Treat an SSD health alert as evidence to investigate alongside storage errors and backups. Record the drive model, serial and exact warning before requesting replacement. A single percentage value is not a reliable prediction of the day a drive will fail.
Read the fields in context
NVMe health reporting includes critical warnings, temperature, spare capacity, endurance estimates and error information. The percentage-used field is an endurance estimate, not the filesystem’s used-space percentage. Vendor and device context still matter when interpreting a report.
| Signal | Next check |
|---|---|
| Critical warning | Preserve data and investigate the specified condition |
| Increasing media errors | Compare with kernel and application I/O failures |
| High temperature | Review sustained load and cooling with the provider |
| Wear estimate increasing | Plan replacement using workload and vendor guidance |
| Filesystem nearly full | Investigate space usage separately from device wear |
Collect useful evidence without a stress test
- Confirm which physical serial corresponds to the affected storage volume.
- Save the read-only health report from the supported OS or controller tool.
- Record the UTC time and related kernel or application errors.
- Check the current backup and array state.
- Send the report and serial through the service support channel.
A VPS may expose a virtual disk rather than the physical drive’s telemetry. A hardware RAID controller can also change how disks are addressed. Do not assume a missing SMART report proves that a drive is healthy or that you have direct device access.
Coordinate recovery before replacement
If data is at risk, prioritize preserving accessible data over running a long benchmark. Agree the replacement target, downtime and rebuild procedure. Never remove a member based only on an unverified device name. After replacement, confirm redundancy, backups and application behavior have recovered.
Use RAID and backup planning to understand the recovery path and the remote-hands checklist when you own the physical machine.
Sources and references
Hosting documentation. Publication and update dates reflect this edition.
Related articles
- RAID Is Not a Backup: Planning Dedicated Server Storage RecoverySeparate availability after a disk failure from recovery after deletion, corruption or a lost server.2 min read
- Writing a Clear Remote-Hands Request for Colocated HardwareGive technicians an exact target, ordered steps, a stop condition and an acceptance check.2 min read
- NVMe vs SATA SSD for Servers: What Changes Your Workload?Compare latency, concurrency, durability and capacity instead of treating one headline throughput number as application performance.3 min read
