Map the physical hardware
Use Drives to filter by server, health, model, serial or label. Define enclosures and bays to map physical positions, including empty and missing slots. Assigned locations appear in drive inventory.
Investigate the evidence
Compare temperatures, SMART state, power-on age, sector indicators and NVMe endurance where reported. Unknown SMART is not a passed test. Diagnosis combines latency, queue depth, throughput, utilization and physical bay links; outliers are investigation hints, not failure predictions.
Use stable identities
Guarded disk actions use short by-id names, backed by GUID and live identity checks. ZFS volume devices such as /dev/zdN are virtual volumes, not physical replacement candidates.
Read I/O history correctly
Disk I/O is sampled from Linux diskstats on normal polls. Missing samples and counter resets create gaps. Measurements do not capture every burst between polls, and busy time alone is not an NVMe saturation score.
Source reference
Read the upstream documentation ↗Reviewed against main on November 1, 2026. Interface labels and capabilities can change.