A Synology reporting one failed disk and grinding audibly from a second bay, with a routine data scrub queued for the Sunday. Scrubbing is genuinely good practice. On that array in that state it would have finished it.
A Synology DiskStation with two 3 TB drives, used as the household and small-business file store. It crashed without warning and began making a loud screeching noise accompanied by several beeps.
The client did what most people do. They powered it off and on again, several times, and the noise persisted. Reasoning that the sound was coming from one particular drive, they removed that drive and fitted a new one in its place, expecting the unit to rebuild.
It did not boot. But the noise had stopped, which felt like progress. It was not.
This is the single most consequential misunderstanding in small NAS ownership, and it is entirely understandable.
The unit was configured as RAID 0. Two 3 TB disks presented as a single 6 TB volume, striped for capacity and speed, with no parity and no mirror — nothing held in reserve at all.
The client's mental model was reasonable: two drives, so if one fails the other has a copy. That is RAID 1, and it is what most people assume a two-bay unit is doing. RAID 0 gives you the larger number on the box, and a great many units ship configured that way or are set up that way during a first-run wizard that presents capacity as the obvious choice.
The practical consequence is severe. On a stripe, every file is split across both disks. Removing the noisy drive did not remove a failed copy — it removed half of every file on the volume, which is why the unit would not boot with a blank disk in its place.
If your two-bay NAS shows the combined capacity of both disks, it is striped. A mirrored pair reports the capacity of one. That single check takes ten seconds and tells you whether you have redundancy or not.
Both drives arrived, including the one that had been removed. The healthy member imaged without incident. The failed member had two distinct problems: a read/write head failure, and minor media damage across the platter surface — the screeching the client had heard was the heads in contact with the disk.
The repeated power cycling is relevant here. Each restart loaded the damaged heads back onto a surface they were already contacting, and every one of those cycles extended the affected area. The media damage found was consistent with a fault that had been made worse after the fact rather than one that arrived that way.
The recommendation was a platter swap: transferring the platter pack into a donor chassis with functioning heads, in filtered air, and reading from there.
Ninety-eight per cent of sectors imaged, and correspondingly ninety-eight per cent of the data extracted. On a striped volume that arithmetic does not distribute the way people expect: the loss is not two per cent of files, it is small gaps spread across many files, because every file crosses both disks.
In practice most documents and photographs came back complete, because a two per cent loss on a 6 TB volume leaves most individual files entirely untouched. Large files — video in particular — were more likely to carry a gap. The client received a full list of what was recovered and what was affected before paying anything.
Had the drive been switched off at the first screech rather than power-cycled repeatedly, that figure would very likely have been higher.
Combined capacity means striped. Single capacity means mirrored. If you are not certain which your two-bay unit is running, that is worth ten seconds today.
A screeching drive is heads contacting the platter. Every restart extends the damaged area. The instinct to try again is the single most expensive one available.
On a stripe there is nothing to reconstruct from. The removed drive was not a spare part — it held half of everything.