What could cause a high upstream_response_time (55 seconds) when the request completion time in the upstream logs is very low (12 ms)? Multiple requests are facing the same issue intermittently under high load, and it also delays the NGINX health checks.