Use HAProxy location for the host header

Since the host header defines which service the push task goes to
in App Engine, this roughly preserves the same behavior (since the
target service has already been selected at this point). There are
some edge cases (especially during service relocation) where this
behaves differently, but the routing pieces need more work to
support those differences.

AppTaskQueue/appscale/taskqueue/push_worker.py

@@ -157,24 +157,27 @@ def execute_task(task, headers, args):
       # Tasks should use HTTP to bypass scheme redirects since they use HAProxy.
       connection = httplib.HTTPConnection(remote_host, url.port)
 
-      skip_host = False
-      if b'host' in headers or b'Host' in headers:
-        skip_host = True
-
       skip_accept_encoding = False
       if 'accept-encoding' in headers or 'Accept-Encoding' in headers:
         skip_accept_encoding = True
 
       connection.putrequest(method,
                             urlpath,
-                            skip_host=skip_host,
                             skip_accept_encoding=skip_accept_encoding)
 
       # Update the task headers
       headers['X-AppEngine-TaskRetryCount'] = str(task.request.retries)
       headers['X-AppEngine-TaskExecutionCount'] = str(task.request.retries)
 
       for header in headers:
+        # Avoid changing the host header from the HAProxy location. Though GAE
+        # supports host-based routing, we need to make some additional changes
+        # before we can behave in a similar manner. Using the HAProxy location
+        # for the host header allows the dispatcher to try extracting a port,
+        # which it uses to set environment variables for the request.
+        if header == b'Host':
+          continue
+
         connection.putheader(header, headers[header])
 
       if 'content-type' not in headers or 'Content-Type' not in headers: