Database
Amazon DynamoDB
Amazon DynamoDB is a serverless key-value and document database with single-digit-millisecond performance at scale. GenAI applications use it for idempotency, job state, rate counters, conversation pointers, and carefully partitioned caches.
Key points
- Partition-key design determines distribution and query access
- Conditional writes support atomic idempotency and optimistic concurrency
- TTL removes expired items asynchronously rather than at an exact instant
When to use it
- Deduplicate retried generation requests
- Store per-session state and authorized response-cache metadata
Exam tips
- Use DynamoDB for known key-based access; use a search engine for ad hoc semantic queries
- Include tenant, model, prompt, corpus version, and authorization context in cache boundaries