Goal: 

As a developer, I want run some level of scale testing by using the fake virt driver, so I can fine tune the thread pool sizes of various nova services running in threading mode.

Acceptance Criteria:

• The CI job created in OSPRH-21006 is extended with more scenarios (see below)
• A blog post or other form of test report exists that details what measurement was done and what suggestions we have about the scaling of our thread pools or other algorithms.
• If the scale testing finds bugs then those bugs are reported upstream.
• If the scale testing suggests different pool size defaults then those opened as bugs as well upstream.

The test report should contain:

• a set of test scenarios
• baseline performance measurement (runtime, memory usage) of those scenarios with eventlet
• performance measurement of the scenarios with threading mode enabled on all nova services that supports them

Scenarios to cover:

• multi cell and multi compute setup checking scheduling performance (scheduler scatter gather) Split out to OSPRH-21006
• multi cell and multi compute setup checking instance listing performance (nova api scatter gather)
• parallel VM lifecycle operatos (like reboot) on multiple VMs on the same compute (compute RPC handling)
• multi compute setup checking novncproxy scaling with multiple long living connections as well as with connection churn
• conductor performance of caching images across multiple cells and computes
• ... TBD

Known limitations:

• Using fake driver means we cannot test nova-compute - libvirt interactions.

This story complements but does not replaces scale testing with real hardware

• https://issues.redhat.com/browse/RHOSSTRAT-788