A well-tuned Postgres or MySQL instance handles 10k-50k QPS. Modern cloud instances with 512 GiB RAM hold enormous working sets without touching disk. Don't shard until you've done the math. OpenAI didn't for ChatGPT.
Redis runs at ~200k ops/sec with sub-millisecond latency. A single memory-optimized instance can hold terabytes. "We need to shard the cache" is almost never the right call.
The interview move: when designing a leaderboard or any Top-K feature, propose Redis Sorted Sets, then follow up with "I'd shard by region and use a separate set per time bucket." That shows you understand both the data structure and its operational limits.
At 100 users, a ranked leaderboard from a relational DB is fine. At 10 million concurrent users, querying and sorting on every request will kill you. Redis Sorted Sets exist for exactly this.
The tradeoff most candidates get backwards: they optimize for interview performance instead of building the depth that makes interview performance easy.
82 Followers 1K FollowingSecuring Agentic Workflows with distributed systems and cryptography. Open for work/collaboration. Always learning always evolving.
1K Followers 907 FollowingBuilder. Older than all dogs. Creating: Hello Interview. Sold 2 startups, failed 1, ready for more. Previously ML eng leader FB, AMZN.