How a 5ms Blocking Call Crashed a Service from 17,189 to 549 req/s

A software engineer discovered that a seemingly harmless 5-millisecond blocking call caused a dramatic drop in service throughput. When executed on a Tomcat worker thread, the service handled 17,189 requests per second without issue. However, the same call made on an event-loop thread caused performance to collapse to just 549 requests per second. The hardware, JVM, and code remained identical across both scenarios — the only variable was the type of thread executing the call. The finding highlights how blocking operations can be catastrophic in event-driven, non-blocking architectures even when the delay appears negligible.