intermediate · latency

Five Seconds at 15% CPU

API p99 is 5 seconds. CPU is at 15%, memory is fine. What is going on?

Practical scenario

A reporting API has a p99 of 5.2 s. Every host graph is green: CPU 15%, memory 40%, network unremarkable. The database team says their metrics look normal too.

What it tests

  • Whether the candidate can distinguish waiting from working
  • Breadth of the "what could it be waiting for" list
  • Whether they know utilisation cannot detect waiting

Lessons behind this question