tezvyn:

Multi-region active-passive DR with Aurora

AI-drafted, machine-checkedSource: interviewadvanced
WHAT IT TESTS

designing cross-region DR with clear RPO/RTO.

OUTLINE

async global replication to a passive region, promote and repoint traffic on failover, and fence the old primary to prevent split-brain.

WHAT THIS TESTS This evaluates whether you can architect region-level disaster recovery, set realistic recovery objectives, and reason carefully about the dangerous split-brain edge case.

REPLICATION MECHANISM Use Aurora Global Database: a primary region serves all reads and writes, and a secondary region holds a read-only replica fed by storage-level asynchronous replication, typically with sub-second lag. Because replication is asynchronous, the recovery point objective is non-zero; a regional loss can drop the last few in-flight transactions. Document the expected RPO and RTO explicitly.

FAILOVER PROCESS On a primary-region outage, detect it via health checks, then promote the secondary's replica to a standalone writer. Repoint the application using a low-TTL DNS endpoint, a global database failover, or a routing layer. Warm connection pools and verify the application is healthy. This is active-passive, so the secondary takes no writes until promotion.

SPLIT-BRAIN AND CONFLICTS The key risk is two writable primaries if the old region recovers and resumes accepting writes. Prevent this by fencing: revoke the old primary's write capability, demote it, or block its endpoint before or during promotion, ensuring only one writer exists. On recovery, do not blindly re-replicate; first reconcile any writes that committed in the old primary but never replicated, using application logic, idempotency keys, or manual reconciliation, then rebuild the old region as the new passive replica.

LIKELY FOLLOW-UPS What RPO does async replication realistically give. How do you automate failover without flapping. How do you detect and reconcile divergent writes. Why not active-active.

ONE CONCRETE EXAMPLE Primary in Singapore, passive in Tokyo via Aurora Global Database. Singapore fails; you promote Tokyo, fence Singapore's writer, and flip the app's Route 53 record. When Singapore returns, you reconcile a handful of unreplicated orders using idempotency keys, then rebuild it as the new Tokyo-following replica.

Read the original → docs.aws.amazon.com

Get five bites like this every day.

Tezvyn delivers a daily feed of 60-second tech bites with quizzes to lock in what you learn.