We are using nginx as a web/comet server to provide push notification to our web-app. We have been running fine on nginx 1.0.11 and nginx_http_push_module verson 0.692, on ubuntu 10.04 servers, for years without issue, but re recently upgraded to nginx 1.4.6 and push module .712, on 12.04. We are seeing an issue when the GET call to the subscriber location on nginx is interrupted it creates a CLOSE_WAIT connection that is never cleared up, but once the GET call resolves once it clears up any connections that were created on that publish channel. In our old set up these connections were being cleaned up after several hours but once we switched over to the new server they persist until nginx is restarted. Is there some way I can get the server to clean these connections as before, or close the connections explicitly in the error state of the call?