NVMe wear: check it before it checks you

SSDs wear out; NVMe reports it:

nvme smart-log /dev/nvme0

Watch percentage_used, data_units_written (in units of 512000 bytes) and media_errors.

A 600TBW drive written 100G a day lasts about sixteen years in theory — in practice controllers and firmware bugs fail first.

Habits that help: keep logs off the SSD or bound them, use a sane fsync policy for databases, and avoid repeated full-disk dd tests. A periodic fstrim reduces write amplification.

On cloud disks the provider owns the hardware, but you should still track percentage_used and plan migrations ahead of it.