Don't use latest stable version for minikube

[oz123]

Sometimes, minikube lags behind kubernetes which can lead
to a breakage of the test suite, since minikube will fail to start
with the latest kubernetes version.
See for example this issue.

This change defaults to leave the decision of which k8s version to use,
to minikube itself. This is defined in:

https://github.com/kubernetes/minikube/master/pkg/minikube/constants/constants.go

However, if one really desires it is still possible to pass

--kubernetes-version=X.Y.Z

to minikube initialization start command via an environment variable
before invoking the test suite:

export $MINIKUBE_ARGS="--kubernetes-version=X.Y.Z"

This allows ofcourse passing other flags to minikube also.

[micw523]

If you take a look at the CI log you'll find that the behavior your changes introduced is to download k8s 1.15 first, and then minikube installs k8s 1.14 again. This is a little redundant in my opinion - I think we could simply pin the k8s version to 1.14 for now(?)

[oz123]

Why would you want to introduce a future bug? Every-time Kubernetes changes behavior you risk minikube breaking. The "working" Kubernetes version for minikube is hardcoded in minikube, and I think you should just minikube choose it's own version.
If the python-client needs to test a specific k8s API there is still the possibility of overriding it with the environment variable.

The script downloads kubectl not kubernetes. kubectl in version n+1 works fine with version n.
After that the script downloads minikube in the stable version, which currently installs the version of kubernetes which is known to work.
If we change this to 1.14 now, someone will need in the future to change this to a version which is known to work with minikube, and this version will work until it won't.
My approach, to leave the decision to minikube is more sustainable.

Also, this documented in my commit message, but just in case, I am repeating this again. kubeadm in its current version installes k8s in version 1.14.3.

[roycaihw]

I have one nit on the link, then LGTM. We should have better mechanism to verify server-client skew, instead of tweaking minikube ad hoc

[roycaihw]

/lgtm
/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: oz123, roycaihw

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [roycaihw]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment