When do you embed data and when do you reference it?

Embedding keeps an aggregate together, so one read returns the whole object and one write is atomic within the document. Referencing stores an id and requires a second lookup or a join, but avoids duplication and unbounded growth.

{
  "_id": "order_1042",
  "customer": { "id": "cus_7", "name": "Asha" },
  "shippingAddress": {
    "line1": "12 MG Road",
    "city": "Bengaluru",
    "postalCode": "560001"
  },
  "items": [
    { "sku": "sku_1", "qty": 2 },
    { "sku": "sku_9", "qty": 1 }
  ]
}

Rules of thumb

• One-to-one or one-to-few with bounded size: embed
• One-to-many with a known small maximum: embed if the parent is always read with the children
• One-to-many with unbounded growth: reference or use a child collection
• Many-to-many: reference, because duplication would explode
• Data shared across many parents and updated often: reference

Consider access frequency as well as cardinality. Embedding 50 line items is fine if the order page always reads them. Embedding 50 items that are only needed on a rare detail view wastes space on every list query.