sysop@hub:~/tools/smart-reader$
smart-reader
Every disk keeps a record of its own health, and the record is hard to read: thirty attributes, three numbers each, and raw values that are sometimes a count and sometimes vendor code. Paste the output of smartctl -a here and get the verdict, the attributes that matter in plain words, and whether a problem is the disk or just the cable. The text is read by your browser and never uploaded.
getting the output
| System | How |
|---|---|
| Linux | sudo smartctl -a /dev/sda, for NVMe sudo smartctl -a /dev/nvme0. The package is called smartmontools. lsblk lists the disks. |
| Windows | Install smartmontools, then in an elevated prompt smartctl --scan to list the disks and smartctl -a /dev/sda. |
| macOS | brew install smartmontools, then smartctl -a disk0. |
| NAS (Synology, QNAP, TrueNAS) | Log in by SSH and run smartctl as above. The web interface shows the same data in its own layout. |
| Proxmox, ESXi | Proxmox: as on Linux. ESXi: esxcli storage core device smart get -d DEVICE gives a shorter list this page does not read. |
| Behind a USB adapter | Add -d sat. Some adapters pass nothing through at all. |
| Behind a RAID controller | The controller hides the disks. Address them through it, for example -d megaraid,0 or -d cciss,0, one number per disk. |
what matters and what does not
| Look at | Why |
|---|---|
| Reallocated, pending and uncorrectable sectors (5, 197, 198) | The three that predict failure best. Zero is healthy. Above zero and rising means replace. |
| CRC errors (199) | A cable or connector problem. Replacing the disk does not help, replacing the cable does. |
| Self-test results | A failed long test is a failed disk, whatever the attributes say. |
| NVMe: critical warning, media errors, percentage used | NVMe drives report health directly instead of through vendor attributes. |
| Temperature | Heat ages every drive. Hard disks like 25 to 45 °C. |
| Not: the raw read and seek error rate (1, 7, 195) | On Seagate and some others these raw numbers are enormous on perfectly healthy drives. Their normalised value is what counts. |
| Not: "PASSED" on its own | The overall result only fails when a threshold is crossed, which is late. Drives die with PASSED on the screen every day. |
SMART is a hint, not a promise: a good share of drives fail without any warning in these numbers, and some with bad numbers run for years. The decision is easy in one direction only: bad sectors that grow mean replace. Either way the answer to a dying disk is a backup that exists before it dies. Vendors use some attributes differently, and for those the page shows the raw value without judging it. Related: RAID calculator, backup planner, ZFS and NAS cheat sheet, Linux disks and LVM.