MongoDB Schema Design: Embedded vs Referenced Documents

Introduction

MongoDB is “schemaless,” but production services benefit from an explicit schema. Embedded vs referenced design is the core fork: store child data inside the parent document (embedded) or in another collection linked by id (referenced). Embedded documents can end reads in one round trip; references control growth and reuse of documents. This article replaces gut feel with measurable criteria: size, frequency, consistency.

After reading this post

• You can apply a checklist for embedded vs referenced designs
• You get hints for 1:N, N:M, i18n/versioning variants
• You understand the 16MB document limit and index interactions

---

Table of contents

1. Concepts
2. Hands-on implementation
3. Advanced usage
4. Performance comparison
5. Real-world cases
6. Troubleshooting
7. Conclusion

---

Concepts

Embedded

Related data lives in one BSON document as nested arrays or sub-documents.

// User + addresses (embedded example)
{
  _id: ObjectId("..."),
  name: "Kim",
  addresses: [
    { label: "home", city: "Seoul", zip: "03000" },
    { label: "work", city: "Seongnam", zip: "13000" }
  ]
}

Pros: If you almost always read together, one query finishes the job. Updates in the same document can leverage atomic updates.

Referenced

Separate collections linked by ObjectId (or similar).

// users
{ _id: ObjectId("u1"), name: "Kim" }
// addresses
{ _id: ObjectId("a1"), userId: ObjectId("u1"), city: "Seoul" }

Pros: Child collections can grow large without bloating the parent. Fits models where many parents share the same child.

BSON document size

A single document is capped at 16MB. Data that could grow without bound should start as separate collections + indexes.

Hands-on implementation

Decision checklist

Question	Favors embedded	Favors referenced
Read together?	Almost always	Rarely or partial reads
Updated together?	Often at once	Frequently independent
Bounded child count?	Small (e.g. a few addresses)	Thousands/unbounded
Shared child across parents?	Rarely	Often
Strong cross-doc integrity?	If it fits one document	Multiple collections + app logic/transactions

Pattern 1: 1:few — embedded

If comments are always shown with a post and counts are limited (or page-sized), consider embedding.

Pattern 2: 1:many (large) — referenced + indexes

// posts collection
db.posts.createIndex({ slug: 1 }, { unique: true });
// comments collection
db.comments.createIndex({ postId: 1, createdAt: -1 });

db.comments.find({ postId: postObjectId }).sort({ createdAt: -1 }).limit(50);

Pattern 3: Middle ground — buckets

For time-series logs, group readings per day to reduce document count.

{
  sensorId: "s1",
  day: ISODate("2026-03-30T00:00:00Z"),
  readings: [
    { t: ISODate("2026-03-30T00:05:00Z"), v: 23.1 },
    // ....batched for the day (design an upper bound)
  ]
}

Embedded vs referenced often lands on embedded-but-bucketed compromises.

$lookup (join)

For occasional joins on referenced models, aggregation $lookup works—but for hot paths, prefer two reads in application code or caching.

Advanced usage

• i18n fields: { title: { ko: "...", en: "..." } } vs locale-split documents—match your translation workflow.
• Partial updates: Use array filters and positional operators to reduce write conflicts on embedded arrays.
• Schema validation: Enforce required fields/types with validator and JSON Schema for operational safety.

---

Performance comparison

Aspect	Embedded	Referenced
Read latency (common path)	Single find	find + find or $lookup
Write contention	Hotspot if many writers to one doc	Can spread writes
Indexes	Nested field indexes—more complex	Per-collection, often simpler
Consistency	Single-doc atomic updates	Application logic + transactions

---

Real-world cases

• E-commerce orders: Header + dozens of line items often embedded; product master referenced.
• Social feeds: Post referenced; like counts denormalized with eventual consistency via events.
• B2B multi-tenant: Include tenantId in every query with a leading index.

---

Troubleshooting

Symptom	Cause / fix
Document near 16MB	Split arrays, archive collections, buckets
Slow update on large embedded arrays	Split documents or move to references
Broken referential integrity	App validation + transactions if needed; explicit delete policies (soft delete)
Index used but still slow	Working set size, projection minimization, covering indexes

---

Conclusion

MongoDB embedded vs referenced design is not “anything goes because NoSQL”—it is locking read patterns and growth boundaries into code. Compare with relational transaction/join trade-offs using the PostgreSQL vs MySQL guide, apply this checklist first, then validate with load tests.

---

자주 묻는 질문 (FAQ)

Q. 이 내용을 실무에서 언제 쓰나요?

A. Choose embedded vs referenced collections using document size, read/write patterns, 1:N growth, and consistency—plus buc… 실무에서는 위 본문의 예제와 선택 가이드를 참고해 적용하면 됩니다.

Q. 선행으로 읽으면 좋은 글은?

A. 각 글 하단의 이전 글 또는 관련 글 링크를 따라가면 순서대로 배울 수 있습니다. C++ 시리즈 목차에서 전체 흐름을 확인할 수 있습니다.

Q. 더 깊이 공부하려면?

A. cppreference와 해당 라이브러리 공식 문서를 참고하세요. 글 말미의 참고 자료 링크도 활용하면 좋습니다.

---

같이 보면 좋은 글 (내부 링크)

이 주제와 연결되는 다른 글입니다.

• PostgreSQL vs MySQL 차이와 선택 가이드 | 스키마·트랜잭션·운영
• TypeORM vs Prisma 비교 | 타입 안전성·마이그레이션·쿼리·성능 실전 가이드
• Node.js 데이터베이스 연동 | MongoDB, PostgreSQL, MySQL

---

이 글에서 다루는 키워드 (관련 검색어)

MongoDB, NoSQL, Schema Design, Embedded, Referenced, Data Modeling 등으로 검색하시면 이 글이 도움이 됩니다.