Server-Sent Events in production: lessons learned after 2 years

we've been using server-sent events (sse) in production for about two years now for all our real-time notifications, and it's been mostly stable. however, we've hit some interesting scaling limits and learned a few lessons. managing 50k concurrent sse connections requires careful tuning of os limits and server memory. connection limits per ip are common issues. client-side, robust reconnection logic with exponential backoff is absolutely critical. we also found sticky sessions to be a must-have for load balancers to ensure clients reconnect to the same server. what are others' practical experiences running sse at scale? any insights into memory management per connection or specific load balancer configurations that helped?