--kubeconfig flag not honored

[rm3l]

/kind bug

What versions of software are you using?

Operating System:
Fedora release 36 (Thirty Six)

Output of odo version:
odo v3.0.0-alpha1 (b7e45da)

How did you run odo exactly?

❯ odo dev --kubeconfig /path/to/a/kubeconfig/other/than/the/default

This always uses the default config, not the one specified by the user. As such, it uses the wrong cluster.

Actually, this affects all commands that expose this flag, like odo deploy or odo list, e.g.:

❯ odo list --kubeconfig /path/to/a/kubeconfig/other/than/the/default -v 4                                                                                                                                                                                         
I0511 16:41:56.599645 1765751 implem.go:103] The path for preference file is /home/asoro/.config/odo/preferences.yaml                                                                                                                                                            
I0511 16:41:56.599700 1765751 util.go:733] HTTPGetRequest: https://raw.githubusercontent.com/redhat-developer/odo/main/build/VERSION                                                                                                                                             
I0511 16:41:56.599724 1765751 segment.go:254] Checking telemetry enable status                                                                                                                                                                                                   
I0511 16:41:56.599767 1765751 util.go:754] Response will be cached in /tmp/odohttpcache for 1h0m0s
                                                                                                                                                                               
I0511 16:41:56.602227 1765751 util.go:1077] using default kubeconfig path /home/asoro/.kube/config

--- truncated ---

Actual behavior

Almost all commands currently expose this --kubeconfig flag in their help message (under the "Additional Flags" section):

For example, with odo help dev:

Additional Flags:
      --kubeconfig string    Paths to a kubeconfig. Only required if out-of-cluster.
  -v, --v Level              Number for the log level verbosity. Level varies from 0 to 9 (default 0).
      --vmodule moduleSpec   Comma-separated list of pattern=N settings for file-filtered logging

Maybe I missed something, but I was expecting this --kubeconfig option to allow me to target a different cluster. Instead, it seems like odo always reads the default ${HOME}/.kube/config file.

After a quick look, it looks like:

• we always initialize a kclient with the defaults:

  odo/pkg/kclient/kclient.go

  Lines 70 to 73 in b7e45da

  	// New creates a new client
  	func New() (*Client, error) {
  	return NewForConfig(nil)
  	}

• in odo list, we always check either the KUBECONFIG env var is set or the default ~/.kube/config does exist :

  odo/pkg/odo/cli/list/list.go

  Lines 69 to 75 in b7e45da

  	// Check to see if KUBECONFIG exists, and if not, error the user that we would not be able to get cluster information
  	// Do this before anything else, or else we will just error out with the:
  	// invalid configuration: no configuration has been provided, try setting KUBERNETES_MASTER environment variable
  	// instead
  	if !dfutil.CheckKubeConfigExist() {
  	return errors.New("KUBECONFIG not found. Unable to retrieve cluster information. Please set your Kubernetes configuration via KUBECONFIG env variable or ~/.kube/config")
  	}

Expected behavior

Make odo honor the user-defined kubeconfig file(s) defined via this flag, for all interactions with a cluster.

Workaround: Set the KUBECONFIG environment variable instead, but it would be nice to make this --kubeconfig option work (taking precedence over the environment variable).

Or if this is not something expected, we should probably remove this flag from the Help messages.

[feloy]

I think that CheckKubeConfigExist should not be used as is, as it does not support KUBECONFIG values with several kubeconfig paths:

KUBECONFIG=/path/to/config:/other/path/to/config

Also, we should rely on genericlioptions.New to do this work, and not do it in the operation specific functions.

[kadel]

That flag is coming from controller-runtime

odo/vendor/sigs.k8s.io/controller-runtime/pkg/client/config/config.go

Lines 38 to 40 in e32024f

	// TODO: Fix this to allow double vendoring this library but still register flags on behalf of users
	flag.StringVar(&kubeconfig, "kubeconfig", "",
	"Paths to a kubeconfig. Only required if out-of-cluster.")

It was not my intention to have it there.

We should figure out a way how to remove it from odo. And then, as a separate issue, have it properly implemented.