Database Design Mental Models

The Core Question

Data modeling starts with access patterns, consistency needs, and change frequency, not with a favorite database engine.

Ask First

• What are the top read paths?
• What are the top write paths?
• Which operations must be transactional?
• Which fields change frequently?
• Which entities grow without bound?
• What must be queryable by support, analytics, or operations?
• What data must be deleted or anonymized?

Relational Bias

Use a relational model when relationships, constraints, joins, reporting, and transactions are central. Postgres is a strong default for many backend systems because it can stay simple while the product changes.

Document Bias

Use a document model when the aggregate shape is natural, nested data is read together, and schema flexibility is more valuable than cross-document joins.

Design Smells

• A table or collection called data.
• JSON fields used to avoid understanding the domain.
• Missing indexes on primary query paths.
• No migration strategy.
• No ownership of retention or deletion.
• Queries that only work because the dataset is still tiny.