Comment 0 for bug 1771517

From this commit [1] nova now needs access to keystone to perform quota (this bug is mostly related to issue we had with quota update).

When keystone is not available the nova-api (running in eventlet) tries to use the endpoints ordered in [keystone]/valid_interfaces, we did not have access to the internal endpoint which caused this issue:

2018-05-14 15:54:46.134 1241 INFO nova.api.openstack.identity [req-8b383cf0-7f99-41e6-9de3-5e694fb24449 f13940ac09924d8582fe6612e838c7a7 9387d3a7be2a487784a90660b6e182cb - default default] Unable to contact keystone to verify project_id

You'll also see:

2018-05-14 15:54:46.419 1241 INFO nova.osapi_compute.wsgi.server [req-e9da4d33-05be-42fe-891d-0d201d2e8311 83e8a17bf7874682a86f9aa58f4c9507 e83ea76e472f48679f6fa6070a8a16e1 - default default] Traceback (most recent call last):
 File "/usr/lib/python2.7/dist-packages/eventlet/wsgi.py", line 512, in handle_one_response
   write(b''.join(towrite))
 File "/usr/lib/python2.7/dist-packages/eventlet/wsgi.py", line 453, in write
   wfile.flush()
 File "/usr/lib/python2.7/socket.py", line 307, in flush
   self._sock.sendall(view[write_offset:write_offset+buffer_size])
 File "/usr/lib/python2.7/dist-packages/eventlet/greenio/base.py", line 385, in sendall
   tail = self.send(data, flags)
 File "/usr/lib/python2.7/dist-packages/eventlet/greenio/base.py", line 379, in send
   return self._send_loop(self.fd.send, data, flags)
 File "/usr/lib/python2.7/dist-packages/eventlet/greenio/base.py", line 366, in _send_loop
   return send_method(data, *args)
error: [Errno 104] Connection reset by peer

Now this is correct, however what happens next is imo not correct, it generates a 200 OK response when it actually failed to perform the requested action.

2018-05-14 15:54:46.420 1241 INFO nova.osapi_compute.wsgi.server [req-e9da4d33-05be-42fe-891d-0d201d2e8311 83e8a17bf7874682a86f9aa58f4c9507 e83ea76e472f48679f6fa6070a8a16e1 - default default] ::ffff:195.74.38.54,172.20.104.11 "PUT /v2/e83ea76e472f48679f6fa6070a8a16e1/os-quota-sets/e83ea76e472f48679f6fa6070a8a16e1 HTTP/1.1" status: 200 len: 0 time: 128.0125880

For us we were able to notice this with 504 gateway error because the time the request took (128 seconds) was too long for our load balancer to allow.

I think atleast catching the exception and setting the return code to 500 would be appropriate, and also output it as an error and not a INFO message.

[1] https://github.com/openstack/nova/commit/1f120b5649ba03aa5b2490a82c08b77c580f12d7