System Design — Mid

1The 6-step framework

0/6▶

Mid level is the interview bar. Run the full framework fluently, including estimates, and narrate every step.

Run it in order

1. Clarify functional + non-functional requirements (scale, latency, consistency).

Primer — How to approach.

2. Back-of-envelope estimates: requests/sec, storage, read:write ratio.

Primer — Estimates.

3. API design — endpoints, verbs, status codes, pagination.

restfulapi.net.

4. Data model — tables/keys/indexes; RDBMS vs NoSQL with a reason.

Primer — Database.

5. High-level diagram: client → LB → app → cache → DB → queue/workers.

Excalidraw.

6. Scale & tradeoffs: caching, replication vs sharding, async, SPOFs.

Primer — index.

2API design

0/4▶

A clean API contract sets up the whole design.

Contracts

Resource modelling, versioning, and pagination (cursor vs offset).

restfulapi.net.

Idempotency keys for safe retries on writes.

Stripe — Idempotency.

Rate limiting & auth at the API edge.

Cloudflare — Rate limiting.

REST vs gRPC vs GraphQL — one-line tradeoffs.

gRPC — intro.

3Load balancing & app tier

0/3▶

How you scale the stateless middle horizontally.

Scale the app tier

L4 vs L7 load balancing (ALB routes by path/host); health checks.

AWS — ELB.

Stateless app servers; sticky sessions vs a shared session store.

Twelve-Factor — Processes.

Auto-scaling groups & how you'd scale a Node/Python API.

AWS — Auto Scaling.

4Caching

0/3▶

The highest-leverage performance lever — and the easiest to get subtly wrong.

Cache well

Layers: CDN → HTTP cache → Redis → query cache; what each holds.

AWS — Caching.

Cache-aside vs write-through vs write-back.

AWS — Caching strategies.

TTLs & invalidation; the two hard problems of caching.

Cloudflare — Caching.

5Databases

0/3▶

Name the right store and scaling lever, and the pain each adds.

Data decisions

Read replicas (scale reads) vs sharding (scale writes) — and their pain.

Primer — replication & sharding.

SQL vs NoSQL tradeoffs; choosing a store per access pattern.

Primer — SQL or NoSQL.

Indexing basics & the N+1 problem.

Use The Index, Luke.

6Queues & async

0/3▶

Decoupling slow work is a core scaling and reliability move.

Work in the background

Message queue + workers — decouple slow work; smooth spikes.

AWS — Message queue.

At-least-once delivery + idempotent consumers; dead-letter queues.

AWS — DLQ.

Pub/sub vs a work queue; when to fan out.

AWS — Pub/Sub.

7Reliability basics

0/3▶

Mid engineers design so one failure doesn't cascade. Mention these unprompted.

Make it not fall over

Timeouts + retries with backoff — why a naive retry worsens an outage.

AWS — Backoff with jitter.

Idempotency so a retried write doesn't double-charge.

Stripe — Idempotency.

Health checks + graceful degradation (serve stale/partial data).

AWS Builders' Library.

8Classic prompts

0/8▶

Do each once with the framework, out loud. Then you can adapt to anything.

Practise out loud

URL shortener — key generation, redirect path, read-heavy caching.

Primer — Pastebin/Bit.ly.

Rate limiter — token bucket vs sliding window; Redis counter.

Cloudflare — Rate limiting.

Notification system — trigger, schedule, fan-out to many users.

ByteByteGo — search "notification system".

Typeahead / autocomplete — trie + caching + ranking.

ByteByteGo — search "autocomplete".

Leaderboard — Redis sorted sets, ties, pagination.

Redis — Sorted sets.

File upload & storage — presigned URLs to S3, metadata in a DB.

AWS — Presigned URLs.

Job scheduler / cron — due-time queue, workers, idempotency.

ByteByteGo — search "distributed scheduler".

Web crawler — frontier queue, politeness, dedup with a bloom filter.

ByteByteGo — search "web crawler".