MongoDB Data Modeling Notes

Mental Model

MongoDB works best when documents represent natural aggregates that are usually read and written together. The unit of design is not a table row; it is the shape of the application access pattern.

Embed When

• Child data is bounded.
• Parent and child are read together.
• Updates usually touch the same aggregate.
• Strong locality is more useful than independent querying.

Reference When

• Child data grows without bound.
• Many parents share the same entity.
• Independent lifecycle or permissions matter.
• The query path naturally starts from either side.

Indexes

Indexes should match real filters and sort orders. A flexible schema does not remove the need for deliberate query design.

Design Smells

• Giant documents that grow forever.
• Collections with no clear aggregate boundary.
• Cross-collection joins simulated repeatedly in application code.
• No plan for schema evolution.
• No clarity around consistency expectations.