Skip to content
All articles

Backup and restore testing: a practical guide

How to test that your backups restore, how often to do it, a restore test checklist and a worked example for a small organisation.

Security6 min read
A hard disk drive with its cover removed

Most organisations have backups. Far fewer have proof that those backups restore, in a reasonable time, to a system that works. A backup you’ve never restored is an assumption, and the day you find out it was wrong is usually the day you need it most.

Restore testing closes that gap. It doesn’t need special tools or a big budget: a separate environment, a checklist and an hour or two on a schedule.

Why backups fail when you need them

Backups rarely fail loudly. More often, something has quietly gone wrong months earlier:

  • The job stopped running. A full disk, an expired password or a server that was moved, and nobody saw the alert.
  • Only part of the system is backed up. The database is there, but not the uploaded files, the configuration or the scheduled jobs that make the application work.
  • The backup can’t be read. It was encrypted with a key that’s no longer kept anywhere, or it was written in a format the current software won’t open.
  • The restore takes too long. Everything is there, but bringing it back from slow storage takes two days when the business can manage two hours.
  • Nobody knows the steps. The person who set it up has left, and the process lives in their head.

Each of these shows up in a test. None of them shows up in a backup report that says “completed”.

Agree how much you can lose, and for how long

Two numbers shape every backup decision:

  • Recovery point objective (RPO): how much recent data you could afford to lose. If it’s one hour, you need backups at least hourly.
  • Recovery time objective (RTO): how long the system can be down before it seriously hurts. If it’s four hours, restoring from a slow archive won’t do.

Agree these with the people who depend on the system as well as the IT team, because they’re business decisions. A finance system at month end and a staff intranet need very different answers.

Keep more than one copy, in more than one place

A widely used rule of thumb is 3-2-1: three copies of your data, on two different types of storage, with one copy kept away from the main system.

Ransomware has made the “away” part more important. Attackers look for backups and encrypt or delete them too. The UK’s National Cyber Security Centre recommends keeping at least one backup offline or otherwise separated from your network, so it can’t be reached from a compromised account. Most cloud providers offer immutable or locked backups that can’t be changed or deleted for a set period, which serves the same purpose.

The four kinds of restore test

Not every test has to be a full rehearsal. Mixing these gives good cover without taking over anyone’s week.

File spot check. Pick a handful of files from different dates and folders and restore them to a separate location. Open each one. This catches broken jobs and missing folders quickly.

Database restore. Restore the database to a separate server, run the database’s own integrity check, and compare record counts and a few known totals against the live system.

Full system restore. Bring the whole application back in a separate environment: database, files, configuration and secrets. Then use it the way a member of staff would, signing in, finding a record and running a report.

Recovery rehearsal. Walk through a realistic incident from start to finish, such as “the main server is gone”, using only the runbook. Time it, and note every point where someone had to guess.

How often to test

There’s no single right answer, so treat this as a starting point and adjust it to your recovery targets:

Test For important systems For everything else
File spot check Monthly Quarterly
Database restore Monthly Quarterly
Full system restore Quarterly Yearly
Recovery rehearsal Yearly When the system changes a lot

Test again after any big change, such as a migration, a new database version or a change of backup provider. Those are exactly the moments a backup quietly stops covering everything.

A restore test checklist

  1. Restore to a separate environment, never over the live system.
  2. Note which backup you used and its date and time.
  3. Restore everything the system needs, not just the database: files, configuration and secrets.
  4. Run the database’s integrity check.
  5. Compare record counts and a few known figures against the live system.
  6. Sign in and use the system as a member of staff would.
  7. Time the restore from start to working system.
  8. Compare that time with your recovery time objective.
  9. Write down what you did, what worked and anything that surprised you.
  10. Fix what you found and update the runbook before the next test.
  11. Delete the test copy when you’re done, as it holds real data.

A worked example

Take a 40 person organisation with a line of business application in the cloud: a SQL database of around 20 GB, a few thousand uploaded documents, and nightly backups kept for 30 days.

They agree an RPO of 24 hours and an RTO of four hours, because staff can work from paper for a morning but not for a day.

For the quarterly full restore, someone from IT restores last night’s database backup and the document store to a separate test environment, copies across the application’s configuration, and signs in as an ordinary user. They open five records created yesterday, run the monthly report, and compare its totals with the live system. The whole thing takes two and a half hours, inside the four hour target.

The test turns up two things. Documents uploaded in the last week of the month were missing, because the document backup ran before a late batch upload; and the restore needed an application setting that was only written down in an old email. Both are fixed that week: the backup schedule moves later, and the setting goes into the runbook. Neither would have been found from the backup report, which showed every job as completed.

Keep a record of every test

A short log is enough, and it’s what an auditor or insurer will ask for:

Date System Backup used Time to restore Result Actions
5 Nov Case management 4 Nov, 23:00 2 h 30 m Working, two gaps Backup time moved, runbook updated

Write the runbook before you need it

When something goes wrong, people are under pressure and memory fails. A short runbook, saying where the backups are, who has access, the exact restore steps and who to tell, means anyone on the team can carry out a recovery. Keep a copy somewhere that doesn’t depend on the system it describes, and use it for every test so it stays accurate.

Watch for silent failures

Make sure a failed backup sends an alert to someone who’ll act on it, and check the backup reports weekly rather than assuming no news is good news. An alert that goes to a shared mailbox nobody reads is the same as no alert.

If you’d like an outside check of your backup and recovery setup, tell us what you have and we’ll test a restore with you. Backups and tested restores are part of our cloud and infrastructure work.

If you’d like a hand with this, send us a few lines about your setup.

Let’s talk

Send an enquiry