CloudGuild · Blog · Cheat sheets · Lessons · Certifications
Amazon DynamoDB — NoSQL at Scale
Serverless key-value/document database with single-digit-millisecond latency at any scale.
What it is
Fully managed NoSQL with predictable low latency, on-demand or provisioned capacity, and no servers.
When I reach for it
- High-scale, simple access patterns: sessions, carts, leaderboards, IoT, user profiles.
- Serverless stacks where a relational engine is overkill.
Key architectural decisions
- Model around access patterns, not normalized tables; choose partition keys that spread load.
- On-demand for spiky/unknown traffic; provisioned + auto scaling for steady, cost-sensitive load.
- Add DAX for microsecond reads on hot items; Streams for change-driven pipelines.
- Global Tables for multi-Region active-active.
Gotchas & exam traps
- Hot partitions throttle — design the key.
- "Single-digit-ms at massive scale, key-value" => DynamoDB; complex joins => relational.
- DAX caches DynamoDB specifically; ElastiCache is generic.
The architect view
DynamoDB rewards up-front access-pattern design and punishes relational habits. When the pattern is simple and the scale is large, it is unbeatable on operations and cost.