressive write test that also refreshes every sector, use the -w flag, but understand that the destructive write test erases the entire drive, so only do that on a disk you have already removed from the pool.
The vendor tool approach
Every major manufacturer ships a diagnostic that knows its own drives better than any generic tool. If you can connect the suspect drive to a Windows or Linux desktop, use the matching one.
| Brand | Tool | Best for |
|---|---|---|
| Western Digital / WD Red | Data Lifeguard / Dashboard | Full read and write surface scans |
| Seagate / IronWolf | SeaTools | Short and long generic tests |
| Toshiba | Fic (Fujitsu/Toshiba diag) | Basic surface verification |
A drive that fails the vendor's own extended test is done. That result also gives you the clean pass or fail language most warranty claims require, which matters if the disk is still covered.
Worth knowing: some NAS drives, especially SMR models sold a few years back, run painfully slowly under sustained write tests and can look like they are failing when they are just choking on their own recording method. If a WD Red in the EFAX range crawls during a write pass but SMART stays clean, that is the drive design, not a defect.
Rule out the cheap culprits first
Here is the mistake I see constantly. People condemn a drive when the real fault was a two dollar cable or a dusty slot.
Before you buy anything, try these swaps. Each one takes five minutes.
- Move the drive to a different bay. If the errors follow the drive, it is the drive. If they stay with the slot, your NAS backplane or a SATA connector is the problem.
- Reseat the disk. Power down, pull the caddy, wipe the connector edge, and slide it firmly back in. Loose contacts throw the exact CRC errors people mistake for failures.
- Check that UDMA CRC count. If ID 199 is high but reallocated and pending sectors are zero, replace the cable or reseat, not the drive.
Reading the verdict
After all this, the picture is usually clear. Zero reallocated and pending sectors, a clean extended test, and a passing badblocks run means the drive is fine and something else tripped the alarm.
Rising pending sectors, read failures in the self-test log, or a hard fail from the vendor tool means you replace it and do not look back. For the deeper meaning behind each attribute, the community reference at the SMART attributes overview is genuinely worth a read, and Backblaze publishes real-world drive failure statistics that show which SMART values actually predict death.
One drive tested this way saved me a needless purchase last spring. The DS920+ was screaming, but every test came back clean, and a reseat fixed it in five minutes.
What to do the moment a test fails
If the drive really is bad, act fast but calmly. Order the replacement, keep the failing disk in the array if it will still respond, and let the NAS rebuild onto the new disk before you pull the old one. Removing a marginal drive too early, while another is quietly weakening, is how single failures turn into total array loss. Test, confirm, then replace in that order every time.
Frequently asked questions
How long does an extended SMART test take on a NAS drive?
It depends on capacity. A 4TB drive typically finishes in 6 to 8 hours, while a 16TB drive can take over 24 hours. The test runs in the background, so your NAS stays online and usable the whole time.
Is running badblocks safe for my data?
The read-only version, badblocks -sv, is completely safe and does not touch your data. Only the write-mode test with the -w flag is destructive, and that one erases the entire drive. Never run the write test on a disk that is still part of your storage pool.
My drive passed SMART but the NAS still flags it. What now?
SMART can pass a drive that still throws errors in real use. Run an extended self-test and a badblocks read pass to dig deeper. Also check the UDMA CRC error count, because high values there usually point to a bad cable or connection rather than a failing drive.
How many reallocated sectors are too many?
A small, stable count of 2 or 3 that never changes is usually harmless. The danger sign is a number that climbs over days or weeks. Any current pending sectors above zero deserve a full extended test right away.
Can I test the drive without removing it from the NAS?
Yes. Both the SMART extended test and a read-only badblocks scan can run while the drive stays in the bay and the NAS stays online. You only need to pull the drive for a destructive write test or to use a Windows-only vendor tool.
Should I trust the vendor tool or SMART more?
Use both. SMART gives you the ongoing health trend, while the vendor tool gives a definitive pass or fail from the company that built the drive. A failure from the manufacturer's own diagnostic is also the language most warranty claims require.
