api-database-mongodb

MongoDB / Mongoose Patterns

Quick Guide: Use Mongoose as the ODM for MongoDB. Define schemas with automatic TypeScript inference, use lean() for read-only queries, prefer embedding over referencing for co-accessed data, place $match early in aggregation pipelines, and always define indexes to match your query patterns.

---

<critical_requirements>

CRITICAL: Before Using This Skill

All code must follow project conventions in CLAUDE.md (kebab-case, named exports, import ordering, import type, named constants)

(You MUST define Mongoose middleware (pre/post hooks) BEFORE calling model() -- hooks registered after model compilation are silently ignored)

(You MUST pass { session } to EVERY operation inside a transaction -- missing session causes operations to run outside the transaction)

(You MUST use .lean() for read-only queries that send results directly to API responses -- skipping lean wastes 3x memory on hydration overhead)

(You MUST use 127.0.0.1 instead of localhost in connection strings -- Node.js 18+ prefers IPv6 and localhost can cause connection timeouts)

(You MUST NOT use findOneAndUpdate / updateOne and expect save middleware to fire -- only save() and create() trigger document middleware)

</critical_requirements>

---

Auto-detection: MongoDB, Mongoose, mongoose.connect, Schema, model, ObjectId, populate, aggregate, $match, $group, $lookup, lean, HydratedDocument, InferSchemaType, MongoClient, Atlas

When to use:

• Defining MongoDB schemas and models with Mongoose
• Building CRUD operations and complex queries
• Designing aggregation pipelines for analytics and reporting
• Managing indexes for query performance
• Connecting to MongoDB Atlas or local instances
• Modeling document relationships (embedding vs referencing)

Key patterns covered:

• Connection setup (Atlas URI, pooling, error handling)
• Schema definition (types, validation, defaults, enums)
• Models with TypeScript (automatic inference, methods, statics, virtuals)
• CRUD operations (create, find, update, delete, lean)
• Query building (filters, projection, sort, limit, populate)

When NOT to use:

• Highly relational data with complex joins and foreign key constraints (use a relational database)
• Strong ACID guarantees across many collections as a primary pattern (use a relational database)
• Simple key-value storage (use a dedicated key-value store)
• Fixed schemas where relational constraints are critical
• Time-series data at scale (use a dedicated time-series database)

Detailed Resources:

• For decision frameworks and anti-patterns, see reference.md

Core Patterns:

• examples/core.md - Connection, schema definition, model creation, TypeScript typing

Query Patterns:

• examples/queries.md - Complex queries, populate, lean, cursor, pagination

Aggregation:

• examples/aggregation.md - Aggregation pipeline, $match, $group, $lookup, $project

Advanced Patterns:

• examples/patterns.md - Schema design (embedding vs referencing), transactions, middleware hooks, virtuals

Indexing:

• examples/indexes.md - Index types, compound indexes, text search, geospatial, TTL, performance

---

Philosophy

MongoDB is a document database. Mongoose provides schema-based modeling on top of it. The core principle: data that is accessed together should be stored together.

Core principles:

1. Schema-first design -- Define schemas before models. Schemas enforce structure, validation, and defaults at the application layer.
2. Embed by default -- Co-accessed data belongs in the same document. Only reference when data is shared across many documents, grows unbounded, or is frequently updated independently.
3. Lean for reads -- Use .lean() for read-only queries. It returns plain objects (3x less memory) instead of full Mongoose documents.
4. Index your queries -- Every query pattern needs a supporting index. Compound indexes follow the Equality-Sort-Range (ESR) rule.
5. Aggregation over application logic -- Push data transformation to the database with aggregation pipelines instead of processing in application code.
6. TypeScript inference -- Let Mongoose infer types from schema definitions. Avoid manually duplicating interfaces unless you need methods/statics/virtuals.

When to use MongoDB / Mongoose:

• Document-oriented data (user profiles, product catalogs, content)
• Flexible schemas that evolve over time
• Hierarchical or nested data structures
• High read throughput with embedding
• Geospatial queries and full-text search

---

Core Patterns

Pattern 1: Connection Setup

Establish a single connection at startup with named constants for pool/timeout config and environment variables for credentials. See examples/core.md for full examples including connection events and graceful shutdown.

const connection = await mongoose.connect(process.env.MONGODB_URI!, {
  maxPoolSize: POOL_SIZE_MAX,
  minPoolSize: POOL_SIZE_MIN,
  serverSelectionTimeoutMS: SERVER_SELECTION_TIMEOUT_MS,
  socketTimeoutMS: SOCKET_TIMEOUT_MS,
  retryWrites: true,
  retryReads: true,
});

---

Pattern 2: Schema Definition with TypeScript

Let Mongoose infer types from the schema definition. Use explicit interfaces only when adding methods, statics, or virtuals. See examples/core.md for full typing examples with HydratedDocument, InferSchemaType, and generic parameters.

// Preferred: automatic type inference
const userSchema = new Schema(
  {
    name: { type: String, required: true, trim: true },
    email: { type: String, required: true, unique: true, lowercase: true },
    role: {
      type: String,
      enum: ["admin", "user", "moderator"] as const,
      default: "user",
    },
  },
  { timestamps: true },
);
const User = model("User", userSchema);

// For methods/statics/virtuals: explicit interfaces with Schema generics
const userSchema = new Schema<IUser, UserModel, IUserMethods, {}, IUserVirtuals>({ ... });

---

Pattern 3: CRUD Operations

Use .lean() for read-only queries, save() when middleware must fire, findByIdAndUpdate with { runValidators: true } for direct updates. See examples/core.md for full CRUD examples.

const user = await User.findById(id).lean(); // read-only, 3x less memory
await User.insertMany(users, { ordered: false }); // bulk insert
await User.findByIdAndUpdate(id, update, { new: true, runValidators: true }); // direct update

---

Pattern 4: Query Building

Use comparison/logical operators for filters, .populate() with field selection and limits. See examples/queries.md for dynamic query builders, cursor-based pagination, and populate patterns.

const post = await Post.findById(id)
  .populate("author", "name email")
  .populate({
    path: "comments",
    options: { sort: { createdAt: -1 }, limit: 10 },
  })
  .lean();

---

Pattern 5: Schema Validation

Add custom error messages, regex validation, and array-level validators. See examples/core.md for complete validation examples.

price: {
  type: Number,
  required: true,
  min: [0, "Price cannot be negative"],
  validate: { validator: (v: number) => Number.isFinite(v), message: "Price must be finite" },
},
sku: {
  type: String,
  required: true,
  unique: true,
  match: [/^[A-Z]{2}-\d{6}$/, "SKU must match format XX-000000"],
},

---

<red_flags>

RED FLAGS

High Priority Issues:

• Mutating a document fetched with .lean() and expecting .save() to work -- lean returns plain objects without Mongoose methods
• Registering middleware after model() call -- hooks are silently ignored
• Running operations in parallel inside a transaction (Promise.all()) -- MongoDB does not support parallel operations within a single transaction
• Using localhost in connection strings on Node.js 18+ -- IPv6 preference causes connection timeouts, use 127.0.0.1
• Missing { session } on any operation inside a transaction -- that operation runs outside the transaction

Medium Priority Issues:

• Using findOneAndUpdate / updateOne and expecting pre('save') hooks to fire -- only save() and create() trigger document middleware
• Unbounded .populate() without limit or field selection -- can return thousands of documents per query
• Not calling runValidators: true on findOneAndUpdate -- schema validation is skipped by default on updates
• Creating indexes in production code instead of migration scripts -- index builds lock the collection
• Using $where or JavaScript expressions in queries -- disables indexes and enables injection

Common Mistakes:

• Forgetting { new: true } on findOneAndUpdate -- returns the old document by default
• Using Schema.Types.ObjectId in TypeScript interfaces instead of Types.ObjectId -- Schema.Types.ObjectId is for schema definitions, Types.ObjectId is for interfaces
• Not handling duplicate key errors (code 11000) from unique indexes
• Calling .lean() on write operations -- lean is for reads only
• Checking doc.isNew in post('save') hooks -- always false after save, use this.$locals.wasNew set in a pre('save') hook

Gotchas & Edge Cases:

• MongoDB has a 16 MB document size limit -- deeply embedded arrays can hit this
• Mongoose buffers operations before connection is established -- queries queue silently if connection fails
• deleteOne / deleteMany do not trigger pre('remove') middleware -- use findOneAndDelete or document .deleteOne() if you need middleware
• Virtual properties are excluded from toJSON() / toObject() by default -- set { toJSON: { virtuals: true } } in schema options
• remove() was completely removed in Mongoose 7+ -- use deleteOne() or deleteMany() instead
• Mongoose 9 dropped callback-based next() in pre hooks -- use async/await instead
• Mongoose 9 renamed FilterQuery to QueryFilter -- update TypeScript imports if upgrading
• Mongoose 9 requires updatePipeline: true for pipeline-style updates -- they throw by default
• Mongoose 9 removed the background index option -- MongoDB 4.2+ builds all indexes in the background by default

</red_flags>

---

<critical_reminders>

CRITICAL REMINDERS

All code must follow project conventions in CLAUDE.md (kebab-case, named exports, import ordering, import type, named constants)

(You MUST define Mongoose middleware (pre/post hooks) BEFORE calling model() -- hooks registered after model compilation are silently ignored)

(You MUST pass { session } to EVERY operation inside a transaction -- missing session causes operations to run outside the transaction)

(You MUST use .lean() for read-only queries that send results directly to API responses -- skipping lean wastes 3x memory on hydration overhead)

(You MUST use 127.0.0.1 instead of localhost in connection strings -- Node.js 18+ prefers IPv6 and localhost can cause connection timeouts)

(You MUST NOT use findOneAndUpdate / updateOne and expect save middleware to fire -- only save() and create() trigger document middleware)

Failure to follow these rules will cause silent data corruption, middleware bypass, or transaction isolation failures.

</critical_reminders>