I'm sure most of us have had to deal with issues reported by end users that we ourselves aren't able to reproduce

This video is an extended case study going through my thought process as I tried to track down and fix a mysterious performance regression which impacted a small subset of end users

I look at the impact of acquiring mutex locks across different threads, identifying hot paths by attaching to running processes, using state snapshot comparisons to avoid triggering hot paths unnecessarily, the memory implications of bounded vs unbounded channels, and much more