The goal is to track long-latency operations, both that are externally visible and internal (e.g., worker threads), so that we (a) know when they happen and (b) an (ideally) identify their root causes.