A Quarterly Restore Drill for Your Business App Database
Short answer: A backup counts only once you have restored it somewhere safe and checked the data. Once a quarter, restore your latest backup into a separate scratch environment, run a handful of checks against known records, time how long it took, and write the result down. The drill finds the usual surprises, such as expired credentials, missing uploaded files or a restore that takes six hours when you assumed one, while there is still no emergency.
What the drill answers
Two numbers matter, and owners rarely know either one.
- How much work could we lose? This is the gap between the last good backup and the moment of failure. Nightly backups mean up to a day. Some databases can also keep a continuous log of changes so you can restore to a chosen minute; the PostgreSQL continuous archiving documentation describes that approach for its own engine, and managed hosts often offer it as a setting.
- How long are we offline while restoring? This includes finding the backup, building a clean server, loading data, reconnecting the app and checking it.
Pick targets you can live with. A bakery's ordering tool might tolerate losing a few hours but not a full day. Then let the drill tell you whether reality matches.
Before you start
- [ ] Agree a date and put it in the calendar for every quarter, for example the first Thursday after quarter end.
- [ ] Name who runs it and who signs it off. Different people is better.
- [ ] Decide the scratch environment: a temporary database server or a staging copy, never the live system.
- [ ] Confirm the scratch environment is as locked down as production, since it will hold real customer data. Delete it afterwards.
- [ ] Tell staff the drill is happening so nobody mistakes the scratch copy for the real app.
- [ ] Note the start time. You are measuring.
The drill itself
Locate the backup without help from memory
Find the latest backup using only your written runbook. If the person who set it up is on holiday and the steps live in their head, that is finding number one. Check where the backup is stored, how old it is, and whether it is in a different place from the live data. A backup in the same account and region as the database shares its fate in many failure scenarios.
Restore into the scratch environment
Run the restore the way you would in an emergency. Record any step that needed a password, key or permission you did not have to hand. Expired access tokens and rotated keys are common blockers.
Run the check list
- [ ] Row counts for the five most important tables are close to the live counts at the backup time.
- [ ] A named customer record you know well is present and correct, including recent notes.
- [ ] The newest order or booking in the backup matches the time you expected.
- [ ] Users can sign in on the scratch copy with a test account, which proves authentication data came across.
- [ ] Attached files, photos and PDFs open. These are often stored outside the database and need their own backup.
- [ ] Scheduled jobs and emails are switched off in the scratch copy so it does not message real customers.
- [ ] Reports run and totals look plausible against a known month.
Stop the clock and write it up
Record total elapsed time and each stage: locate, restore, verify. Add anything that surprised you. Then compare with the target you set.
A one-page drill record
| Item | Example entry |
|---|---|
| Date and people | 2 October, run by Aroha, signed off by Tom |
| Backup used | Nightly, 31 hours old (an illustration) |
| Time to restore | 1 hour 40 minutes |
| Time to verify | 35 minutes |
| Gaps found | Uploaded images not in backup |
| Fixes and owners | Add file storage to backup, Tom, by end of month |
| Next drill | 8 January |
The numbers above are examples only. Yours will differ with data size, host and plan.
Typical findings and what to do
- Files missing. Add the file storage bucket to your backup plan and test it too.
- Restore is much slower than expected. Ask the host about faster tiers, trim old data, or plan a staged restore with the essential tables first.
- Nobody can find the encryption key. Store recovery keys in a password manager with two trusted holders. A backup you cannot decrypt is no backup.
- Backups silently stopped. Add an alert for "no new backup in 26 hours" and check the alert fires.
- Environment settings are undocumented. Write down every setting and secret name the app needs, without the secret values, so a rebuild is not archaeology.
Beyond the database
A real recovery needs the application code, configuration, domain and DNS access, email settings and payment provider keys. Once a year, widen the drill to a full rebuild on a spare account, even if it is rough. Your host's help centre will list which parts of your plan are covered, so read it as of November 2026 rather than assuming.
FAQ
Is a quarterly drill too often for a tiny app?
Twice a year is fine for a low-change tool. Run it extra after changing host, plan, database version or backup method.
Can the hosting provider do the drill for us?
They can restore, but you need to verify the data is right. Ask them to restore to a separate instance and do the checks yourself.
Do we need a drill if the host says backups are automatic?
Automatic means created, not proven usable. The drill checks your side too: access, keys, files and timing.
How long should drill records be kept?
Keep a few years. They help in audits, insurance questions and handovers between developers.
If you would like a backup plan and a first restore drill set up for your app, get in touch with us.