Hi all,

So I have an nginx server that's working over https with Sinatra. When I try to download a jnlp file in a configuration that works fine over Mongrel and http (no s), the nginx server fails to serve the file with a 504 error. Subsequent checking of the logs states that this error is due to overflowing the available number of file handles, ie, "24: too many open files". Running

sudo lsof -p <nginx worker pid>

gets me a huge list of files, all looking like:

nginx   1771 nobody   11u     IPv4           10867997         0t0      TCP localhost:44704->localhost:https (ESTABLISHED)
nginx   1771 nobody   12u     IPv4           10868113         0t0      TCP localhost:https->localhost:44704 (ESTABLISHED)
nginx   1771 nobody   13u     IPv4           10868114         0t0      TCP localhost:44705->localhost:https (ESTABLISHED)
nginx   1771 nobody   14u     IPv4           10868191         0t0      TCP localhost:https->localhost:44705 (ESTABLISHED)
nginx   1771 nobody   15u     IPv4           10868192         0t0      TCP localhost:44706->localhost:https (ESTABLISHED)
nginx   1771 nobody   16u     IPv4           10868255         0t0      TCP localhost:https->localhost:44706 (ESTABLISHED)
nginx   1771 nobody   17u     IPv4           10868256         0t0      TCP localhost:44707->localhost:https (ESTABLISHED)
nginx   1771 nobody   18u     IPv4           10868330         0t0      TCP localhost:https->localhost:44707 (ESTABLISHED)
nginx   1771 nobody   19u     IPv4           10868331         0t0      TCP localhost:44708->localhost:https (ESTABLISHED)
nginx   1771 nobody   20u     IPv4           10868434         0t0      TCP localhost:https->localhost:44708 (ESTABLISHED)

Increasing the number of files that can be opened is no help, because then nginx just blows right past that limit. And no wonder, it looks like it's in some kind of loop to pull all available files.

Any idea what's going on, and how to fix it?

EDIT: nginx 0.7.63, ubuntu linux, sinatra 1.0

EDIT 2: Here's the offending code. It's sinatra serving jnlp, which I finally figured out:

get '/uploader' do
  #read in the launch.jnlp file                                                               
  theJNLP = ""
  File.open("/launch.jnlp", "r+") do |file|
    while theTemp = file.gets
      theJNLP = theJNLP + theTemp
    end
  end                                                                    
  content_type :jnlp
  theJNLP
end

If I serve this with Sinatra via Mongrel and http, everything works fine. If I serve this with Sinatra and nginx via https, I get the above error. All other parts of the website appear to be equivalent.

EDIT: I have since upgraded to passenger 2.2.14, ruby 1.9.1, nginx 0.8.40, openssl 1.0.0a, and no change.

EDIT: The culprit appears to be infinite redirects due to using SSL. I don't know how to fix this, other than hosting the jnlp file in the root directory of the server (which I'd rather not do, since it limits me to one jnlp-based app at a time).

The relevant lines from nginx.conf:

# HTTPS server                                                                            
#                                                                                         
server {                                                       
    listen       443;                                                
    server_name   MyServer.org
    root         /My/Root/Dir;
    passenger_enabled on;
    expires           1d;

    proxy_set_header X-FORWARDED_PROTO https;
    proxy_set_header X_FORWARDED_PROTO https;#the almighty google is not clear on which to use   


   location /upload {
      proxy_pass https://127.0.0.1:443;
   }
}   

The funny thing about this is, first, I was putting the jnlp into a directory called 'uploader', not 'upload', but that still appeared to trigger the problem, since that proxy_pass directive appeared in the logs. Second, again, moving the jnlp into root avoided the problem, because there wasn't any of this proxying due to ssl.

So, how can I avoid the infinite proxy_pass loop in nginx?