A portable returns recovery case
Fictional Marlow Equipment accepts a constant 10 returns per second for this exercise. It uses one provider today and requires a tested recovery path to another. Its supplied objectives are four hours to recover and at most 30 minutes of lost data. A retried return must remain unique. These are case inputs, not measured service performance.
Question: What does an accepted return mean? Commit the business record before acknowledging it. If the response is lost, a stable return identity lets the client retry without creating another return. Choose one authoritative writer during normal operation and recovery.
Question: What do the supplied numbers prove? The supplied rate is constant. An hour contains 36,000 records; 30 minutes contains 18,000. That arithmetic doesn’t establish bytes, export lag or import duration. Measure those and reconciliation time before claiming the objectives are achievable.
Recovery path and promotion evidence
Draw the intake path through the gateway, service and authoritative database, then draw a separate export worker publishing a consistent committed checkpoint, immutable objects and a manifest that identifies what the importer may reconstruct. A recovery importer reads that contract into the alternate database.
Question: Why not write to four providers? The requirement asks for a tested exit path. Four independent writers introduce conflict and reconciliation work that the brief doesn’t justify. Choose a primary provider from operating requirements, then test the recovery adapter; similarly named managed services don’t establish equivalent behavior.
Question: When can the alternate accept writes? Check the dataset first. Verify checksums, schema, source coverage and unique return IDs. Reconcile writes after the export checkpoint, validate the client path and fence the previous writer. A stored object or successful sign-in alone doesn’t prove recovery or permission to decrypt it.
A stored object alone proves no recovery.
Interview and underlying practice
Open the 55-minute cloud architecture round to draw and defend the design. The broader system-design interview area covers other original problems. The round is independent interview practice. It doesn’t run a cross-provider recovery test or reproduce employer questions.
Use the AWS, Azure, Google Cloud and OCI courses for provider-native configuration boundaries. Their local consoles don’t deploy the architecture or prove a recovery time. The AWS recovery planning reference and Azure recovery guidance provide primary reading for a separately tested recovery plan.