Writing an Instrumented Test Harness for System Calls
Understanding the Problem
You’ve just gotten paged at 2 AM because your service is burning CPU in system time.
perf topshows 40% in kernel code, but which syscalls?stracewould tell you, except it makes your already-struggling service 10x slower and miss its SLOs entirely. You need instrumentation that measures syscalls without destroying performance.
Why Standard Tools Fall Short
