Neutron external sanity checks
Recently, a new helper script was added to neutron to facilitate testing whether or not a running system supports all of the Neutron configuration options that have been enabled. Before this script, ad hoc code kept being added to Neutron to check component version numbers at runtime. Due to the increasing number of features added and the brittleness inherent in testing version numbers for feature availability, this functionality was moved into an external check that could be run once at deployment time and specifically test for system suitability by trying to use the required features. For example, if VXLAN tunnels are enabled, the script can be run to verify that the installed version of Open vSwitch is capable of creating VXLAN ports by actually trying to create the port and removing it.
Checks can be run either by passing CLI arguments to test a system for specific feature support, or by passing in config files from the system and automatically testing each of the enabled features covered by the test script.
For example, to test specifically for Open vSwitch VXLAN support from the CLI:
To run all tests that map to enabled configuration options, pass in the config-file options for each of the configs to be tested. For example:
Adding new tests
To add a new test, first define the option in neutron/cmd/sanity_check.py in the OPTS variable using the existing options as a guide. The BoolOptCallback is just like a standard BoolOpt, but also allows you to pass a callback function which actually defines the test that should be run by the script.
Next, add a check to enable_tests_from_config() to override the CLI option for the test defined in OPTS to True based on applicable config values.
Next, add the new test to neutron/cmd/sanity/checks.py. Tests should test the actual feature required and not rely on version numbers. Tests should also return the system state to the way it was before the test was run. Return value should be True for passing, False for failing.
Next, go back to neutron/cmd/sanity_check.py and add the callback function that was defined in OPTS. This function should take no arguments and should log a helpful error message on failure. If testing via config options is supported, it should use the cfg.CONF option that the test depends on. The return value should be True for passing, False for failing. The reason this function is separate from the neutron/cmd/sanity/checks.py version is that the tests in sanity_check.py must all have the same arguments and generally will use values from cfg.CONF, whereas for testing purposes other argument values may be necessary.
Finally, define a functional test in neutron/tests/functional/sanity/test_ovs_sanity.py which ensures that they defined check function runs through to completion. This will alert developers if an API change is made that breaks the sanity check functions. For example: