U3 Storage and Data
Backup strategy: copies, versions and restores you have tested
Nobody cares about backups. Everybody cares about restores. This guide builds a backup strategy around the moment you actually need it: the day a drive dies, a disk fills with encrypted files, or someone deletes the wrong folder.
At a glance
- RAID and mirrors keep you running when a drive fails. They do nothing against deletion, ransomware, fire, theft or a corrupted filesystem.
- Keep at least three copies of important data, on at least two separate systems, with at least one off-site. Then test them.
- Separate the credentials. The system being backed up should not be able to delete its own backups.
- Encrypt off-site copies before they leave home, and store the key somewhere that will survive losing the house.
- An untested backup is a hope, not a plan. Schedule restore drills the way you schedule the backups.
RAID is not a backup
Redundancy protects against one specific failure: a drive dying. Every other kind of loss reaches every drive in the array at the same moment, because the array is one system. The table below shows what each layer actually protects against.
| Threat | RAID or mirrors | Snapshots | Local backup on another system | Off-site backup |
|---|---|---|---|---|
| One drive fails | Yes | No | Yes | Yes |
| Accidental deletion or overwrite | No | Yes | Yes, if versioned | Yes, if versioned |
| Ransomware on a workstation | No | Usually | Yes, with separate credentials | Yes |
| Compromised NAS admin account | No | No | Yes, with separate credentials | Yes |
| Controller or filesystem corruption | No | No | Yes | Yes |
| Power surge, flood or fire | No | No | Rarely | Yes |
| Theft | No | No | Rarely | Yes |
No single layer covers every row. That is the whole argument for a strategy rather than a product.
The 3-2-1 rule, interpreted for a home
The classic rule says three copies of your data, on two different types of media, with one copy off-site. In a homelab, the spirit matters more than the letter. Two separate systems with separate credentials are worth more than two kinds of media in the same box. Treat 3-2-1 as the minimum, and add the step most people skip: tested.
Decide what deserves protection
Not everything needs three copies. Back up by how hard data would be to replace and by what a lost day would cost you.
| Tier | Examples | Protection |
|---|---|---|
| Irreplaceable | Family photos and videos, personal documents, your own creative work, password vault exports | Full 3-2-1, versioned, encrypted off-site, restore-tested |
| Costly to recreate | Service configurations, VM images, container volumes, home automation setup, notes | Local versioned backup plus off-site copy of configurations |
| Replaceable | Media you could re-rip, installers, downloadable images, caches | Redundancy and maybe a second local copy, often no off-site copy |
Build the layers
Snapshots for fast undo
Snapshots on the NAS are your first line: instant, cheap and easy to restore from. Expose them to users as previous versions of files where your platform supports it. They protect against mistakes and most ransomware on workstations, provided the workstation cannot reach the storage administration interface.
A local backup on a second system
The second copy belongs on a different machine: a second NAS, a backup server or even a small box with a couple of large drives. It should keep versions, so an encrypted or corrupted file does not overwrite the good copy, and it should have its own credentials.
An off-site copy
The third copy leaves the building. Good options for a home:
- A reciprocal arrangement with a friend or relative. Each of you hosts a small encrypted backup for the other, connected over a VPN or mesh overlay. Seed the first copy locally, then send only changes.
- Object storage from a cloud provider. Inexpensive per terabyte, but watch request costs, retention settings and your monthly bill.
- Rotated offline drives. A drive you take to work or a family member's house. Simple and effective, if you actually rotate it.
Myth
"File sync is a backup." Sync faithfully copies deletions, corruption and encryption to every device within seconds. Sync with versioning and snapshots on the hub is a useful layer, but it is not one of your three copies.
Separate the credentials
Ransomware operators and intruders look for backups first. If the account that runs your NAS can also delete your backups, one stolen password ends both. Design so that no single credential can destroy every copy.
- Pull, do not push. Let the backup system connect to the source and pull data with a read-only account. The source never holds credentials for the backup.
- Write but not delete. Where the backup target supports it, give the hypervisor or NAS permission to add backups but not to remove or prune them. Pruning runs on the backup system itself.
- Different admin passwords and multi-factor authentication on every backup system, stored in your password manager, not reused anywhere.
- Keep the admin interface of backup systems off the household network.
- Rotate every credential after any intrusion, before restoring. Otherwise the intruder can return and do it again.
Caution
"Immutable" backups are only as immutable as the account that controls them. Retention locks and protected snapshots can usually be removed by someone with full administrative access. Protect those accounts as carefully as the backups themselves.
Encryption and keys
Encrypt every off-site copy on your own system before it leaves home, so you never depend on someone else's promises about privacy. Then solve the problem that has destroyed more backups than any failed drive: the key.
- Store the encryption key or passphrase in at least two places that are not the system being backed up, such as your password manager and a sealed paper copy kept with important documents.
- Test that you can decrypt a backup on a different machine using only the stored key.
- Tell one trusted person where the key and the restore instructions are kept.
Virtual machines, containers and configurations
- Back up every VM on one nightly schedule so new VMs are never forgotten. Use the stop or snapshot mode your hypervisor offers; stopping briefly gives the most consistent result for a lab that sleeps at night.
- Dump databases to files before backing them up instead of relying on a live disk image of a running database.
- For containers, back up the volumes and the compose files. The images can be pulled again; the data and configuration cannot.
- Export network configurations from the firewall, switches and any controller after every change, and include them in your backups. These files often contain VPN keys and passwords, so encrypt them.
- Back up the documentation that tells you how to rebuild everything, and keep a copy you can read when the lab is down.
Watch the backups, not just the data
Backups fail quietly. A job that stopped running in March looks exactly like a job that runs every night until the day you need it.
- Have each backup job send a check-in ping to a monitoring service only when it finishes successfully. Alert when the ping does not arrive. This catches jobs that never start, which error alerts cannot.
- Review backup reports and sizes weekly. A backup that suddenly shrinks or grows is worth investigating.
- Check cloud bills. An unexpected jump can mean duplicate uploads or a retention setting that never deletes anything.
Test restores on a schedule
- Monthly: restore a few files. Pick files from different datasets and dates, restore them to a scratch location and open them.
- Quarterly: restore a service. Restore one VM or container stack to an isolated network and confirm it starts and its data is intact.
- Yearly: run a full drill. Pretend the house is gone. Using only the off-site copy, your stored key and your written instructions, restore something important to a different machine.
- Write down what went wrong. Every drill finds something. Fix it, update the runbook and record the date.
Our backup verification checklist turns this into a printable routine, and The Restore Nobody Tested tells the story of why we take it so seriously.
Everything degrades
Writable optical discs fade, flash storage loses charge when left unpowered, and hard drives on a shelf for years may not spin up. Treat every medium as perishable. Keep backups on systems that check their own integrity, read old backups back periodically, and migrate archives to new media before the old media fails.