A successful backup job is only the first check

Third Level IT · IT field notes

Start with the business service

“We back up the server” is not a complete recovery objective. Identify what people need to use: a shared folder, an accounting application, a directory service or a complete virtual machine. Each has different dependencies and a different test.

Agree on an acceptable amount of data loss and an acceptable period of interruption. These objectives guide the test; they are not results until measured.

Separate three levels of evidence

• Job evidence: the backup process finished and the repository contains the expected recovery point.
• Readback evidence: selected data or configuration can be retrieved and decrypted.
• Service recovery evidence: the restored workload starts, its dependencies work, and an authorized user can perform the required task.

All three are useful. None should be reported as another. Restoring one file does not establish that a virtual machine will boot or that its application will work.

Design an isolated test

A restored system can conflict with production if it brings back the same network address or starts scheduled jobs. Plan the recovery environment before powering it on. Identify which connections should be available and which should remain isolated. For sensitive data, define who may access the restored copy and when it will be removed.

Check the dependencies

Applications may depend on DNS, certificates, service accounts, licensing, databases or an external provider. Record the dependencies the test actually exercised. If a dependency was simulated or unavailable, say so. Measure elapsed recovery time only for the steps that ran.

Keep a short, useful test record

• Recovery point and target system.
• What was restored and which keys or accounts were required.
• Start and finish times for the tested steps.
• Application and user checks that passed or failed.
• Unverified dependencies, corrective work and next test date.

Our own example

In a TLI internal Windows server engagement, an encrypted backup completed and configuration readback succeeded. A full machine restore was not tested. Our engineering practice note preserves that limit rather than describing the system as fully recovery-tested.

Use the same standard when reviewing a provider’s report: ask exactly what was recovered, not only whether the backup status is green.