To pitch in some preachy advice: with some time and effort, staging and production environments should be essentially identical. Sure, production is generally going to have more servers handling each service (more web nodes, more db/storage nodes, multiple load balancers, etc.), but the setup and deployments should not be any different.

All too often, we think of staging as nothing more than a clone of the development environment on a remote server where QA can get at it. If you run your load balancer (haproxy), web server (apache, nginx), database (mysql, postgres, mondo, elasticsearch), caching (memcached, redis) all on a single server for staging, you're eventually going to stumble on this kind of hard-to-diagnose problem.

One of the main disservices you are doing to yourself with a single-server staging is that all of your traffic is going to be travelling over localhost or, worse, over unix sockets. You're not even testing basic network latency or performance.

Staging should really be considered a first-level production. Its configuration and maintenance should be handled with the same attention to detail as production.