We were able to trace down a problem that is crashing our NGINX server running Magento until the following point:

Background info: Magento Backend has a CMS function with a WYSIWYG editor. This editor loads some pictures via a controller in magento (cms/directive).

When we set the NGINX error_log level to info, we get the following lines (line break inserted for better readability):

2012/10/22 18:05:40 [info] 14105#0: *1 client closed prematurely connection, 
  so upstream connection is closed too while sending request to upstream, client: 
  XXXXXXXXX, server: test.local, request: "GET 
  index.php/admin/cms_wysiwyg/directive/___directive/BASEENCODEDIMAGEURL,,/ 
  HTTP/1.1", 
  upstream: "fastcgi://127.0.0.1:9024", host: "test.local" 

When checking the code in the debugger, the following call does never return (in ´Varien_Image_Adapter_Abstract::getMimeType()`

# $this->_fileName is http://test.local/skin/adminhtml/base/default/images/demo-image-not-existing.gif` 
# $_SERVER['REQUEST_URI'] = http://test.local/admin/cms_wysiwyg/directive/___directive/BASEENCODEDIMAGEURL

list($this->_imageSrcWidth, $this->_imageSrcHeight, $this->_fileType, ) = getimagesize($this->_fileName);

The filename requests is

• an URL to the same server which is requesting the script
• a link to a static .gif that is not existing.

Sample URL:

http://test.local/skin/adminhtml/base/default/images/demo-image-not-existing.gif

When the above line executed, any subsequent request to the NGNIX server does not respond any more. After waiting for around 10 minutes, the NGINX server starts answering requests again.

I tried to reproduce the error with a simple test script that only calls getimagesize() with the given URL - but this not crash. It simple leads to an exception saying that the URL could not be loaded (which is fine as the URL is wrong)