Add tests for k0scontrolplane controller

[apedriza]

Changes

• add unit and integration tests on the k0scontrolplane controller. They use envtest to spin-up to kube-apiserver+etcd without using a cluster. In some of these tests it is enough to use fakeclient for the creation of objects and subsequent state check but since for others it is necessary to use envtest (an apiserver is needed) and in order to reduce the different ways of creation of the environment I have chosen to use envtest for any test. For the test implementation I have taken inspiration from how upstream projects do this job.

  The idea is, after agreeing on how we want this type of testing for the project, to add the rest of the tests for the other controllers.

  More detailed testing of the machine reconciliation logic can be added but since it is undergoing changes I have decided to add only basic testing for the time being.

• updates docs about testing and moving some line to the proper place

[apedriza]

Ready for review, also I added some docs changes about testing

[apedriza]

I think all this should be in contribute workflow instead testing so I moved it there

[makhov]

Could you elaborate a bit on why the logic was changed? The result looks more or less the same

[apedriza]

Yes, so basically I found a bug in the logic when running the proposed test TestReconcileMachinesSyncOldMachines. Bug comes from two places: logic calculating machinesTodDelete var + new machine name calculation.

Reproducible when:

• desired kcp replicas = 3
• current desired replicas: <machine-name>-1
• outdated replicas: <machine-name>-0 <machine-name>-2

👉 You can repro it by running new test TestReconcileMachinesSyncOldMachines with the current reconcileMachines func.

In this case, desired machine (<machine-name>-1) is added to machineToDelete variable even others machines which doesn't has same version should be delated first. In this case, <machine-name>-1 is added becauseactiveMachines.Len() > int(kcp.Spec.Replicas)+len(machineNamesToDelete) is true but others machines not updated should be removed first so condition activeMachines.Len() > int(kcp.Spec.Replicas)+len(machineNamesToDelete) has been moved out of the loop to be sure a desired machine is not added until all others machine that could be added to machineToDelete are checked.

In this case described above, machineName func was calculating a bad machine name, facing an infinite loop an returning a machine name which was already created:

• machines to delete: <machine-name>-1, <machine-name>-2
• desired machines: <machine-name>-4, <machine-name>-5
• output: <machine-name>-5 ❌ (is already desired)

but I think all this comes from the error adding the wrong machine into machinesToDelete so the fix is there and not in the naming logic.

[makhov]

Great catch!