Cannot assign requested address

An API gateway pod begins throwing intermittent 'cannot assign requested address' / connect failures to a backend after a refactor that 'simplified' the HTTP client by creating a fresh client per outbound call instead of reusing a pooled one. Errors climb with traffic and disappear after a pod restart, then return. CPU/mem are fine. ss/netstat shows tens of thousands of sockets in TIME_WAIT to the backend's IP:port, and the count tracks request volume. No backend change. How do you triage and mitigate?