Pemodelan Dokumen dengan Beanie ODM

Beanie is an asynchronous ODM (Object-Document Mapper) for MongoDB built directly on top of Pydantic and the async motor driver. In a FastAPI backend it gives you typed, validated documents that feel just like the Pydantic models you already use for request and response bodies.

• Each document class maps to one MongoDB collection.
• Each instance maps to one document (a JSON-like record).
• Validation, serialization, and JSON Schema come for free from Pydantic v2.

Because everything is async, Beanie pairs naturally with FastAPI's async route handlers and avoids blocking the event loop on database I/O.

A Beanie model subclasses Document instead of Pydantic's BaseModel. Fields are declared with normal type hints, and Beanie automatically gives every document an id field backed by MongoDB's _id (an ObjectId).

• Required fields have no default; optional fields use Optional[...] or a default value.
• The collection name is derived from the class name unless you override it.

Below, Product becomes a collection of product documents.

Beanie reads optional configuration from an inner class Settings. The most common option is name, which sets the MongoDB collection name explicitly instead of relying on the class name.

• name — the collection name.
• use_state_management — track changed fields for partial saves.
• validate_on_save — re-run validation when saving an existing document.

Pinning the collection name keeps your schema stable even if you later rename the Python class.

Before any document can talk to MongoDB you must call init_beanie once, passing your motor database and the list of document models. In FastAPI this belongs in the lifespan handler so it runs at startup.

• AsyncIOMotorClient creates the async connection.
• document_models registers every model so Beanie can build indexes and queries.

This is framework/server wiring — it needs a live MongoDB, so treat it as a setup pattern, not a standalone script.

Because a Document is a Pydantic model, every validation tool you know still works: Field constraints, custom validators, and rich types like EmailStr or HttpUrl.

• Field(gt=0) rejects non-positive prices.
• Field(min_length=...) enforces string length.
• Validation runs when the object is constructed, so bad data never reaches MongoDB.

This snippet is pure Pydantic-style validation and runs on its own.

Indexes make queries fast and can enforce uniqueness. Beanie offers two styles. The simplest is the Indexed wrapper applied to a field's type, which creates a single-field index.

• Indexed(str, unique=True) builds a unique index — perfect for an email or SKU.
• Beanie creates the index automatically during init_beanie.

Use unique indexes to push integrity rules down into the database rather than relying only on app checks.

For multi-field or advanced indexes, declare them in Settings.indexes using PyMongo's IndexModel. This is how you build compound indexes, control sort direction, or add a TTL.

• List the fields with directions: pymongo.ASCENDING / DESCENDING.
• Pass unique=True or expireAfterSeconds=... through the IndexModel.

Order matters: a compound index on (category, price) optimizes queries that filter by category and then sort by price.

MongoDB stores nested objects inside a single document. In Beanie an embedded structure is just a plain Pydantic BaseModel used as a field type — it is not a separate collection and has no id.

• Embed when the nested data is owned by the parent and always read together (an address inside a user).
• The whole structure is validated and serialized as one document.

Here Address is embedded inside the User document.

A document field can hold a list of embedded models, which is ideal for one-to-many data that belongs entirely to the parent — think order line items or comments.

• list[OrderItem] validates every element on construction.
• Each item is serialized inline, so reading the order needs no extra query.

Prefer embedding lists when the collection is bounded and read with the parent; use references when it grows unbounded.

When related data lives in its own collection and is shared or large, use a reference instead of embedding. Beanie's Link[OtherDocument] stores a pointer (a DBRef) and can fetch the linked document on demand.

• Link[Category] keeps the category in its own collection.
• Use fetch_links=True on a query, or await doc.fetch_link(...), to resolve it.

Rule of thumb: embed owned, read-together data; link shared or independently-queried data.

The core design decision in document modeling is whether to embed data or reference it. There is no universal answer — it depends on access patterns and data size.

• Embed when: the child is owned by the parent, always loaded together, and bounded in size (address, order items).
• Reference when: the data is shared across documents, queried on its own, or can grow without limit (categories, authors, audit logs).

MongoDB documents have a 16 MB cap, so unbounded embedded arrays eventually break — another reason to reference large, growing collections.

You learned how to model MongoDB data with Beanie on top of Pydantic:

• Document subclasses map a Python class to a collection; class Settings sets the collection name and options.
• init_beanie must run at startup (in FastAPI's lifespan) with your motor database and document_models.
• Pydantic Field constraints and validators keep bad data out of the database.
• Index with the Indexed wrapper for single fields or Settings.indexes + IndexModel for compound and unique indexes.
• Embed owned, bounded, read-together data as nested BaseModels; reference shared or growing data with Link[...], mindful of the 16 MB document cap.

With these tools you can design typed, validated, query-efficient MongoDB schemas for your FastAPI backend.