MERN ENGINEERING PLAYBOOK

Modern Best Practices for Scalability, Maintainability & Performance. Basically: “If someone joins my startup tomorrow — this is how we write React, Node, Express, Mongo in 2026 so that the codebase doesn’t become a nightmare in 12 months.” Not theory. Not tutorials. Just: what to do / what not to do in production-grade MERN apps.

---

🟦 PART 1 — REACT (CLIENT ARCHITECTURE)

This section answers: How should we write React apps in 2026 so they remain scalable after 50+ features?

1. Component Philosophy

✅ Use:

• Functional Components only
• Composition over inheritance
• Custom Hooks for logic reuse
• Controlled components for forms
• Presentational vs Container separation
• Headless UI approach (logic separated from UI)

// Example:
const useCategories = () => {
  return useQuery(categoryQueryKeys.list(), fetchCategories)
}

// UI stays clean:
const CategoryList = () => {
  const { data } = useCategories()
}

❌ Avoid:

• Class Components
• Large “God Components” (Massive 400-line components)
• Prop Drilling → use Context/Zustand
• Business logic inside UI components
• API calls inside JSX files
• Components managing unrelated state

If your component fetches data, filters it, sorts it, validates it, and mutates it, that is NOT a component anymore. That’s a mini-backend 😭. Move logic to: hooks/, queries/, store/, utils/.

---

2. TypeScript Rules

✅ Always write like this:

type CategoryFormProps = {
  onSubmit: (data: CategoryInput) => void // (or data: Category)
}

const CategoryForm = ({ onSubmit }: CategoryFormProps) => {}

❌ Never use: const CategoryForm: React.FC<Props>

Why? React.FC adds implicit children, messes with generics (bad generic inference), breaks defaultProps typing, makes inference harder, and makes DevTools debugging messier.

---

3. State Management Strategy (2026 Standard)

Modern React has 4 types of state:

State Type	Example	Tool
UI State	modal open	Zustand
Server State	users from DB	Tanstack Query
Form State	login form	React Hook Form
URL State	page=2	React Router / nuqs
Global Auth	—	Zustand + Persist

🔥 GOLDEN RULE: Server State ≠ Client State

❌ NEVER: Store API response in Zustand, store React Query data in Redux, or duplicate backend data locally. Let React Query own server data lifecycle: caching, retries, background refetch, deduping, and stale logic. Avoid Redux unless enterprise-level scale, and avoid useState for shared state.

---

4. Data Fetching Pattern

❌ Old Pattern: Fetch inside useEffect

useEffect(() => {
  fetch('/api/users')
}, [])

This causes: race conditions, double fetching, stale UI, no caching, bad retries, and manual loading states.

✅ New Standard: Always use query.ts, mutation.ts, and queryKeys.ts. Use useQuery(), useMutation(), Query Key Factory pattern, Optimistic Updates, and Suspense Mode.

// Example:
export const categoryQueryKeys = {
  all: ['categories'] as const,
  list: () => [...categoryQueryKeys.all, 'list'] as const,
}

export const useCategories = () =>
  useQuery({
    queryKey: categoryQueryKeys.list(),
    queryFn: fetchCategories,
    staleTime: 5 * 60 * 1000
  })

---

5. Folder Structure (Feature First)

✅ Use:

src/
  modules/
    category/
      components/
      hooks/
      api/
      types/
      category.query.ts
      category.mutation.ts
      category.store.ts

❌ Avoid: Root-level components/, pages/, utils/, services/. Type-based structure doesn’t scale. Feature-based = modular.

---

6. Forms (Huge Source of Bugs)

✅ Use: React Hook Form and Zod Resolver.

❌ Never: Manage form state manually, use useState for inputs, or validate in onSubmit.

---

7. Performance Rules

✅ Use: Lazy loading for routes (with Suspense), React.memo only when needed, useCallback only for child prop stability (handlers passed down), Virtualized Lists (react-virtual) for large tables.

❌ Avoid: Premature memoization (memoizing everything), inline anonymous functions/heavy computations in heavy lists/JSX, derived state stored in state.

• Bad: const [total, setTotal] = useState(cart.reduce(...))
• Good: const total = useMemo(() => cart.reduce(...), [cart])

---

8. Global State (Zustand Pattern)

Use Zustand ONLY for: auth state, theme, UI toggles, and local preferences.

Never: backend data, form inputs, server responses.

---

9. Error + Loading UI

• Never: Handle loading manually everywhere.
• Use: React Query Suspense and Error Boundaries.

---

10. API Layer Pattern

Never call axios inside component.tsx. Always use api/category.api.ts.

---

🚨 FINAL REACT RULES CHECKLIST

✔ Hooks > Classes ✔ Tanstack Query for server state ✔ Zustand for UI state ✔ React Hook Form for forms ✔ Zod for validation ✔ Feature-based folder structure ✔ API layer abstraction ✔ No fetch in useEffect ✔ No React.FC ✔ No global server state

---

🟩 PART 2 — NODE.JS BACKEND ARCHITECTURE (2026 STANDARD)

1. First Principle: Your Backend is NOT MVC

Classic MVC (route → controller → model) looks nice in tutorials but becomes controllers with 300 lines, models called directly, business logic duplicated in 7 controllers, impossible testing, and tight coupling in real apps. Avoid MVC-only backend (doesn’t scale well).

✅ Use This Instead: Layered Architecture OR Domain Driven Modular Architecture. route → controller → service → repository → database. Each layer has ONLY ONE responsibility.

Layer	Responsibility
Route	define endpoint
Controller	HTTP handling
Service	business logic
Repository	DB interaction
Model	schema only

🚨 Rule: Controllers should NEVER talk to database directly. If you see User.find() inside controller → architecture violation.

---

2. Project Structure (Scalable)

❌ Avoid: controllers/, models/, routes/, services/. This type-based structure breaks at scale.

✅ Use Domain-Based Structure:

src/
  modules/
    user/
      user.route.ts
      user.controller.ts
      user.service.ts
      user.repository.ts
      user.schema.ts
      user.dto.ts

Everything related to “user” lives in one module. Now: features are isolated, testing becomes easy, onboarding is faster, and less merge conflicts.

---

3. Controller Rules

Controllers are the translation layer between HTTP world and business world.

✅ Controller SHOULD: Validate input, Call service, Return response.

❌ Controller SHOULD NOT: Query DB, Run business logic, Transform domain logic, Call external APIs, Perform calculations.

• Bad: const user = await User.findById(id); user.balance += 200; await user.save()
• Good: await userService.addBalance(id, 200)

---

4. Service Layer (Your Real Backend)

Service Layer contains: business rules, workflows, calculations, multi-db operations, external integrations.

Example: await walletService.transferMoney(sender, receiver, amount). Inside: validate balance, debit, credit, transaction logging, rollback logic. None of this belongs in controller.

---

5. Repository Layer (DB Abstraction)

Service should NOT care if you are using Mongo, Postgres, Redis, or Firestore. Repository handles DB.

Example: userService → userRepository → mongoose. Now: You can replace Mongo later without rewriting business logic.

---

6. Async Handling Pattern

❌ Avoid: Nested awaits or blocking operations (sync fs, crypto).

✅ Use: async/await everywhere.

// Promise.all for parallel tasks
await Promise.all([
  save(),
  sendMail(),
  log()
])

Parallel execution = faster APIs.

---

7. Error Handling System

Never: throw raw strings, send stack traces to client, or write try/catch in every controller.

✅ Create Custom Error Class and Global Middleware:

class AppError extends Error {
  constructor(message, statusCode) {
    super(message)
    this.statusCode = statusCode
  }
}

// Controller:
next(new AppError('User not found', 404))

Use Error Codes (not only messages).

---

8. DTO Pattern (MANDATORY)

Never trust req.body. Use DTOs (e.g., createUser.dto.ts) to ensure validation, shape consistency, type safety, and prevent overposting attacks.

// Controller:
const dto = createUserSchema.parse(req.body)
await userService.create(dto)

Use Zod; avoid manual validation.

---

9. Logging Strategy

Never use console.log() in production. Use Pino (fastest) for structured JSON logs.

{
  "level": "info",
  "service": "auth",
  "userId": "123"
}

Logs should track: requestId, userId, module, latency.

---

10. Background Jobs (Don’t Block APIs)

Never await sendEmail() inside a signup API. The user should not wait for the email service.

Use BullMQ, Worker Threads, or Message Queue. Flow: signup → queue email job → API returns immediately.

---

11. Dependency Injection (Important)

Avoid import userRepository inside service directly.

Instead, inject dependency: new UserService(userRepository). Now it is testable, replaceable, and loosely coupled.

---

12. Config Management

Never use process.env.JWT_SECRET everywhere inside 25 files. Use config/env.ts and config/db.ts to export validated config once.

---

🚨 FINAL NODE RULES CHECKLIST

✔ Domain-based modules ✔ Controller thin ✔ Service = business logic ✔ Repository = DB access ✔ DTO validation ✔ Custom Error System ✔ Global Error Middleware ✔ Structured logging ✔ Background jobs for async tasks ✔ Promise.all for parallel work ✔ Dependency Injection ✔ Central config management

---

🟨 PART 3 — EXPRESS API LAYER (2026 STANDARD)

Express is NOT your backend logic. Express is your Transport Layer.

It should only: receive request, validate, authenticate, pass to controller, send response. Nothing else.

---

1. Request Lifecycle (MANDATORY FLOW)

Every API request should follow:

Request
  ↓ Global Middleware
  ↓ Auth Middleware
  ↓ Validation Middleware
  ↓ Controller
  ↓ Service
  ↓ Repository
  ↓ DB
  ↓ Response Formatter

If this flow breaks → codebase becomes unpredictable.

---

2. Route Design Pattern

❌ Never: Put logic in routes, DB calls inside controller, or business logic inside route.

• Bad: router.post('/user', async (req,res)=>{ const user = await User.create(req.body) })

✅ Routes Should Only: Map endpoint → middleware → controller.

router.post(
  '/',
  validate(createUserSchema),
  authMiddleware,
  userController.createUser
)

Routes = traffic police 🚦 They just direct the request. Controllers validate input, call service, return response. Services handle business logic and DB calls.

---

3. Middleware Layering

Your Express app should have 3 middleware types:

Type	Purpose
Global	CORS, Helmet
Route	Auth
Feature	Validation

• App Level: app.use(cors()); app.use(helmet()); app.use(requestLogger)
• Route Level: router.use(authMiddleware)
• Feature Level: router.post('/', validate(schema), controller)

---

4. Validation Flow

Never check manual req.body in controllers (e.g., if(!req.body.email)).

✅ Use: Zod / Joi and Celebrate middleware.

export const validate =
  (schema) =>
  (req, res, next) => {
    schema.parse(req.body)
    next()
  }

Now controller only receives valid data.

---

5. Auth Middleware Pattern

Never decode JWT inside controller (e.g., jwt.verify()).

✅ Create: auth.middleware.ts. Attach user to request: req.user = decodedUser. Now downstream layers use: req.user.id. Clean separation.

---

6. Response Formatter (VERY IMPORTANT)

Avoid: res.json(user), res.send(data), res.status(200).json() — Different devs = different response shapes 😭

✅ Create: response.util.ts. Example: res.success(data), res.error(message).

Now every API response becomes predictable frontend handling:

{
  "success": true,
  "data": {}
}

---

7. API Versioning

Never use /api/users. Use /api/v1/users. Later: /api/v2/users. Avoids breaking mobile apps, old clients, and frontend crashes.

---

8. Rate Limiting & Security (Production Mandatory)

Use Helmet, CORS config, JWT Rotation, Bcrypt with salt rounds ≥ 10, and express-rate-limit. Avoid storing secrets in code or sending stack traces to client.

Protect: login, OTP, public endpoints. Example: Attach loginLimiter → router.post('/login', loginLimiter, controller).

---

9. Centralized Error Handling

Never use res.status(500) inside controller.

• Controller: next(new AppError('Not found', 404))
• Global Middleware: error.middleware.ts handles formatting, status code, logging.

---

10. Async Wrapper

Avoid try/catch everywhere. Use asyncHandler(controller) to wrap controller once for a cleaner codebase.

---

11. Request Logging

Track: requestId, userId, route, latency. Attach requestId: req.id = uuid(). Now logs become traceable in production.

---

12. Route Protection Strategy

Route Type	Middleware
Public	none
Private	auth
Admin	auth + role

Example: authorize('admin').

---

🚨 FINAL EXPRESS RULES CHECKLIST

✔ No logic in routes ✔ Validation middleware ✔ Auth middleware ✔ Versioned APIs ✔ Global error handler ✔ Async wrapper ✔ Response formatter ✔ Request logging ✔ Rate limiting ✔ Role-based route protection

---

🟥 PART 4 — MONGODB + MONGOOSE (2026 STANDARD)

1. Golden Rule of MongoDB

MongoDB is a Query-first database. Not relation-first, normalization-first, or table-first. You design based on: How the data will be read most frequently.

---

2. Embed vs Reference (MOST IMPORTANT DECISION)

✅ EMBED WHEN:

• One-to-Few relationship, frequently read together, data size is small, data rarely updated separately.
• Examples: user → addresses, product → tags, order → shipping status.

✅ REFERENCE WHEN:

• One-to-Many relationship, data grows over time, frequently updated independently, used across collections.
• Examples: user → orders, product → reviews, course → students.

🚨 Avoid: Unbounded arrays (e.g., comments: []). This can break Mongo’s 16MB document limit later.

---

3. Hybrid Pattern (Production Standard)

Use Hybrid when needed. Embed frequently read fields, reference heavy dynamic data.

Example order: userId, total, itemsSummary (embedded), itemsRef (referenced). Fast reads + scalable writes.

---

4. Must Use Schema Options

Always use timestamps: true and versionKey: false. Also use toJSON: { virtuals: true } and toObject: { virtuals: true } — needed for computed fields.

---

5. Index Strategy (NON-NEGOTIABLE)

If your query filters → it needs an index. Avoid find() without index.

• Index: email, status, createdAt, foreign keys, sort fields.
• Compound Indexes: userId + createdAt needed for pagination and dashboard queries.

🚨 Avoid: Indexing everything. Each index increases write time and consumes memory.

---

6. Query Optimization: Use lean() for Read APIs

Normal Mongoose returns heavy document instances. You don’t need that for APIs.

Always use .lean() for reads (e.g., User.find().lean()). Lean is 30–50% faster, uses lower memory, and returns plain JSON.

Avoid lean when: using virtuals, using middleware, or modifying document.

---

7. Pagination Pattern

Never use skip(50000) — Mongo will scan 50k docs first 😭

Use Cursor Pagination based on createdAt or _id.

// Pattern: Scales infinitely.
find({ createdAt: { $lt: lastSeen } })
  .limit(10)
  .sort({ createdAt: -1 })

---

8. Aggregation Pipeline Rules

Aggregation is Mongo’s SQL JOIN + GROUP BY equivalent.

Always:

1. Use $match first / early.
2. Use $project early to reduce payload.
3. Use indexed fields.
4. Use $facet for dashboards / analytics.
5. Use $lookup only when necessary / sparingly.

🚨 Avoid: Aggregation without index support or pipeline before filtering ($lookup → $match explodes memory usage).

---

9. populate() Usage

Never use .populate('orders') for lists (Populate = hidden join). Use aggregation or manual join for large datasets. Populate okay only for detail pages.

---

10. Soft Delete Pattern

Never use User.deleteOne(). Avoid hard deletes in production apps.

Use deletedAt: Date | null. Query: { deletedAt: null }.

---

11. Virtual Fields

Use for fullName, computed totals, derived stats. Example: userSchema.virtual('fullName'). No DB storage needed.

---

12. Projection (Huge Performance Gain)

Never use User.find() blindly. Use .select('name email') for less payload → faster APIs.

---

13. Pre/Post Middleware

Use for hashing password, audit logs, timestamps, cascading soft delete. Avoid heavy logic inside middleware.

---

14. Transaction Usage

Use for payments, wallet updates, inventory deduction with session.startTransaction(). Never rely on Mongo atomicity for multi-document ops.

---

🚨 FINAL MONGOOSE RULES CHECKLIST

✔ Embed small related data ✔ Reference growing data ✔ Hybrid for balance ✔ Always index query fields ✔ Use lean() for reads ✔ Cursor pagination ✔ Match early in aggregation ✔ Avoid populate for lists ✔ Use soft delete ✔ Use projections ✔ Use transactions for multi-doc ops ✔ Virtuals for computed data

---

🟪 PART 5 — TYPESCRIPT ENGINEERING GUIDELINES (2026)

1. tsconfig.json (MANDATORY BASELINE)

If this is not strict → your whole type safety is fake.

✅ Must Enable:

{
  "compilerOptions": {
    "strict": true,
    "noImplicitAny": true,
    "strictNullChecks": true,
    "strictFunctionTypes": true,
    "strictBindCallApply": true,
    "strictPropertyInitialization": true,
    "noImplicitThis": true,
    "alwaysStrict": true,
    "isolatedModules": true,
    "forceConsistentCasingInFileNames": true,
    "skipLibCheck": true
  }
}

---

2. Type Inference Philosophy

❌ Avoid: const name: string = "Abhishek" (Redundant).

✅ Use: const name = "Abhishek" (Let TS infer internally).

🚨 Exception: Always type function parameters, public API responses, DTOs, service returns.

---

3. Interfaces vs Types (Production Rule)

• Use interface when: defining object shapes, extendable domain models, class contracts.

interface User { id: string; email: string }

• Use type when: unions, tuples, mapped types, utility composition.

type Role = 'admin' | 'user'

---

4. Never Use any

❌ Bad: function process(data: any) — Removes autocomplete, compile checks, runtime safety.

✅ Use: unknown and narrow later.

function parse(data: unknown) {
  if (typeof data === "string") {
    return data.toUpperCase()
  }
}

---

5. DTO Pattern (Full Stack Safety)

Backend should NEVER accept req.body directly. Use DTO validation via Zod.

// createUser.dto.ts
export interface CreateUserDTO {
  email: string
  password: string
}

Now: controller safe, service safe, DB safe. Prevents overposting attack (User sending admin: true manually).

---

6. Async API Pattern

Always type API responses.

❌ Bad: async function getUser()

✅ Good: async function getUser(): Promise<User>

Frontend: useQuery<User>(). Full stack type sync achieved.

---

7. Generics for Reusability

Use in repositories, API clients, services.

async function fetchData<T>(url: string): Promise<T>

Now reusable across users, orders, products.

---

8. Type-Only Imports (Performance)

• Avoid: import { User } from './types'
• Use: import type { User } from './types' — Prevents runtime bundle pollution.

---

9. const Assertions (Union Safety)

const ROLES = ['admin', 'user'] as const
type Role = typeof ROLES[number]

Strongest possible union.

---

10. Type Guards

Needed for external APIs, unknown input, JSON parsing.

function isUser(value: unknown): value is User {
  return typeof value === 'object'
}

---

11. Null Handling (Huge Runtime Crash Source)

• Never: user.name.length
• Use: user?.name ?? "Anonymous"

---

12. Avoid Deep Type Magic

❌ Avoid: Recursive mapped types everywhere. Slows IDE, build time, compile time.

✅ Prefer: Partial<User>, Pick<User, 'id'>, Readonly<User>.

---

13. Dependency Injection Friendly Types

interface PaymentGateway {
  charge(amount: number): Promise<boolean>
}

Service depends on interface — not implementation. Testable architecture achieved.

---

14. Folder Structure for Types

Never use a generic types.ts. Use domain based typing:

user/user.types.ts
user/user.dto.ts

---

15. Testing Types

Use @ts-expect-error to test invalid cases.

---

🚨 FINAL TYPESCRIPT RULES CHECKLIST

✔ Strict mode enabled ✔ Inference > explicit types ✔ Interface for models ✔ Type for unions ✔ No any ✔ unknown for external data ✔ DTO validation ✔ Typed async returns ✔ Generic APIs ✔ type-only imports ✔ const assertions ✔ Type guards ✔ Safe null handling ✔ Domain based typing ✔ Avoid deep mapped types

---

⚙️ CROSS-STACK BEST PRACTICES

• API: Use REST naming consistency, Pagination, DTO Pattern.
• Caching: Use React Query cache, Redis for backend caching.
• Auth Flow: Use Access Token (short-lived), Refresh Token (long-lived), HTTPOnly cookies.
• Testing: Use Vitest (frontend), Jest (backend), Supertest (API).
• Dev Tooling: Use ESLint, Prettier, Husky, Lint-staged.