gthread workers hang at worker_connections limit with keepalive connections

Workers using gthread become unresponsive when nr_conns reaches worker_connections limit while futures queue is empty, particularly during high concurrent load with keepalive connections. Worker stops polling for new connections and processing requests, entering infinite loop.

Increase worker_connections configuration value to handle concurrent keepalive connections. Monitor for workers that stop processing requests while maintaining active connections. Restart unresponsive workers as immediate mitigation. Review keepalive timeout settings to reduce connection retention.