Use a couple of environment variables to figure out what wsgi app to really run in nginx_mod_wsgi.py. Now you can run any wsgi app you want easily in nginx. Also, delay calling start_response until the first hunk of the body is sent since nginx doesn't let us change the response codes or headers after the first time it is called.

eventlet/support/nginx_mod_wsgi.py

@@ -1,5 +1,9 @@
 
 import sys
+import traceback
+
+sys.path.insert(0, '/Users/donovan/Code/mulib-hg')
+sys.stdout = sys.stderr
 
 from eventlet import api
 from eventlet import httpc
@@ -7,14 +11,15 @@ from eventlet import httpc
 from eventlet.hubs import nginx
 
 
-def real_application(env, start_response):
+def old_real_application(env, start_response):
     #result = httpc.get('http://127.0.0.1:8081/')
     start_response('200 OK', [('Content-type', 'text/plain')])
     #sys.stderr.write("RESULT %r" % (result, ))
-    return 'hi'
+    return 'hello'
 
 
 def wrap_application(master, env, start_response):
+    real_application = api.named(env['eventlet_nginx_wsgi_app'])
     result = real_application(env, start_response)
     ## Should catch exception and return here?
     #sys.stderr.write("RESULT2 %r" % (result, ))
@@ -22,6 +27,14 @@ def wrap_application(master, env, start_response):
     return None, None
 
 
+class StartResponse(object):
+    def __init__(self, start_response):
+        self.start_response = start_response
+
+    def __call__(self, *args):
+        self.args = args
+
+
 def application(env, start_response):
     hub = api.get_hub()
 
@@ -34,17 +47,21 @@ def application(env, start_response):
     hub.current_application = api.getcurrent()
 
     slave = api.greenlet.greenlet(wrap_application)
+    response = StartResponse(start_response)
     result = slave.switch(
-        hub.current_application, env, start_response)
+        hub.current_application, env, response)
 
     while True:
         #sys.stderr.write("RESULT3 %r" % (result, ))
         if result is None or result == (None, None):
             yield ''
         else:
+            start_response(*response.args)
             if isinstance(result, tuple):
-                yield result[0]
+                for x in result[0]:
+                    yield x
             else:
-                yield result
+                for x in result:
+                    yield x
             return
         result = hub.switch()