Refresh the backup first, then replace fast: same capacity or larger, CMR only, NAS/enterprise class — brand does not matter. A degraded RAID5 has zero redundancy until the rebuild finishes, and the rebuild itself is the riskiest thing you will ask the surviving drives to do. Live in-stock replacements by capacity are right below.
A degraded array keeps serving files, which is precisely why people let the state persist — the red light is on but nothing seems wrong. Here is what the state actually means. On RAID5 or single-parity setups (SHR-1, RAIDZ1, classic RAID5), the failed member consumed the array's entire failure budget: one more member failure and the array is unrecoverable. The surviving drives are typically the same age as the one that died, often from the same purchase, and they are about to be asked to work harder than they ever have.
That extra work is the rebuild: to reconstruct the new member, the array reads every sector of every surviving drive, end to end, for hours or days. Two risks live inside that window. The dramatic one is a second drive failure under the sustained load. The statistical one is quieter: a single unreadable sector (URE) on any survivor can abort or corrupt a single-parity rebuild — and the probability of encountering one grows with the terabytes read, which is why the risk conversation changed as drives crossed into the double-digit-TB era. We state that risk directionally rather than quoting a fake precise probability: big arrays reading tens of terabytes end-to-end give rare errors real odds, and the odds get worse with each capacity generation.
This is also the structural argument for dual parity at modern capacities: with RAID6/RAIDZ2/SHR-2, the array survives both a second failure and rebuild-window read errors. If your array is 16TB-class members or larger and still single-parity, treat this failure as the prompt to fix that on the next rebuild — dual parity costs one drive of capacity and buys the failure mode that actually kills arrays. In the meantime, the sequencing for a degraded array is fixed: verify the backup first (the rebuild is when arrays die — do not start it as your only copy's custodian), then replace the member, then rebuild.
Live replacement options · prices updated every 4-5 hours · last checked 26 min ago · CMR-family drives only (RAID-safe)
| Drive | Cap | Cond | Warranty | $/TB | Price | |
|---|---|---|---|---|---|---|
| Toshiba MG Series 8TB Enterprise SATA | 8TB | REFURB | 5 yr | $10.63 | $85.00 | Buy → |
| MDD 3TB SAS 6G 7200RPM | 3TB | NEW | 5 yr | $11.33 | $33.99 | Buy → |
| Seagate Constellation ES.2 3TB SAS | 3TB | REFURB | see listing | $11.63 | $34.90 | Buy → |
| Seagate 3TB Enterprise Capacity SAS 6G | 3TB | REFURB | see listing | $11.67 | $35.00 | Buy → |
Same size or larger. Arrays rebuild onto a member at least as large as the one that failed. Larger works everywhere (the platform uses the matched portion until enough members grow), and in 2026's market buying one size up is frequently the better $/TB anyway — a failure is a legitimate excuse to start the capacity upgrade, and the upgrade planner prices exactly that "replacing a failed drive? size up while you're at it" move.
CMR is mandatory. Not preferred — mandatory. Shingled (SMR) drives collapse under rebuild write patterns, stretching a one-day rebuild into many and risking mid-rebuild timeouts that get the new member ejected. The picks above are filtered to conventional recording for this reason, and our CMR guide lists which retail models are safe if you shop elsewhere. Mixing brands is fine; mixing SMR in is not. Arrays are indifferent to labels, and mixed batches actually reduce correlated-failure risk — the thing arrays genuinely fear is a member that answers slowly.
NAS/enterprise class for array duty. Desktop drives without time-limited error recovery can stall an array with minutes-long internal retries and get themselves kicked — the intermittent-dropout pattern covered on the symptom page. Everything in our picks is NAS or enterprise class; if a bargain elsewhere is a desktop model, it is not a bargain for this job.
Synology: Storage Manager marks the failed slot; the repair flow is deliberately simple — hot-swap the member (on hot-swap models), then Storage Manager → Storage Pool → Repair, selecting the new drive. SHR pools accept larger members gracefully. Let the repair finish before resuming heavy workloads; DSM will happily serve files during a rebuild, but every I/O extends the window.
UnRAID: the parity model changes the math in your favor — only the failed disk's contents are at stake (each data disk is an independent filesystem), and the emulated disk keeps serving while you swap. Replace, assign, and let parity reconstruct. The caveat mirrors the parity design: parity must be valid, so if you have been skipping parity checks, this is the education. Our UnRAID drive guide covers the array-suitability details.
TrueNAS/ZFS: zpool status names the failed device; replace it and ZFS resilvers. Two ZFS-specific notes: resilver only copies live data (often faster than a block-level rebuild on a part-full pool), and scheduled scrubs are the mechanism that finds weak sectors on survivors before a resilver does — a pool that scrubs regularly rebuilds with far fewer surprises. RAIDZ1 at large member sizes carries the same single-parity caveats as everything above.
Sometimes the drives are fine and the box around them is the corpse — a NAS that no longer powers on, a backplane slot gone bad, a controller that boot-loops. The drives usually carry the array intact: Synology arrays migrate between Synology units, UnRAID arrays follow their disks, ZFS pools import on any system that speaks ZFS. Check your platform's migration documentation, move the disks in order, and — the one commandment — never accept a new system's offer to initialize "unknown" disks. Below: the cheapest in-stock devices if the box needs replacing today.
Live replacement options · prices updated every 4-5 hours · last checked 26 min ago
If the failed drive is in warranty, you are owed a replacement — but the RMA clock (commonly weeks, per community reports) runs while your array sits degraded, which is why the standard play for arrays is buy-now, RMA-anyway: end the exposure today, file the claim, and bank the returned drive as a cold spare that — in a shortage market — holds real value. The full decision math, with live per-capacity prices, lives in our RMA-or-replace guide (publishing next in this cluster), and the healthy drives you eventually retire still book money on the used-drive tracker. One last piece of rebuild hygiene while the new drive ships: pause noncritical workloads, hold off on scrubs and balance jobs until the array is whole, and resist the urge to reboot the NAS repeatedly — a degraded array wants calm, not attention. The rebuild will stress the survivors enough on its own; everything you defer until afterward is risk you chose not to take.
It works — and that is exactly the trap. A degraded RAID5 has zero remaining redundancy: the next member failure, or a single unreadable sector during the eventual rebuild, and the array is gone. Every day of degraded operation is a day of betting all the data on the health of the surviving drives, which are the same age and often the same batch as the one that just failed. Use it, but treat the state as an emergency in slow motion.
No. Arrays care about capacity (same or larger), recording technology (CMR), and suitability (NAS/enterprise class with sane error-recovery timing) — not the label. Mixing Seagate, WD and Toshiba in one array is routine and even reduces the odds of correlated same-batch failures. What you must never mix in is SMR.
Yes — same size or larger is the rule, and larger is often the smart buy in 2026's market. Most platforms use only the matched portion until enough members grow (Synology SHR and UnRAID handle mixed sizes gracefully; classic RAID uses the smallest member as the unit). Buying one size up turns a failure into the first step of a capacity upgrade — see the upgrade planner.
Shingled drives handle sustained random writes — exactly what a rebuild is — by shuffling data internally, and their write speed collapses under that load. Rebuilds that take a day on CMR can take many days on SMR, with timeouts that can get the new member kicked from the array mid-rebuild. Every drive in our picks is CMR-family for this reason; the /cmr-drives guide has the full story.
The rebuild reads every sector of every surviving member — the largest sustained read those drives have ever done, at the moment redundancy is zero (RAID5). The documented risk is hitting an unreadable sector (URE) on a survivor mid-rebuild, and it grows with array size: reading tens of terabytes end-to-end gives even rare per-bit error rates real odds. This is the structural argument for RAID6/dual parity at modern capacities, and for fresh backups before you start.
Usually both: buy now to end the degraded window today, then RMA the failed drive and keep the returned unit as a cold spare. An RMA is free but commonly takes weeks — weeks your array spends one failure from total loss. The full math (including what a cold spare is worth in a shortage) is in our RMA-vs-replace guide.
Two honest paths: replace the enclosure and migrate the drives (Synology-to-Synology and UnRAID migrations preserve arrays well; check your platform's migration documentation before assuming), or mount the drives directly in a PC to copy data off first. The live device picks below cover the replace-the-box path; never initialize 'unknown' disks the new system offers to set up — that offer is a data-loss button.
No — same rules as the replacement: capacity equal or larger, CMR, NAS/enterprise class. A cold or hot spare from a different brand or batch is arguably better than a twin, for the same de-correlation reason arrays benefit from mixed batches. What matters is that it is present before the next failure, not that it matches.