Tempest is designed to be useful for a large number of different environments. This includes being useful for gating commits to OpenStack core projects, being used to validate OpenStack cloud implementations for both correctness, as well as a burn in tool for OpenStack clouds.
As such Tempest tests come in many flavors, each with their own rules and guidelines. Below is the proposed Havana restructuring for Tempest to make this clear.
Each of these directories contains different types of tests. What belongs in each directory, the rules and examples for good tests, are documented in a README.rst file in the directory.
API tests are validation tests for the OpenStack API. They should not use the existing python clients for OpenStack, but should instead use the tempest implementations of clients. This allows us to test both XML and JSON. Having raw clients also lets us pass invalid JSON and XML to the APIs and see the results, something we could not get with the native clients.
When it makes sense, API testing should be moved closer to the projects themselves, possibly as functional tests in their unit test frameworks.
Scenario tests are complex “through path” tests for OpenStack functionality. They are typically a series of steps where complicated state requiring multiple services is set up exercised, and torn down.
Scenario tests should not use the existing python clients for OpenStack, but should instead use the tempest implementations of clients.
Stress tests are designed to stress an OpenStack environment by running a high workload against it and seeing what breaks. The stress test framework runs several test jobs in parallel and can run any existing test in Tempest as a stress job.
Many openstack components include 3rdparty API support. It is completely legitimate for Tempest to include tests of 3rdparty APIs, but those should be kept separate from the normal OpenStack validation.