3-2-1 backup: copies that actually restore
Field: backup, disaster recovery, business continuity
For: SME owners who want to sleep soundly, and their IT person
Experience: the backup chain that protects my own servers, checked every night
A backup that has never been restored is not a backup: it is a hope. The 3-2-1 rule, encryption, retention and above all the restore test make the difference between a reassuring copy and a copy that saves the business. Here is the 3-2-1 backup as I practise it, and the questions a business owner should ask their IT person.
Background
Failing disk, human error, ransomware: the question is not whether a server will lose data, but when. I pointed out in my article on ransomware that backups remain the first way out. They still have to exist, be intact, and you have to know how to use them on the day.
The problem: fake backups
In SMEs, I often come across setups that look like a backup without being one:
- The copy on the same server: it disappears with it.
- Synchronisation: a shared folder that updates everywhere also replicates deletions and files encrypted by ransomware.
- Copying a running database: files copied while the database is writing do not guarantee a restore.
- The backup that fails silently: it worked the first month, then nobody looked again.
- The erasable backup: if the server can delete its own copies, an attacker who takes over the server can too.
The solution: six simple rules
1. Apply the 3-2-1 rule
Three copies of every important piece of data (the original and two backups), on two different media, one of them off your premises. This rule, recommended by the US cybersecurity agency, remains the basis. An off-site backup service, like the one I present in my article on Swiss Backup, is a simple way to have the remote copy.
2. Encrypt before sending
The data is encrypted on your server, before it leaves. The provider that stores your backups cannot read them. You keep the key, and a copy of the key in a safe place: without it, the backup is lost.
3. Back up the database properly
For an ERP, every night, I export the database into a consistent file, then read that file back to check that it is complete. Only then does the general backup take it off the server.
4. Stop the server from erasing its backups
The server can add backups, never delete them. Old copies are pruned by another machine. If an attacker takes control of the server, they cannot destroy your safety net.
5. Keep several points in time
An error is sometimes discovered weeks later. I keep one copy per day for a week, one per week for a month, one per month for a year, and one per year for two years. Thanks to deduplication, this takes far less space than you might think.
6. Test the restore
Every night, part of the backed-up data is read back to check its integrity. And a real restore of the database is carried out in a disposable environment, cut off from the network, then compared with production: number of tables, customers, invoices. I run it again at least once a quarter and after every major update. It is the only test that counts.
Questions to ask your IT person
- When did we last do a full restore, and how long did it take?
- In the worst case, how many hours of work do we lose?
- Who is warned if a backup fails, and how quickly?
- Can an attacker who takes over our server erase our backups?
- Where is the encryption key, and who can use it if you are away?
What I learned running it
- A copy can look like a backup without being one. My general backup also took the raw database files, copied while running. They took up space every night and gave a false sense of security. I removed them: the verified export is what counts.
- A new database is backed up by nobody. If someone creates a test database, it is not part of any routine. My backup now flags any database that nothing covers.
- The service can be "active" and the backups failing. A change at a provider broke the sending of backups without anything looking abnormal. What protects you is an alert on the age of the last successful backup, linked to monitoring.
Under the hood (for your IT person)
- Database: pg_dump in custom format, read back with pg_restore --list (a truncated export makes the script fail), local rotation, and a check for orphan databases.
- Scheduling: the database export runs before the general backup; never move one without the other.
- General backup: restic, client-side AES-256 encryption, deduplication, to a REST server in append-only mode, reached through the VPN.
- Retention: forget with 7 daily, 4 weekly, 12 monthly and 2 yearly, then prune, run by the machine that hosts the repository, not by the client.
- Checks: restic check on a sample of the data every night, freshness check (snapshot less than 48 hours old), notification on success as well as failure.
- ERP attachments: immutable and named by hash, they can be copied while running; that is not the case for the database engine's files.
What I take away
A good backup is judged on the day you need it, and on that day it is too late to improve it. Encrypted, off-site, impossible to erase from the server, monitored and tested: that is what turns a disaster into a bad day. Together with a server you know how to rebuild, it is the heart of the recovery plan, and the foundation of the map of an SME taking back control with open source.
Going further
I can audit your current backups, carry out a real restore test with you and set up a complete chain. See Your tools and your data, or book a 20-minute call.
The step-by-step guide
For your IT person, my guide no. 2, encrypted server backup (CHF 39), details the setup: a daily encrypted restic backup to Swiss online storage, a retention policy and a mandatory restore test. It is part of a complete stack, documented and tested in production, gathered in the Guides Pack (CHF 149).
Sources
- Data Backup Options (3-2-1 rule), US-CERT, now CISA, 2012, accessed 15 November 2026.
- restic, References (AES-256 encryption), official documentation, accessed 15 November 2026.
- restic, Removing backup snapshots (retention policy), official documentation, accessed 15 November 2026.
- restic, Working with repositories (integrity check), official documentation, accessed 15 November 2026.
- restic REST server (append-only mode), official repository, accessed 15 November 2026.
- PostgreSQL, SQL Dump and File System Level Backup, official documentation, accessed 15 November 2026.
- Photo: Denny Müller / Unsplash (unsplash.com/photos/white-and-silver-hard-disk-drive-1qL31aacAPA), Unsplash licence.
Tags: backup, 3-2-1 rule, restore test, restic, encryption, ransomware, business continuity, SME