Connection pool holds idle

At 12:10 your reporting API starts returning 500s under moderate load. Logs are full of HikariPool-1 - Connection is not available, request timed out after 30000ms. The Postgres server itself is healthy, CPU 30%, only 60 of its max_connections=200 in use, no long-running queries on its side. The app pool is configured for 50 connections and they're all checked out, but DB-side activity is low, so the connections appear idle-in-pool-but-held by the app. A feature shipped this morning added a new 'export full history' endpoint. Throughput on the rest of the API has also degraded. How do you triage and fix this?