Data protection & recovery
Data rebuild
A data rebuild is the process by which a storage system recreates the data that was held on a failed drive or node, using the redundant information stored elsewhere, and writes it to healthy hardware. Until the rebuild finishes, the affected data has less protection than it was designed to have.
How does a rebuild work?
When a drive or server fails, the system identifies which pieces of data it held. For replicated data, it copies a surviving replica to a new location. For data protected by parity or erasure coding, it reads the remaining fragments, recomputes the missing ones and writes them out. When every affected piece has been recreated, redundancy is back to its intended level.
How do RAID and distributed rebuilds differ?
The approach depends on the architecture.
| Aspect | Traditional RAID | Distributed storage |
|---|---|---|
| Rebuild target | A replacement or spare drive | Free space across many nodes |
| Data read from | The other drives in the same group | Many drives across the cluster |
| Parallelism | Limited to the group | High, many sources and destinations |
| What is rebuilt | The whole drive, including empty space | Only the data actually stored |
| Waiting for a replacement part | Often required | Not required to start |
Why does rebuild time matter?
Rebuild time is the window of exposure. If further failures occur within it and exceed what the protection scheme tolerates, data is lost. Larger drives lengthen the window, because there is more to recreate at a write speed that has not grown in proportion. A RAID group of very large drives can take days to rebuild. Durability calculations therefore depend on rebuild time as much as on drive failure rate, and schemes that tolerate two or more simultaneous failures have become standard for that reason. Data durability covers the arithmetic.
What affects rebuild speed and its impact?
Three things set the pace: how much data has to be recreated, how many drives can take part in parallel, and how much of the system's capacity is given to the rebuild. Rebuilds compete with client traffic for disk and network, so systems throttle them, trading a longer exposure window for steadier performance. Many prioritize the data at greatest risk, the objects that have lost the most redundancy, ahead of the rest. A system near full has less room to rebuild into, which is one reason to keep headroom. Object storage rebuilds looks at the trade-offs.
Frequently asked questions
Is data available during a rebuild?
Yes, in a correctly designed system. Reads of affected data are served by reconstructing it from the remaining pieces, which is slower.
Does a rebuild start automatically?
In most distributed systems, yes, after a short delay to rule out a transient fault. Traditional RAID may wait for a spare or replacement drive.
What is the difference between a rebuild and a rebalance?
A rebuild restores lost redundancy after a failure. A rebalance redistributes healthy data, for example after adding capacity.
Related terms
- Erasure coding: the scheme rebuilds recompute from.
- RAID: where rebuilds are confined to a drive group.
- Data durability: what rebuild time directly affects.
- Annualized failure rate (AFR): how often rebuilds are triggered.
- Bit rot: silent damage that a rebuild may uncover.