Phoenix WebSocket connection scaling: Memory overhead vs. reconnection frequency

Phoenix Channels leverage the Cowboy web server to maintain persistent WebSocket connections, utilizing the BEAM process model to isolate each client. In environments targeting millions of concurrent idle connections, there is a design trade-off between the memory footprint of maintaining these long-lived processes and the CPU/network overhead of aggressive reconnection strategies.

While the BEAM is optimized for lightweight processes, the cumulative memory overhead per connection can impact overall system stability when scaling to extreme volumes. This becomes critical when determining the optimal heartbeat interval and timeout configuration to prune stale connections without triggering a reconnection storm.

• What is the precise memory overhead per idle WebSocket connection in recent Phoenix versions?
• How does the cost of maintaining an idle connection compare to the resource spike caused by frequent mass reconnections?