Restructure kubectl plugin execution

[twz123]

Description

This PR will enable better integration with kubectl plugins and will prepare k0s for KEP-3638. See #2879.

K0s used the NewDefaultKubectlCommandWithArgs function to obtain a kubelet command to be embedded into k0s. That function merely executes its plugin logic unconditionally, i.e. it won't wait until that command is being executed by Cobra. This lead to a situation in which kubectl plugins not only shadowed kubectl commands, but also k0s commands. An inttest was added to prove that, before this change, an executable named kubectl-airgap would effectively replace the builtin k0s airgap command.

Address this by using a two-pass approach when executing the embedded kubectl command. First, during the Cobra bootstrap phase, call NewKubectlCommand to create a Cobra command without triggering the plugin execution logic. Then patch the command's callbacks, so that when the command is actually invoked by Cobra, the NewDefaultKubectlCommandWithArgs function is called. That function will execute plugins and never return, if any. Otherwise it will just return a new Cobra command, which k0s will simply discard and use the previously created one.

Also don't print kubectl usage on unknown subcommands. The basic kubectl command will never accept any arguments. This will be handled more or less automatically by Cobra when used as a root command. When run as a subcommand, Cobra will instead pass all of the arguments to the command's run methods, which will trigger some unexpected output. Address this by specifying cobra.NoArgs for the kubectl command.

Reorder persistent flag registration for k0s kc, so that they are grouped together. Also add InitLogs/FlushLogs, which is done by k8s.io/component-base/cli, which k0s doesn't use.

Streamline kubectl integration test: The common test matrix wasn't checking for stderr contents so far. Address this by capturing both stdout and stderr when executing the test commands and forward those along with the execution error to the check function. Add asserts for stderr contents to all the tests in the matrix. Moreover, the integration test environment wasn't completely isolated. For example, the support-bundle plugin that's installed in the footloose image was leaking into the test setup. Address this by setting up a private PATH and use that when invoking the test commands. That way, the plugin-list and symlink-warning test cases can be integrated into the common test matrix as well.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update

How Has This Been Tested?

• Manual test
• Auto test added

Checklist:

• My code follows the style guidelines of this project
• My commit messages are signed-off
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published in downstream modules
• I have checked my code and corrected any misspellings

[mikhail-sakhnov]

Side note: I was totally sure we use component-base/cli, I wonder where does this false memory come from

[twz123]

I digged a bit in the GitHub history when I wrote that comment and found a superseeded PR of yours. That might be the reason:

• Use k8s.io/base-component/cli for k0s kubectl command #1408