Consistency and the replicated journal

How AWS S3 is built

Topic 5 of 13

Published Jan 21, 2026 · Clip 20:15–26:04 (5:49)

The guest explains that eventual consistency primarily optimized for availability. She details the indexing subsystem that holds object metadata and serves every API call, its quorum-based replication across availability zones, and how spreading data avoids correlated fault domains. For strong consistency, the team invented a replicated journal—a distributed data structure chaining nodes sequentially so writes flow through with sequence numbers—enabling strong consistency without compromising availability.