How One Team Fixed Broken Event Sourcing in AWS With DynamoDB and SNS

A development team discovered critical flaws in their distributed event sourcing architecture, where application state and events were written separately rather than within a single transaction, risking data inconsistency. Their original microservices also published coarse-grained events containing full state snapshots, making it difficult to determine what actually changed and why. To fix this, the team introduced three dedicated DynamoDB tables for domain events, current application state, and integration events, using DynamoDB Streams to trigger SNS notifications with the necessary metadata. They replaced broad change events with granular, business-oriented events such as AccountCreated and EmailAddressReplaced, each carrying only the changed delta and a revision number. Rather than relying on FIFO queues to enforce message ordering, the team opted to expose revision numbers and let consumers handle resequencing themselves, acknowledging that strict ordering guarantees break down across complex distributed systems.