So this day has come - your production server gets out of memory and you didn't notice millions of requests. After quick investigation you see that your ruby processes are just too fat(300MB? 500MB? 900MB?). You may wonder that some of the requests causes memory leaks.
If you're using passenger the quick workaround (just to minimize effects for now) could be killing worker after it grows too much: PassengerMaxRequests
Still we don't know what caused that. You might try to search for long requests and assume that they the one but long response doesn't need to mean memory leak.
You can easily count the number of objects created during the request.
If you are using ruby 2.1 you can gather more detailed information: http://stackoverflow.com/questions/20956401/how-do-i-track-down-a-memory-leak-in-my-ruby-code
