Schema Design: Embedding vs Referencing

The single biggest modeling decision in MongoDB: should related data live inside the same document (embedding), or in a separate collection linked by an id (referencing)?

Embedding: keep related data together

A blog post and its handful of comments are read together almost every time the post is viewed, so embedding the comments directly inside the post document avoids a second query entirely:

JS an embedded document

{
  _id: ObjectId("..."),
  title: "Hello, MongoDB!",
  comments: [
    { author: "Sam", text: "Great post!" },
    { author: "Noah", text: "Thanks for this." }
  ]
}

One findOne() call returns the post and every comment on it — no join, no second round trip.

Referencing: keep related data separate

A user and every order they've ever placed is a different story — there could be thousands of orders, they're usually queried independently of the user record, and you rarely need "the user and all their orders" in one shot:

JS referenced documents

// users collection
{ _id: ObjectId("u1"), name: "Priya" }

// orders collection
{ _id: ObjectId("o1"), userId: ObjectId("u1"), total: 42.50 }
{ _id: ObjectId("o2"), userId: ObjectId("u1"), total: 18.00 }

Each order stores a userId pointing back to the user — the same idea as a foreign key in a SQL table. Fetching a user's orders is a separate query: db.orders.find({ userId: ObjectId("u1") }).

Choosing between them

Embed when the related data is bounded, usually small, and almost always read together with its parent. Reference when the related data can grow without bound, is queried on its own, or is shared across multiple parent documents.

The 16MB ceiling: every MongoDB document has a hard maximum size of 16MB. An embedded array that grows without bound — comments on a post that goes viral, log entries appended forever — can eventually hit that limit and start failing inserts. That's a concrete, practical reason to reference instead of embed once a one-to-many relationship doesn't have a natural upper bound.

PreviousThe Aggregation Pipeline NextUsing a Driver