mongodb-schema-design

Embed when data is always accessed together, such as one-to-one or bounded one-to-few relationships like user addresses. Reference when data is accessed independently, relationships are many-to-many, or arrays can grow without bound, such as posts and comments.

Avoid converting tables one-to-one into collections. Identify tables that are always joined together and denormalize them into single documents, keeping separate only what is accessed independently. This enables single-query reads and atomic updates.

The 16MB BSON limit is typically hit by unbounded arrays, large embedded binaries, or deeply nested objects. Move unbounded data to separate collections with a reference field, and monitor document sizes using the $bsonSize aggregation operator.

Run an aggregation with $indexStats to find indexes with zero access counts, then hide the index with hideIndex, monitor for a full workload cycle, and drop it once confirmed safe. A compound index like {a:1, b:1} makes a single-field index {a:1} redundant.

Add $jsonSchema validation when creating new collections or after discovering inconsistent document structures. Start with validationLevel moderate and validationAction warn on existing collections, then tighten to strict and error once violations are resolved.

Avoid embedding when arrays grow unbounded, child documents are large relative to the parent, data is queried independently of the parent, or child data has a different lifecycle. Use references or patterns like outlier and bucket instead.