Ephemeral port exhaustion from disabled keep-alive
A high-throughput billing worker that makes many short-lived HTTPS calls to a payment processor starts logging 'cannot assign requested address' (EADDRNOTAVAIL) on outbound connects at peak, ~16:00 daily. The processor's endpoint and the worker IP are fixed. `ss -s` shows ~28,000 sockets in TIME_WAIT to a single (dst IP, dst port 443) tuple. `net.ipv4.ip_local_port_range` is the default 32768–60999. The team recently 'optimized' an HTTP client by disabling keep-alive to 'avoid stale connections' after an unrelated incident. Throughput is up 20% week over week. Triage and fix.
Stop the bleeding first (mitigate), then form hypotheses from real signals. Separate root cause from symptom, communicate status as you go, and close with what prevents a repeat.