Heat output-show performance linearly depends on stack size

[Ilya Shakhat]

The performance of heat output-show operation depends linearly on number of items in the stack. I'd expect it to remain pretty constant since the operation shouldn't deal with the whole collection. On hardware environment the following numbers are collected:

 * 150 instances (300 outputs) -> 0.5 seconds
 * 300 instances (600 outputs) -> 1 second
 * 450 instances (900 outputs) -> 1.5 seconds

According to log, the request is accepted by Heat API service and then forwarded by RPC to Heat Engine. In Engine it is processed, but there is no information what exactly is going on:

2016-06-27T13:09:57.016442+00:00 debug: 2016-06-27 13:09:57.007 12004 DEBUG oslo_messaging._drivers.amqpdriver [-] received message msg_id: 492f6e6d55dc40a4aa609f7fa3b7f7fc size: 9268 reply to: reply_9d5be52c1aa94449b82c4e25e4b9b8c6 __call__ /usr/lib/python2.7/dist-packages/oslo_messaging/_drivers/amqpdriver.py:230

2016-06-27T13:09:58.468949+00:00 debug: 2016-06-27 13:09:58.462 12004 DEBUG oslo_messaging._drivers.amqpdriver [req-23fd156d-13bf-4f89-ba14-60099a6310db - admin - default default] REPLY msg_id: 492f6e6d55dc40a4aa609f7fa3b7f7fc size: 272 reply queue: reply_9d5be52c1aa94449b82c4e25e4b9b8c6 time elapsed: 1.452448347s _send_reply /usr/lib/python2.7/dist-packages/oslo_messaging/_drivers/amqpdriver.py:96

Steps to reproduce:
1. Prepare heat stack that boots instances and retrieves outputs (e.g. similar to this https://github.com/openstack/shaker/blob/master/shaker/scenarios/openstack/l2.hot)
2. Create the stack
3. Retrieve stack outputs via 'heat show-output' command. Check the duration

Environment: 6 HW nodes, 48 cores, 256 Gb RAM, MOS 9 ISO 495