Why average latency hides evening checkout pain
Daily average latency on a checkout API can look calm while evening shoppers wait through TLS queues and cold cache misses. In several Korean retail paths we reviewed, the midday sample sat under 200 ms while the 20:00–22:00 window stretched past a second at the 95th percentile.
The fix was not “add more dashboards.” It was agreeing on a capture window that matched marketing calendars, then timing each hop from edge termination through application handlers. Cache warmers and pool limits only made sense after the path showed where wait time actually sat.
If your monitors report health while support hears “pay button froze,” schedule measurement for the hours customers actually buy — not the hours that are convenient for change tickets.