I wonder which combination of HTTP reverse proxy (e.g. nginx, cloud load balancer) and application server (e.g. Python WSGI server) is best at using bounded queues and signaling backpressure (e.g. HTTP 503) rather than just letting requests pile up.