Monitor to make decisions before failure
A drive may report SMART PASSED yet show read errors or disappear from the controller. Conversely, an old counter may remain stable for a long time. Decisions require trends, real errors, data criticality, and the condition of other drives.
This guide covers SMART collection and Unraid replacement preparation. Device names are fictional. Confirm your Unraid version, filesystem, topology, and applicable official procedure before changes. These commands are not automatic diagnosis or reliable failure prediction.
1. Identify the physical drive and role
In Main, record slot, model, serial, capacity, and role: array data, parity, pool member, or unassigned device. Map the serial to a physical bay. A /dev/sdX letter can change after reboot.
Array parity does not protect data residing only in a pool. A nonredundant pool requires migration or restore rather than reconstruction from array parity.
| Role | Protection to verify | Procedure |
|---|---|---|
| Array data | Valid parity and readable remaining disks | Slot rebuild |
| Parity | Data health and other parity | Parity recalculation |
| Btrfs/ZFS pool | Actual redundant profile/topology | Pool-specific replacement |
| Unassigned device | Independent backup | Appropriate copy/restore |
lsblk -d -o NAME,SIZE,MODEL,SERIAL,TRAN
ls -l /dev/disk/by-id/2. Read SMART and keep a baseline
Open the disk under Main and export its report. smartctl reads health exposed by the device and controller; administrative access may be needed. Reports contain serial numbers, so redact them before public sharing.
Use the smartmontools version bundled with your Unraid and its local help. An HBA, USB bridge, or different protocol may require a suitable device type. Do not guess options. Failed collection and standby skips need separate reporting.
Retain time, temperature, errors, tests, and relevant counters outside the boot device. Reading logs does not change disk files but may wake the drive; account for spin-down behavior.
# Replace with the verified WHOLE device ID, not a partition.
smartctl --version
smartctl -x /dev/disk/by-id/ata-EXAMPLE_DISK_SERIAL
# May avoid waking an ATA disk in standby:
smartctl -n standby -a /dev/disk/by-id/ata-EXAMPLE_DISK_SERIAL
# Inspect the result and exit status; skipped is not healthy.3. Interpret counters for the model
For ATA HDDs, prioritize reallocated, pending, and uncorrectable sectors alongside tests and recent errors. Growth warrants investigation and backup. Raw and normalized values are different; units and identifiers depend on the manufacturer.
Growing UDMA CRC often points to the transfer path. Inspect cables, connections, and controller and correlate resets; it does not alone prove damaged media. Historical counts may remain after repair.
Power_On_Hours reflects accumulated usage; Load_Cycle_Count usually tracks head load/unload cycles. Compare model specifications without converting a counter into a failure date. SSD/NVMe wear and media-error indicators require their own interpretation.
4. Correlate health with server symptoms
Record read/write errors, disabled drives, link loss, unusual temperatures, and concurrent tasks. Download Tools → Diagnostics before reboot where possible and retain it securely.
A disabled drive is no longer used normally by Unraid; content may be emulated when protection permits. The fault can involve drive, cabling, power, or controller. Replacement without addressing the hardware path may repeat the failure.
Fictional case: D3 accumulates CRC after cabling work with no new media error. Inspect the link in a shutdown window and observe the trend. Another case: increasing pending sectors and a failed test; prioritize data and replacement planning.
5. Test without stacking heavy operations
A short or extended test starts internal work and adds load; it is not metadata-only collection. Schedule based on drive condition. A severely failing drive holding the only copy should not undergo repeated tests before recovery.
Extended tests can take hours. Avoid overlapping rebuilds, scrubs, and parity checks. A successful test does not guarantee future reliability. Read the result after the stated completion time rather than relying on the start confirmation.
# CHANGE: starts an internal test; adds drive load.
# Confirm identity, backup, health and maintenance window first.
smartctl -t short /dev/disk/by-id/ata-EXAMPLE_DISK_SERIAL
# After the indicated completion time, READ the result:
smartctl -l selftest /dev/disk/by-id/ata-EXAMPLE_DISK_SERIAL
# Optional extended test only when appropriate:
# smartctl -t long /dev/disk/by-id/ata-EXAMPLE_DISK_SERIAL6. Prepare backups, alerts, and maintenance
Test restoration of important files to a separate location. Back up boot configuration, share settings, and application data with appropriate consistency. Parity also follows deletion and formatting; it does not retain an independent version.
Enable and test Unraid notifications. Acknowledging an alert repairs nothing and does not replace history. Track changes and collection failures. Device limits and update rights depend on licensing; check Registration before capacity changes.
Record slots and serials, encryption-key access, outage expectations, replacement drive, and backups. Stopping the array affects shares and applications; plan with users.
7. Replace array data: choose the right scenario
For a planned upgrade on a healthy array, begin with the parity verification required by the official procedure and investigate errors. For an already failed drive, do not automatically run a lengthy check or correction: examine emulation and remaining drives first.
Unraid distinguishes replacement, parity, and parity swap. Replacement data capacity must cover the old disk and fit the smallest parity limit. Retain the old drive unchanged until validation. Do not Format an Unmountable disk; reconstruction also reproduces filesystem state.
The following is a change workflow to adapt to your version and official procedure. If emulation is inaccessible, stop and establish the appropriate recovery plan.
- Verify serials, backup, protection, and emulated data.
- Stop the array; for planned replacement follow Unraid’s unassign/start-without-drive sequence.
- Stop again; power down for physical work unless hardware supports hot-swap.
- Assign the replacement to the exact slot and start rebuild.
- Monitor progress/errors, then validate files and applications.
8. Apply separate parity and pool procedures
Parity replacement recalculates protection from data and creates a risk window. When desired capacity exceeds existing parity, use the applicable parity upgrade or swap procedure; do not improvise assignments.
For pools, verify actual profile and vdevs before removal. Multiple devices do not automatically imply redundancy. Follow documented Btrfs or ZFS steps, then verify rebuild, balance, or resilver completion and pool health.
Preclear is an optional community test that destroys target contents. It is not required for a rebuild and must never target useful data. This guide supplies no erase commands.
9. Validate rebuild and prepare recovery
Check completion, errors, mount status, important files, share permissions, and applications. A successful rebuild does not prove every file was correct before the incident. Compare existing checksums or reference restores where available.
Rollback is not automatically swapping the old disk back: emulated writes may have made it stale. Retain it without writes and use a validated restore or controlled recovery plan.
Do not use New Config to rebuild. Unraid warns that it clears the state needed to offer reconstruction. If another disk fails or the job stops, preserve diagnostics and reassess protection before retrying.
Before-and-after checklist
Storage and backup consulting can establish these checks and a replacement plan matched to your data. The goal is evidence and workable recovery rather than waiting for a SMART counter to predict failure.
- Physical identity, slot, and role confirmed.
- Independent backup and restore tested.
- Remaining drives and emulation assessed.
- Capacity, license, and topology compatible.
- Exact procedure selected; no improvised formatting.
- Results, important files, applications, and alerts validated.
About the author
Marc-Antoine Lambert
Montréal-based senior IT consultant and Team Lead specializing in architecture, cybersecurity, Microsoft infrastructure, networking and business continuity.
Discuss your environment →