The dashboard shows a green tick: “backup completed”. The database is backed up every night and there is plenty of storage space. Everything seems under control until the only question that matters arrives: how long would it take us to get back up and running?
If nobody has restored the system in an isolated environment and checked the result, the answer is guesswork.
A backup is one ingredient. Recovery involves people, access, configuration, data, applications and decisions. This guide explains how to move from “we have backups” to “we know what we can recover, how long it will take and the maximum data loss involved”.
Backups and Recovery Answer Different Questions
A backup asks: “have we saved the data?” Recovery asks: “can we rebuild a usable service with that data?”
There is quite a lot of work between the two:
- Creating the necessary infrastructure.
- Deploying a compatible version of the application.
- Restoring the database and files.
- Recovering configuration and secrets without exposing them.
- Reconnecting integrations and DNS.
- Verifying that important business operations work.
A database restored on an empty server does not yet restore the service to your customers.
Before Touching a Tool: Define RPO and RTO
Two acronyms help turn a technical conversation into a business decision.
RPO: How Much Information Can You Afford to Lose?
The Recovery Point Objective defines the point in time to which you need to return. If your RPO is one hour, the system and backups must allow you to recover with no more than that window of data loss in the anticipated scenario.
An informational website is not the same as a system that records orders or payments. Backup frequency should derive from that tolerance, not from a generic template.
RTO: How Long Can You Be Down?
The Recovery Time Objective is the target time for restoring the service after an interruption. It includes detecting the problem, organising the response, restoring the system and validating it. It is not just the time needed to copy a database.
These objectives should be documented for critical services and accepted by the business. “As soon as possible” does not help you decide architecture, budget or priorities.
What to Save Besides the Database
The database usually gets all the attention, but a real application has more components:
- Files and objects: documents, images, attachments and exports.
- Code and deployable versions: the repository, artefacts and container images needed to rebuild the service.
- Configuration and infrastructure: definitions of networks, services, scheduled jobs and deployment parameters.
- Secrets: credentials and certificates, protected and backed up with specific controls, separately from the code.
- Managed services: queues, storage, search engines and other components with their own backup mechanisms.
- Operational knowledge: contacts, external dependencies and recovery steps.
The inventory should also include what you will not back up because you can regenerate it. This distinction prevents you from paying to store temporary data while forgetting something irreplaceable.
The 3-2-1 Rule Is a Good Starting Point
The 3-2-1 strategy recommends keeping three copies of your data, using two different media or systems, and storing one copy outside the primary environment. It is a useful reference, not an automatic security certification.
Two further questions are worth asking:
- Can a failure or an attacker delete every copy using the same credentials? If the answer is yes, there is no real separation.
- Are the copies protected against modification or deletion for the required period? Immutability or an offline copy may help, depending on the risk.
Separating accounts, credentials and permissions is especially important in the event of ransomware or an administrative error. A backup that shares the same destination as production does not provide a sufficient boundary.
How to Run a Useful Restore Test
A useful test does not consist of opening a dashboard and checking that a file exists. It consists of recovering a service and measuring the result.
1. Define the Scenario
Choose a specific situation: loss of the database, deletion of documents or complete unavailability of the environment. Do not combine every possible disaster in the first test.
Record which service will be recovered, which backup will be used and what result is considered acceptable.
2. Use an Isolated Environment
Restore to separate infrastructure so that you do not overwrite production or send real emails, charges or messages. Block or simulate external integrations that could produce side effects.
Isolation also makes it possible to test more frequently without turning the exercise into another incident.
3. Time the Entire Process
Start when recovery begins, not when the copy finally starts. Record the time spent locating access, preparing infrastructure, restoring data and validating the result.
If someone has to spend forty minutes looking for a password, that time is part of the actual RTO.
4. Verify Data and Business Operations
Technical checks are necessary: tables available, files readable and application deployed. But functional tests are also needed:
- Log in with a test user.
- Query records from before and after the chosen recovery point.
- Create a test operation end to end.
- Check documents, permissions and simulated integrations.
- Make sure automated jobs do not run twice.
Recovery does not end when the server responds; it ends when the business can operate with consistent data.
5. Compare the Result with RPO and RTO
Document the point recovered, the time taken and any differences from the objectives. If they are not met, there are three basic levers: change the procedure, change the technology or revisit the objective with the business.
Every deviation needs an owner and a deadline. A report without actions merely records the risk.
Failures That Successful Backups Do Not Reveal
These issues tend to emerge during a restore, not on the daily dashboard:
- The backup includes the database but not documents stored separately.
- The current application is incompatible with the old schema and a controlled migration is missing.
- Nobody knows which version of the code corresponds to the backup.
- Secrets have expired or only one person knows them.
- The restore works, but the data volume makes it take much longer than expected.
- Scheduled jobs run again and duplicate operations.
- The backup is encrypted and the recovery key is unavailable.
- Storage has been rejecting data due to lack of space without anybody reviewing the alert.
A successful backup run does not necessarily detect any of these problems.
Automate Backups, but Assign Recovery Responsibilities
Backups should run and be monitored automatically. Recovery also needs clear owners:
- Who declares the incident and coordinates the response.
- Who can authorise a restore.
- Who performs the technical work.
- Who validates the data and business processes.
- Who communicates the status to affected customers or teams.
Names and deputies must be available even if the main system is down. The only runbook stored inside the failed service is not much help.
A Minimum Plan That Fits on One Page
For each critical service, document:
| Element | What it should answer |
|---|---|
| Scope | Which application, data and dependencies it includes |
| RPO | What maximum data loss has been agreed |
| RTO | What recovery time is being targeted |
| Backups | Where they are, who administers them and how long they are retained |
| Access | Where it is stored and who can recover it |
| Procedure | Where the up-to-date runbook is located |
| Validation | Which tests demonstrate that the business works |
| Responsibilities | Who executes, validates and communicates |
| Evidence | When it was tested and which fixes remain open |
Test frequency should reflect criticality and the pace of change. A system that changes every week needs its recovery reviewed more closely than a stable one; in both cases, an old test loses value as evidence.
In Summary
The success of a backup is not measured by the file generated, but by the ability to restore a useful service within the agreed limits.
Define RPO and RTO, include every part of the system, separate backups from production and rehearse recovery in an isolated environment. Measure, validate business operations and fix whatever emerges. Peace of mind does not come from a green tick: it comes from having made recovery work.
If you have backups but have never timed a complete restore, we can help you review coverage and prepare an initial test without putting production at risk. Let’s talk about your infrastructure.
Keep reading: Observability for Small Teams, to detect failures earlier and know when to activate the recovery plan.