Black Friday autoscaling exhausts database connections

Black-Friday traffic ramps and your stateless API tier autoscales smoothly from 30 to 90 pods, CPU stays comfortable, and yet p99 latency climbs and a growing share of requests time out. Each pod holds a fixed DB connection pool of 10. Postgres max_connections is 400 and the pg_stat_activity count is pinned right at 400, with many sessions idle-in-transaction or waiting on a lock. App dashboards show request latency dominated by 'time waiting to acquire a DB connection from the pool.' Adding more pods made it worse, not better. How do you triage and what do you change?