Connection pool storms after failover

Aurora MySQL. A 20-second primary failover happened at 14:00. The DB recovered, but for the next 8 minutes the new primary sat at Threads_connected = max_connections, CPU pegged, and the app threw 'too many connections' even though real query volume was normal. App tier is ~200 stateless pods, each with a 20-connection HikariCP pool and aggressive reconnect-on-error. Connections that do get in run fast. Walk through what happened and how you'd both recover now and prevent it.