Cloud disaster recovery: how to design recovery after an outage
Cloud disaster recovery starts with recovery objectives per application and ends with tests that show the recovery really works.
Cloud disaster recovery is the design by which systems are restored after a major outage within an agreed time and with an agreed amount of data loss. You determine the RTO and RPO per application, choose a recovery strategy and test the plan regularly.
Who is it for: Architects, operations teams and risk managers who need to organise recovery after the loss of a site, service or region.
What are RTO and RPO?
RTO stands for recovery time objective: the time within which a service must be available again. RPO stands for recovery point objective: how much data loss, expressed in time, is acceptable. An RPO of one hour means you can lose at most the last hour of work.
These objectives belong to business impact, not to the technology. Discuss them with the owners of the business processes, and support them with the consequences of an outage. Stricter objectives call for more design, more capacity and more management.
Which recovery strategies are there?
The strategy varies in speed and complexity. Choose per application what suits the objectives and the effort involved.
- Backup and restore: data is restored into a new environment; relatively simple, but slower.
- Pilot light: a minimal environment is ready and is scaled up in a disaster.
- Warm standby: a scaled-down, running copy takes over and scales up as needed.
- Active-active: several sites run at the same time; fastest recovery, highest complexity.
Where does it go wrong in practice?
Plans rarely fail because of the chosen pattern, more often because of forgotten dependencies. Think of identity services, DNS, certificates, secrets, licences and links with external parties. If those sit in the same region or under the same administrator as the application, the recovery fails too.
Also take cyber incidents into account. With ransomware, copies must be untouched, and you must have a clean environment to restore into.
How do you test the recovery plan?
A plan that has not been tested gives a false sense of security. Test at several levels and record the outcomes.
- First check that backups can be read and restored.
- Carry out a tabletop exercise with all the teams and roles involved.
- Test the technical recovery of an individual application in an isolated environment.
- Practise a full scenario, including communication and return to the normal situation.
- Adjust the plan based on findings and schedule the next test.
Strategy and trade-offs
| Strategy | Recovery speed | Complexity and management |
|---|---|---|
| Backup and restore | Longer, depending on the amount of data | Low; mainly good backups and procedures |
| Pilot light | Medium | Moderate; core is ready, the rest is built up |
| Warm standby | Short | Higher; keeping two environments up to date |
| Active-active | Very short | Highest; requires data synchronisation and design for consistency |
Who is it not ideal for?
- Applications with low business impact, for which backup and restore is usually sufficient and expensive recovery strategies are not needed.
- Teams that confuse disaster recovery with high availability; the latter protects against small outages, recovery against large ones.
- Organisations without capacity for tests, because an untested plan often does not work in a real crisis.
Frequently asked questions
Is high availability the same as disaster recovery?
No. High availability absorbs the failure of components. Disaster recovery covers recovery after the failure of a larger unit, such as a site or region, or after an incident.
How often should you test?
That depends on the business impact and on changes to the system. Test at least after major changes and according to a fixed schedule.
Who decides that the recovery plan is activated?
Record a clear decision-maker and a deputy, with criteria. That way no time is lost on discussion during an incident.
