Course contentsShow
← Computer Science
Lesson 1724 of 2,87239. NoSQL and Modern DatabasesFree lesson

Embedding vs Referencing

Choosing between embedding related data in a single document versus storing references to separate documents.

Embedding vs Referencing

You'll learn: How to decide whether to nest related data inside a document or link to separate documents in NoSQL databases.

The Core Decision

In document databases, you face a fundamental modeling choice: should you embed related data directly within a parent document, or should you store it in separate documents and connect them with references (like foreign keys in SQL)?

Embedding: Keeping It Together

Embedding means nesting related data structures inside a single document. Think of it like keeping all your travel information—flight details, hotel reservation, and car rental—in one folder.

Advantages:

  • Faster reads: One query retrieves everything
  • Atomic operations: Update all related data in one transaction
  • Data locality: Everything lives together on disk

When to embed:

  • Data that's always accessed together
  • One-to-few relationships (not thousands of sub-items)
  • Data that rarely changes independently

Referencing: Keeping It Separate

Referencing stores related data in different documents and links them using IDs—like having separate files for flight, hotel, and car rental, each with a booking confirmation number.

Advantages:

  • Flexibility: Each piece can be updated independently
  • Avoids duplication: Shared data exists once
  • Handles large relationships: No document size limits

When to reference:

  • Many-to-many relationships
  • Data accessed independently
  • Large or unbounded arrays
  • Frequently updated data that's shared across documents

The Hybrid Approach

You can combine both strategies—embed frequently accessed data while referencing the complete dataset elsewhere.

Key Takeaway: Embedding optimizes for read performance and data that belongs together; referencing provides flexibility and prevents duplication—choose based on your access patterns and data relationships.