p99 latency on a critical API jumped from 120ms to 3s an hour ago. p50 is unchanged. How do you find the cause?

Core Summary

An unchanged p50 with a blown p99 means most requests are fine and a specific subset is not — which rules out broad causes like CPU saturation and points at something correlated: one dependency, one shard, one node, one customer, or garbage collection. The shape of the latency distribution is the diagnosis.