PostgreSQL connection pool exhaustion under burst traffic

Running a pool of 20 connections (pgBouncer in transaction mode) behind a Node.js API. Under normal load (~50 req/s) it's fine. But during cron-job bursts (~400 req/s for 30s) we hit pool exhaustion — connections queue up, response times spike to 8-12s, then the healthcheck fails and the orchestrator restarts the pod. Options we're weighing: 1. Increase pool size (but then we hit DB max_connections) 2. Add a request queue with backpressure at the API layer 3. Separate the cron workers onto a different pool/database replica Has anyone dealt with this pattern? We're on PostgreSQL 16 + pgBouncer 1.20, deployed on K8s with HPA. Interested in what worked at scale.