Latency tests are described in our customer documentation here:
https://docs.redhat.com/en/documentation/openshift_container_platform/4.21/html/scalability_and_performance/cnf-performing-platform-verification-latency-tests#cnf-measuring-latency_cnf-latency-tests

If the LATENCY_TEST_CPUS parameter is not supplied, the test is skipped with the following message:
[SKIPPED] Skip the test, the requested number of CPUs should be even to avoid noisy neighbor situation

This is due to the following check:
https://github.com/openshift/cluster-node-tuning-operator/blame/7916d0fc178a08ff83794f3f5fff9779885340c4/test/e2e/performanceprofile/functests/4_latency/latency.go#L84

I think the code is attempting to use all the available CPUs minus 1, which results in an odd number and the failure to run the test.

Note that the example in the documentation for the hwlatdetect test omits the LATENCY_TEST_CPUS parameter, so any customer following these instructions will hit this issue.

Seen in OCP 4.20  

Always

1. Run a hwlatdetect test following the instructions in the customer documentation referenced above    

The test is skipped    

The test runs

We'll need to decide on an appropriate default value for LATENCY_TEST_CPUS if it is not supplied. Using all the CPUs that are available is probably not a good idea as we do not recommend running latency tests on all or most of the CPUS in a server. The other option would be to change LATENCY_TEST_CPUS to be a mandatory parameter, but that is a bit unintuitive for the hwlatdetect test, which actually runs the test on ALL CPUs (using a kernel tracer), regardless of how many CPUs the cnf-tests container is using.

Also note that the workaround is just to provide the LATENCY_TEST_CPUS parameter to the command. Given that there is a simple workaround I would recommend that we only fix this in the current release.

Also note that this will likely require a customer documentation update if we change the default value for LATENCY_TEST_CPUS - the documentation says this:
LATENCY_TEST_CPUS: Specifies the number of CPUs that the pod running the latency tests uses. If you do not set the variable, the default configuration includes all isolated CPUs.