Designing an API Performance Monitoring System at Scale

How do you track latency, throughput, and error rates across millions of requests per second — without slowing down a single one? This is one of those system design questions that feels straightforwa