Restic

Restic takes the backup problem from the right end: it sends your data encrypted to a remote repository (object storage, SFTP, external disk), transfers only what changed, and keeps every backup as a dated snapshot you can reach into for a single file. For an SMB, the stake isn't having a backup — almost everyone has one. The stake is knowing how long it takes to bring the service back from it, and whether anyone has ever checked.

• →Borg / BorgmaticSame philosophy, SSH repository rather than object storage: excellent when the destination is a machine you own
• →Host snapshotsImmediate and install-free, but hosted in the same account as production
• →PostgreSQL PITRFor a critical database, replication and point-in-time recovery complement file backupView page
• →AnsibleTo deploy and hold the same backup configuration across a whole fleetView page

1. 01

   3-2-1 rule: three copies, two media, one off-site — and outside the account that hosts production

2. 02

   Client-side encrypted repository, key kept somewhere other than the machine being backed up

3. 03

   Consistent application dump before the snapshot, never a hot copy of the files of a database being written to

4. 04

   Declared retention, pruned automatically, with scheduled integrity checks on the repository

5. 05

   Restore tested and timed, with the procedure written for someone who didn't build the system

• What does Restic add over a plain file copy?

  Three things a copy doesn't give you. Deduplication: only changed blocks go out, so thirty daily backups don't weigh thirty times the same data. The dated snapshot: you restore the state of a precise day, not the last known state that may already have overwritten the data you lost. Encryption before transfer: the storage provider holds unreadable blocks.
• Aren't my host's snapshots enough?

  They're useful and instant, but they live in the same hand as what they protect: a compromised account, a billing error or a deleted project takes both at once. They cover hardware failure and human error. They cover neither ransomware walking up the shares nor the loss of the account itself. A copy outside the provider stays necessary.
• Where are backups stored, and what does that storage cost?

  In S3-compatible object storage (OVH, Scaleway, Backblaze B2, Wasabi), over SFTP, or on a remote disk. That's a running cost, billed on the volume retained: on the order of a few euros a month for a few hundred gigabytes, sometimes with egress fees on a large restore — worth checking before picking the provider, not on the day of the incident.
• What does setting up a backup strategy cost?

  The tool doesn't drive the price; what has to be protected and how fast it has to come back does: the number of servers and databases, the volume, the retention to hold, whether monitoring exists, and the level of restore to guarantee — one file, one database, or the whole service. The amount is set in the quote, after a free initial scoping session that puts the scope in writing before any commitment.
• How often should a restore be tested?

  At every significant infrastructure change, and at least twice a year outside of any change. A restore test isn't a box to tick: it measures a duration, exposes forgotten dependencies (certificates, environment variables, permissions) and verifies that the procedure can be run by someone other than whoever wrote it.