Document vs Relational Thinking
MongoDB's document model encourages embedding related data, which reduces joins and improves read performance. However, embedding everything leads to document bloat and write amplification. The key is knowing when to embed and when to reference.
The Embedded Document Pattern
Embed when: the embedded data is only accessed via the parent, the cardinality is one-to-few, and the data doesn't need independent querying.
// Embed address inside user — rarely queried independently
{
name: "Alice",
address: { street: "123 Main", city: "NYC" }
}The Reference Pattern
Reference when: data is accessed independently, the cardinality is one-to-many, or multiple documents reference the same entity. This is what we use for posts → categories and posts → author.
Indexing Strategy
Every query should hit an index. Run db.collection.explain("executionStats") to verify. Key indexes for a blog:
- Unique index on
slug - Compound index on
{ status: 1, publishedAt: -1 }for listing published posts - Text index on
{ title: "text", content: "text" }for search
The Bucket Pattern
For time-series data (like view counts per hour), bucket documents instead of one document per event. This reduces document count dramatically and improves aggregation performance.
Schema Versioning
Add a schemaVersion field to every document. When you need to migrate, you can do it lazily on read — transform the old shape to the new one, save it back, and never need a big-bang migration.