See inside running services.

Attach to a live process, capture arguments, return values and timings where the problem actually happens, and detach — no restart, no redeploy, no code change.