How to Fix Slow API Response Time When the Database Looks Fine

Last month I had a production API that looked healthy in all the wrong places. CPU was low, PostgreSQL was not showing lock waits, and the endpoint still returned 2 to 4 seconds at p95 during the morning traffic spike. The frustrating part was that a single manual request in staging came back fast, so the first reaction was to blame the network or the CDN. I started by pulling trace IDs from the…