Introspect and manage disks

Common commands for inventorying, monitoring and driving a host’s disks. smartctl is provided by the administrator profile, the other tools are part of the base system.

Inventory

Command	What it gives
lsblk -o NAME,SIZE,MODEL,TRAN,ROTA,MOUNTPOINTS	Disk tree, transport (usb, nvme, sata), rotating or not
findmnt /mnt/data	Source, type and actual options of a mount
duf	Free space, readable
ls -l /dev/disk/by-id/	Stable identifiers (wwn-, ata-, usb-)

Always designate a disk by its stable identifier

/dev/sda changes from one boot to the next. In a configuration or a udev rule, prefer /dev/disk/by-id/wwn-0x..., which follows the physical drive, or the UUID of the filesystem, which follows the formatting.

SMART health

Overall verdict first, detail next :

sudo smartctl -H /dev/sda          # PASSED ou FAILED
sudo smartctl -a /dev/sda          # tous les attributs
sudo smartctl --scan               # disques vus, avec leur type (-d)

A self-test is triggered on demand and read back later :

sudo smartctl -t short /dev/sda    # ~2 min, non bloquant
sudo smartctl -l selftest /dev/sda # résultats des derniers tests

USB drives

External enclosures go through a SAT layer. If autodetection fails, force the type with -d sat. smartctl --scan shows the type retained.

Is the disk spinning?

sudo smartctl -n standby -i /dev/sda

Response	State
Power mode is: ACTIVE or IDLE	Platter spinning
Device is in STANDBY mode, exit(2)	Motor stopped
Device is in SLEEP mode, exit(2)	Motor stopped, one level lower

Never omit -n standby

Without this option, smartctl wakes the disk up to query it : the answer will always be “active”, and the test will have caused what it was measuring.

Setting or forcing standby

1. Read the current settings

   sudo smartctl -n standby -g all /dev/sda

2. Allow standby, then arm the internal timer

   sudo smartctl -d sat -s apm,127 /dev/sda      # niveau APM >= 128 : jamais d'arrêt
   sudo smartctl -d sat -s standby,240 /dev/sda  # 240 x 5 s = 20 min d'inactivité

3. Or stop the disk right away

   sudo smartctl -d sat -s standby,now /dev/sda

Volatile settings

APM and the standby timer are lost every time the disk loses power. Replay them from a systemd unit triggered by a udev rule on the disk identifier, which also covers host startup.

When the internal timer never fires

Any command sent to the disk resets its idle counter, including a SMART read. An exporter querying the disk every minute is enough to push the deadline back indefinitely. The remedy is an active stop : a systemd timer compares the I/O counters at regular intervals and sends standby,now itself when nothing has moved. A disk that is already stopped is not woken up for all that, -n standby making the command exit without emitting anything.

Measuring actual activity

grep " sda " /proc/diskstats

Fields 4 and 8 : completed reads and writes. Comparing two readings taken apart gives the disk’s actual activity. SMART commands do not appear there, which makes it the right sensor for judging idleness.

For history rather than a snapshot, the metric node_disk_io_time_seconds_total is collected on every monitored host, see Supervision.

On-demand mounts

An x-systemd.automount mount is carried by two units :

systemctl status mnt-data.automount   # le déclencheur
systemctl status mnt-data.mount       # le montage lui-même
findmnt /mnt/data                     # les deux lignes, autofs puis le vrai FS

Automatic unmount rarely takes effect

x-systemd.idle-timeout only unmounts if nobody crosses the mount point anymore. On a monitored host, the metrics exporter performs a statfs() there on every scrape, which resets the counter continuously. No consequence for the hardware : a statfs() is served from memory and does not wake a sleeping disk.

btrfs filesystems

sudo btrfs filesystem usage /mnt/data  # occupation réelle, données et métadonnées
sudo btrfs device stats /mnt/data      # compteurs d'erreurs par périphérique
sudo btrfs scrub start -B /mnt/data    # vérification complète, bloquante

Backups

For restic repositories and the disks that carry them, see Network backups.