This guide draws on the Vitaver Playbook’s practice of identifying the evidence that changes a decision. Ask what the standby can deliver with the primary location unavailable.
Decide how much recovery the business will pay for
Put the service objective on paper before buying a second site. Specify the recovery time objective, or RTO, as the target time to restore an agreed level of service. Define the recovery point objective, or RPO, as the maximum acceptable age of recovered data relative to the interruption. A reporting dashboard may tolerate a long wait. An order service may need to accept new work soon and recover details of transactions that users already submitted. Have the service owner approve both objectives and the minimum capacity needed at the alternate site.
AWS’s recovery strategy guidance describes options including rebuilding from backups, running a minimal footprint that can be expanded, maintaining a warm standby, and serving from more than one site. Use the options to compare cost and complexity. A warm standby can reduce the work required at activation and introduces a second environment to patch, validate and monitor. Serving writes across multiple sites requires an explicit way to handle consistency and conflicts. Use observed recovery performance to decide whether more investment is warranted.
Place the alternate site outside the likely failure
Choose a location based on what could fail together. Consider geographic exposure, utilities, network routes, the supplier’s operating boundaries, and whether local responders can reach the site during the same event. Two geographically distant server rooms can still depend on one carrier, a shared management network, or one team with the only recovery credentials. Record the failure your selection is meant to survive. Check capacity reservations, network connectivity and any approved data-location boundaries before committing to it.
List what the application needs after traffic arrives. Include identity, DNS, certificates, key management, databases, storage, code and package registries, observability, and any service needed to take or confirm an order. Then list what the response team needs to make the switch. AWS’s multi-region dependency guidance discusses common failures across workloads and dependencies. Microsoft’s review of the October 29, 2025 Azure Front Door incident describes connection timeouts and DNS issues across multiple regions. Test the complete route users and operators depend on wherever those functions are hosted.
Decide where writes are allowed
For a service with a single active writer, the recovery procedure needs a reliable way to keep the old primary from accepting writes before the standby is promoted. Verify which copy is eligible for promotion and record the last consistent recovery point. If the primary’s state cannot be confirmed, apply the approved rule for the uncertain state. A temporary read-only mode can give operators time to resolve it. Where writes are allowed at multiple sites by design, test how the application resolves competing updates after connectivity returns.
Use an illustrative test order. Record its reference, whether the customer saw confirmation, what entered the database, and what reached any downstream system. Fail over, then verify which stage survived. Decide who checks transactions that arrived around the switch and who can authorize replay. Before failing back, reconcile changes made in the alternate environment and confirm the next site is safe to accept writes. Give the service owner both an acceptance test and a way to stop the return.
Prove the alternate service at useful capacity
The platform lead should demonstrate an end-to-end transaction with the primary location treated as unavailable. Have a business owner verify the result, including user access, downstream confirmation, data freshness, and an agreed minimum workload. Record when the team declared failover, how long the transaction path took to become usable, and which dependencies still required the primary environment. Confirm that the standby has enough workers and provider quotas for the tested load. Repeat the check after a material dependency or recovery target changes. You can use those observations to decide whether switching providers or building an additional location would address a real gap.