sysop@hub:~/explainers/snapshot-is-not-a-backup$
snapshot-is-not-a-backup
Snapshots, RAID and replication all protect data, each against something different. None of them is a backup, and the difference only shows on the day something goes wrong. Below are eight such days. Click a cell to see why a protection holds or does not.
what survives what
| What went wrong | Snapshot | RAID | Replica | Backup, same site | Backup, off site | Off site and immutable |
|---|
Click a cell.
yes it saves you partly with luck or with losses no it does not help
why a snapshot is not a copy
A snapshot does not copy your data. It tells the storage: from now on, do not overwrite the old blocks, write changes somewhere else. Taking it costs nothing and takes a second, which is why it is so useful. It also means the snapshot is made of the same blocks, on the same disks, in the same system, managed by the same admin account as the data it protects. When any of those is lost, the snapshot goes with it.
Snapshots also get more expensive the longer they live. Every change since the snapshot is kept, so they grow. On most hypervisors a virtual machine with an old snapshot gets slower, and when the datastore fills up because of forgotten snapshots, every machine on it stops.
what makes something a backup
- A separate copy. Real data, not pointers to the original.
- On a separate system. Other disks, another device, ideally another place.
- Behind separate credentials. If your domain admin password can delete it, so can whoever steals that password.
- With history. Several points in time, reaching back further than it takes you to notice a problem.
- Restored at least once. Until you have restored from it, you have a hope, not a backup.
how to use each of them
| Good for | Rule | |
|---|---|---|
| Snapshot | A quick way back during a change: updates, migrations, risky edits. | Take it right before, delete it within days. Never the only protection. |
| RAID | Keeping the system running when a disk fails. | Monitor it. A degraded array nobody noticed protects nothing. |
| Replica | Getting back to work fast when hardware or a site fails. | It copies mistakes as faithfully as data. Keep it for uptime, not for history. |
| Backup | Everything else: deletions, ransomware, mistakes, slow damage. | Three copies, two systems, one off site, one that cannot be changed, and tested restores. |
Product names blur the terms. A "snapshot" that a cloud provider stores in a separate storage system, or snapshots that are replicated to a second machine, are closer to a backup. The questions stay the same: is it a separate copy, on separate hardware, behind separate credentials, with enough history? Related: backup planner, RAID calculator, backup and disaster recovery checklist, war story: the backup gap, ZFS and NAS cheat sheet.