System Design

System Design: Revision Sheet

One-page rapid review. Read this the night before your interview.

Answer Framework (R.U.D.T.S.)

1. Requirements — functional + non-functional, ask clarifying questions
2. Estimates — users, RPS, storage (rough math is fine)
3. Design — draw boxes: client → API → DB → cache
4. Deep dive — 1–2 areas interviewer picks
5. Tradeoffs — what changes at 10x scale

Scalability

• Vertical: bigger machine (limits, single point of failure)
• Horizontal: more machines (needs load balancer, stateless apps)
• Start simple → scale what breaks first

Availability

• Redundancy, health checks, no in-memory sessions
• 99.9% ≈ 8.7 hrs downtime/year

Latency Reduction

Cache → CDN → indexes → fewer round trips → compression

CAP Theorem

During network partition: pick Consistency OR Availability

• Bank balance → consistent
• Like counts → eventually consistent OK

Load Balancer

Distributes traffic across servers. Algorithms: round-robin, least connections.

• L4: IP/port | L7: URL/header routing
• WebSockets need sticky sessions OR Redis pub/sub

Caching Layers

Browser → CDN → Redis → Database

• Cache-aside: check cache → miss → DB → write cache
• TTL: auto-expire stale data
• Invalidate on write for user-owned data

CDN

Static assets (JS, CSS, images) at edge locations near users. Cache busting: app.a3f2b1.js (content hash in filename).

API Design Quick Hits

• Nouns not verbs: /users not /getUsers
• Cursor pagination for feeds (not offset)
• Status codes: 200, 201, 400, 401, 403, 404, 429, 500
• Rate limit auth endpoints

Auth Flow

Login → validate bcrypt → session cookie (HTTP-only) OR JWT
→ middleware verifies → attach user → authorize by role

Never: plaintext passwords, JWT in localStorage

Database

• PostgreSQL default for relational data
• Index: foreign keys, WHERE columns, ORDER BY columns
• Read replicas for read-heavy workloads
• Denormalize counts (like_count) for fast reads

WebSockets (Chat)

Client ←WS→ Server → Postgres (persist)
              ↓
         Redis pub/sub (multi-server fan-out)

REST for history, WebSocket for real-time

URL Shortener Pattern

POST long URL → base62(auto-id) → store → return short link
GET /:code → Redis cache → 302 redirect (async analytics)

Social Feed Pattern

• Pull model: query on read (simple, start here)
• Fan-out on write: precompute feed (scale path)
• Cache feed page 1 in Redis (60s TTL)
• Images: S3 + CDN, not in DB

Frontend Design Hits

• Feature-based folder structure
• React Query for server state, local state for UI
• Code splitting (React.lazy), virtualization (react-window)
• Infinite scroll: cursor pagination + IntersectionObserver
• Autocomplete: debounce 300ms, min 2 chars

Security Checklist (Always Mention)

• HTTPS
• bcrypt passwords
• Input validation
• Auth on every protected route
• Rate limiting
• No secrets in git

Quick Math

Designs to Practice

1. URL Shortener (easiest — 20 min)
2. Chat App (WebSockets + persistence)
3. Social Media Feed (pull vs fan-out)

Last check: Draw URL shortener on paper in 5 minutes. Say every box out loud.