Putting It All Together: Choosing the Right Pattern

The Decision Framework

Advanced DynamoDB modeling is not about memorizing patterns — it is about matching the right pattern to the right problem. Here is a practical decision guide:

Use Adjacency Lists when:
- You have Many-to-Many relationships (social graphs, permissions, enrollments).
- You need to traverse relationships in both directions.
- You want to fetch an entity and its relationships in a single query.

Use Inverted Index (GSI with flipped keys) when:
- You need to traverse a relationship in the reverse direction.
- The forward direction is served by the main table's PK/SK.
- Example: "Who follows Alice?" when the main table answers "Who does Alice follow?"

Use Write Sharding when:
- A single Partition Key receives more than ~800 writes/second (leave headroom below the 1,000 WCU limit).
- The hot key is predictable (a known popular item, a leaderboard entry, a live event counter).
- You can tolerate client-side aggregation on reads.

Use Sparse Indexes when:
- You need to query a small, dynamic subset of items (open orders, pending jobs, flagged content).
- The subset membership changes over time (items enter and leave the subset).
- A full table scan or a fat GSI would be prohibitively expensive.

The Patterns Are Composable

Real-world DynamoDB schemas often combine multiple patterns. A social platform might use:

• Adjacency lists for the follower graph.
• Inverted index to answer "who follows me?".
• Write sharding for celebrity accounts with millions of followers writing simultaneously.
• Sparse indexes to surface only accounts currently flagged for review.

The skill is recognizing which problem each pattern solves and applying them in combination without creating conflicting access patterns.