Help for owners
I have backups, but nobody has ever tried restoring one
Short answer
A backup is proven only when someone has restored it somewhere safe and the site or app worked. Ask for a dated restore test that covers the database, uploaded files and settings. Providers can schedule restore tests themselves; AWS Backup, for example, documents automated restore testing.
Does this sound like your situation?
- The host or my developer says backups run every night, but I have never seen one opened.
- Nobody can tell me how long a restore would take.
- The backup covers the server, but I am not sure it includes the database or customer uploads.
- The person who set the backups up has left.
- I would not know whom to call at three in the morning to bring the site back.
Why does this happen?
A backup job succeeding is not the same as a restore succeeding
A job can report success while it saves an empty folder, an old path or a file that is half written. You only find out when you try to use it.
Parts of the system are missing from the copy
Databases, uploaded files, configuration, secrets and the DNS settings are often kept in different places. A copy of one of them does not bring the business back.
Restores need access and permissions nobody checked
AWS documents common restore failures that have nothing to do with the data itself, such as a restore role that cannot be assumed or a missing encryption key permission. The backup exists but cannot be opened.
Confirmed by docs.aws.amazon.com
Nobody owns the restore
Backups are set up once and forgotten. No one has written down the order of steps, the timing or who decides to restore.
What can I do about it today?
- Write down what the business cannot loseOrders, customer records, uploaded documents, the website content. This list tells everyone what a backup must contain.
- Ask four plain questions in writingWhat is backed up, where the copies are stored, who can open them, and the date a restore was last tried.
- Do not delete or overwrite anything to testA test restore must go to a separate, empty place. Never restore over the live site to see what happens.
- Check that the copies are not only on the same serverIf the backup lives on the machine that fails, it fails with it. Ask where else it is kept.
- Ask for a short restore reportA date, what was restored, how long it took and what did not work. If nobody can produce one, you have your answer.
When should I bring someone in?
- No one can say what is covered or where the copies are.
- The business takes money or orders online and an hour of lost data would hurt.
- A test restore failed, or the person who knows the setup is gone.
How do I handle it?
- List every part of the systemDatabase, files, configuration, secrets, domain settings and scheduled jobs, each with a named copy and a place it is stored.
- Restore into an empty copyI bring the latest backup up on a separate machine, never over the live one, and check the app works with real-looking data.
- Write the steps and the time they tookA short runbook anyone technical can follow, with the measured restore time, not a guess.
- Fix what the test exposedMissing folders, wrong permissions, copies kept on the same server. Each gap is closed and the test repeated.
- Repeat on a scheduleA restore test is a recurring check with a date and a result, kept where you can read it.
The price depends on what I find, so I quote in writing after a short discovery call. Tell me what is happening and I reply within 24 hours.
What do owners ask about this?
How often should a restore be tested?
There is no single right interval. Test after any change to the system and on a regular schedule you can keep. AWS Backup lets you set a frequency for automated restore tests if you use it.
Does the host's backup count?
It counts only once you know what it includes and have restored from it. Many provider backups are copies of the whole disk and may not match what you expect for the database or uploads, so confirm each part.
Can I test it myself?
You can ask for the report and check the questions above. The restore itself should be done by someone technical, into a separate place.
Sources
How was this page made?
I drafted this page with AI assistance from the official documents listed under Sources. Found a mistake? Tell me.