System Design & Highload
A compact, no‑fluff field guide for a distributed‑systems interview: the theory you must recall, the vocabulary that signals seniority, diagrams you can redraw on a whiteboard, and ready‑to‑speak answers to the classic scenario questions.
How to use this workbook
Read top‑to‑bottom once for coverage, then re‑read only the 🔑 key idea, 🎤 say‑this, and cheat sheet blocks the night before. Every chapter follows the same rhythm:
Theory
The mental model in a few tight paragraphs — enough to reason from first principles.
Practice
Concrete numbers, code, HTTP examples and diagrams you can reproduce live.
Say this
Crisp phrasings that make you sound like you have run this in production.
Chapters
Distributed Systems
The building blocks: stateless design, caching, load balancing, retries, backpressure, idempotency, circuit breakers and graceful degradation.
API Design
REST principles, idempotency keys, pagination, versioning, auth and rate limits — plus the async job pattern (POST /tasks → poll → webhook).
Reliability Engineering
SLI vs SLO vs SLA, error budgets, the availability “nines”, p95/p99 latency, retries with exponential backoff and jitter, and incident handling.
Scenario Playbooks
Structured, out‑loud answers to “latency doubled”, “a model provider is failing”, and “traffic just went 10×”. Mitigate first, root‑cause later.
Design Drills
Five whiteboard‑ready walkthroughs: an API for millions of clients, a multi‑provider LLM routing layer, a distributed rate limiter, a highly available REST service, and a telemetry pipeline — each with architecture, tradeoffs and talking points.
The 60‑second mental model
Almost every highload question decomposes into the same loop. Keep this diagram in your head and narrate against it.
“Push state to the edges (cache/DB/queue), keep compute stateless so it scales horizontally, protect every hop with timeouts, retries and circuit breakers, and measure everything with SLOs on p99 so I know when to mitigate.”