Storage
Amazon S3
Amazon S3 stores objects with massive durability, flexible lifecycle, event, replication, and security controls. It is the usual system of record for GenAI documents, datasets, model artifacts, logs, and evaluation evidence.
Key points
- Versioning preserves prior object versions and supports recovery
- Bucket policies, access points, and Block Public Access form layered access controls
- Event notifications are asynchronous and can be duplicated
When to use it
- Store a versioned RAG corpus and ingestion manifests
- Retain immutable evaluation inputs and generated reports
Exam tips
- Choose S3 for object storage, not low-latency file locking or relational queries
- Record object version IDs in lineage and use idempotent event consumers