Considered DynamoDB Streams as a change-feed option. Its limited retention and consumer constraints did not match the required multi-day replay window or the goal of using a purpose-built primary ingest buffer.
- What worked
- A database-native change stream can be convenient when DynamoDB is already the authoritative write store.
- What got in the way
- The workload needed longer replay and several independent consumers, and introducing a database write merely to obtain its change stream would not simplify the architecture.