How many reallocated sectors is too many?

There is no safe number of reallocated sectors above zero, and there is no number that means the drive dies tomorrow. What decides it is the direction. A small count that stays flat means keep the drive, check your backup and watch it. A count that keeps rising means replace it. So do pending sectors that will not clear.

The figures behind that are not small. Google found that drives were over 14 times more likely to fail within 60 days after their first reallocation. We counted Backblaze's published data for April to June 2026. Toshiba, WD and HGST drives with one reallocated sector on 1 April failed by 30 June at 2.15%. Drives with none failed at 0.16%.

What a reallocated sector is#

A hard drive stores data in small blocks called sectors. When the drive decides a sector is damaged, it moves that address to a spare sector from a reserve pool. That move is a reallocation. Google's paper puts it this way: the drive "can remap the faulty sector number to a new physical sector drawn from a pool of spares". The count only goes up.

Seagate's openSeaChest documentation says "all drives have a fixed number of sectors available for reallocation". The pool size varies by capacity and vendor, and no vendor publishes it in SMART.

SMART is the drive's own health report. You read it with smartctl -A /dev/sda. Each line is one attribute, with a normalized value, a vendor threshold and a raw value. For sector counts, the raw value is the real number. Three attributes carry this story:

Backblaze also tracks 187 (errors reported to the computer as uncorrectable) and 188 (command timeouts). Only about a third of its drives reported those two on 1 April 2026.

The difference between 5 and 197 matters most. The smartd manual describes a pending sector as one the drive "would like to mark as 'bad' and reallocate", usually because a read failed. So "there is some unreadable data on the disk". A reallocated sector is damage the drive already dealt with. A pending sector is data it cannot read right now.

Two answers circulate, and both leave something out#

The first answer says a few reallocated sectors are normal. Its best evidence is the SMART status itself. That status only says FAILED when a normalized value falls below the vendor's threshold. Seagate says that "as a matter of policy, Seagate does not publish attributes and thresholds". Scrutiny's README says some thresholds are "so high that they can only be used to confirm a failed drive". So a drive with hundreds of reallocations can still read PASSED, and people take that as permission.

The second answer says the threshold is one. That comes from Google's paper, and the paper means something narrower. It defines a critical threshold as the value that raises the chance of failure in the next 60 days at least tenfold, compared with drives at zero. For reallocations, that value was one. It is a statement about risk, not about death. The same paper found about 85% of drives still running more than 8 months after their first reallocation.

So the second answer is right about the risk, and the first is right that the drive usually keeps going. Neither tells you what to do. The direction of the count and the pending sectors do.

What the numbers say#

Google's "probational count" means sectors on probation until they are reallocated or work again. The paper gives no SMART attribute numbers, so read it as close to 197, not as 197 exactly.

Backblaze's 2016 post says a SMART 5 raw value of two "means little until combined with other factors". Its reports give no failure rate per sector count, so we counted one.

Our count: failure within 91 days, by reallocated sectors on 1 April 2026#

We took every hard drive in Backblaze's 1 April 2026 snapshot and its attribute 5 value. Then we checked which drives had a failure record by 30 June. Seagate counts are divided by eight and rounded up, for the reason in the next section.

Three things stand out. Risk rises with the count at every step. On Toshiba, WD and HGST drives, one sector was already about 13 times the zero rate, close to Google's figure. And pending sectors were worse than a moderate reallocated count. Across all makers, drives with both reallocated and pending sectors failed at 12.40% within 91 days. Drives with reallocations and no pending sectors failed at 1.45%.

That is the part forums miss. A drive with a few remapped sectors and nothing pending has dealt with its damage. A drive with sectors pending is still finding more. Note the window, too: 91 days, not a year.

Why a Seagate count of 8 can mean one sector#

Many drives use 4,096-byte physical sectors but present them as eight 512-byte logical sectors. That format is called 512e. Seagate's openSeaChest wiki says such a drive "will most likely mark all of these bad and reallocate them all at once". It continues: "This is a single physical sector reallocation, but since this counter is for logical sectors, it grew by 8."

Backblaze's data shows it. Of 9,219 Seagate drives with a nonzero count on 1 April, 92% held a multiple of 8. Among 9,036 Toshiba, WD and HGST drives it was 6%. The median nonzero count was 26 on Seagate and 4 on the others. So "my drive has 24 reallocated sectors" may mean three on a Seagate.

Pending sectors that clear on write#

A pending sector can end two ways. openSeaChest says a write lets the firmware "either rewrite it or reallocate it depending on if it can be rewritten or not". If it reallocates, attribute 5 goes up. If the rewrite works, only the pending count drops. Google's paper agrees: "sectors on probation may indeed never be reallocated."

In Backblaze's data, 7,073 hard drives had pending sectors on 1 April. By 30 June, 377 of them showed zero pending and had not failed. Only 86 of those 377 had a higher reallocated count. For the other 291, the sectors went back into service without a new reallocation.

So the useful move with a pending sector is to get it written. On a redundant pool, a scrub reads every block and repairs a bad one from the good copy. ZFS, btrfs, mdadm or one disk has the scrub command for ZFS, btrfs, mdadm and SnapRAID. On a single disk, the data in that sector is already lost. The smartd manual says you can force the reallocation by writing to it, "but at the price of losing the 512 bytes of data stored there". The hdparm manual calls its --write-sector option "VERY DANGEROUS" for the same reason.

Then read 197 again. Back to zero with 5 flat: keep watching. Back to zero with 5 up a little: the drive retired the sector, which is what it should do. Still pending after the write, or new ones appearing every week: replace the drive.

A worked case: two readings a week apart#

Take a 12 TB Seagate drive in a ZFS mirror. Suppose smartctl -A shows these raw values on a Monday:

  5 Reallocated_Sector_Ct   ...   16
197 Current_Pending_Sector  ...   0
198 Offline_Uncorrectable   ...   0

Sixteen on a Seagate is two physical sectors. Nothing is pending. That falls in the 2 to 5 row, 1.32% within 91 days against 0.20% at zero. The ruling is to keep it, confirm the off-site backup restores, and set an alert on any change.

Now suppose the next Monday shows 5 at 40, 197 at 8 and 198 at 8. Forty on a Seagate is five physical sectors, so the size alone still looks small. But the count rose in a week, and sectors are pending. Drives with both reallocated and pending sectors failed at 12.40% within 91 days. Run zpool scrub. If 197 is still above zero afterwards, or 5 keeps rising, order the replacement now. Resilver onto it while the mirror still has two readable halves.

The same case on a WD drive changes one thing. Sixteen there means sixteen sectors, which is the 6 to 50 row at 6.71%. Same rule, less time.

When to replace a hard drive: the rulings above in one table#

How to get told when the count moves#

smartd, part of smartmontools, does this without a dashboard. -R 5! warns whenever the raw reallocated count changes; the manual gives it as the way "to warn when new sectors are reallocated". -C 197+ reports only when the pending count goes up, because "some disks do not reset this attribute when a bad sector is reallocated."

/dev/sda -a -R 5! -C 197+ -U 198+ -m you@example.com

Scrutiny does the same over time, with a history graph per drive. Its README says it uses "real world failure rates" rather than manufacturer thresholds. The minimum viable monitoring stack covers getting that alert to your phone.

What this answer does not cover#

These figures are for hard drives. Our table leaves SSDs out, and their attributes are vendor specific, so none of the rates apply to them. None of the sources reports SMR drives separately, so nothing here is specific to SMR. SAS drives report a grown defect list instead of these attributes. All the data comes from datacenter drives that run all day; nothing here measures a desktop drive that spins down.

Backblaze's failures are partly its own decision. Its Q3 2025 report says its drive sentinel counts uncorrectable errors, and "if it exceeds a threshold, access to the drive will be removed." So a drive can count as failed because of the errors we sorted by. Our count is one quarter of Backblaze's raw daily files, boot drives included, so it shows 1,522 failures where Backblaze's report lists 1,498. Google's drives were 80 to 400 GB, watched in 2005 and 2006.

None of this changes the ruling. A rising count or a pending sector that will not clear is a drive to replace. What you have to confirm yourself is that your backup restores. Backups that actually restore covers that.

What to do next#

Run smartctl -A on every disk today and write down 5, 197 and 198. Set the smartd line above so the next change reaches you. If a pool has a disk with pending sectors, scrub it this week, not this month. And if a disk is already climbing, buy the replacement before the rebuild needs it.