Love this, but you know how every system design interview goes: “Nice toy URL shortener… now make it handle 5B redirects/day, multi-region, zero downtime, analytics, abuse protection and no data loss.” At that point your 160 lines in C turns into something like: Public edge: DNS + Anycast → global entry CDN/edge layer to cache popular short links close to users WAF + rate limiting so your “minimal” service doesn’t die to a botnet Read path: Edge cache hit for hot URLs Miss → API tier behind load balancer In-memory cache (Redis/Memcached) for short→long lookups Fallback to primary datastore (say DynamoDB / Postgres with sharding) Write path: POST /shorten → API tier ID generation via Snowflake/segment allocators to avoid collisions Persist mapping in DB Push event to Kafka/Kinesis for async stuff (analytics, fraud checks, email hooks etc) Cold storage + analytics: All click events streamed to S3 as cheap, durable storage Presto/Trino/Athena over S3 for reporting Aggregated stats materialized back into a “stats” store for fast dashboards Reliability: Multi-region active/active or active/passive S3 as backup for mapping snapshots + disaster recovery Background jobs to rehydrate caches, reindex, move old data to cheaper tiers Interviewer brain: “Cool, now talk about: hot key mitigation schema migrations without downtime rolling deploys / canaries how you test this without taking prod down”