Updated to use default Kubernetes client discovery logic

[nicknezis]

The defaultClient() call incorporates the following logic:
If $KUBECONFIG is defined, use that config file.
If $HOME/.kube/config can be found, use that.
If the in-cluster service account can be found, assume in cluster config.
Default to localhost:8080 as a last resort.

With this code change, we'd no longer be using the scheduler config item for K8s scheduler URI. But the config item is used to create Job Links. So I couldn't get rid of that config item.

Any thoughts on how I could fully remove the scheduler URI config item?
Do you think the defaultClient() code is good enough? It seemed to work in my Minikube testing and I figured worst case that last option would point to kubectl-proxy if we update it to use port 8080. I assume if someone needs to run the Heron API Server outside of the k8s cluster, then can configure with the $KUBECONFIG environment variable.

[nicknezis]

Hopefully this addresses part of what was asked for in #3259

[joshfischer1108]

I haven't looked at the code yet, but does this PR relate to the issue listed here? #3259

[nicknezis]

Yes, I tested it in Minikube and was able to spin up a StatefulSet without the kubectl-proxy. I'll update this PR to remove that container from the Kubernetes deploy yamls and Helm chart. I wanted to also remove the Scheduler URI as was suggested in #3259, but the config item was used in the Job link. But the current implementation does not generate job specific job links.

[joshfischer1108]

Ahh yeah I see you mentioned the issue right above where I mentioned the issue. Consider that a palm to forehead moment 😄 . I haven't tested the code, but It sounds like it addresses the issue after the proxy container is taken out.

[joshfischer1108]

@cckellogg If you have some time could you review this?

[joshfischer1108]

@nicknezis It looks like the build is failing because of checkstyle errors.

[nicknezis]

I think the test failure might not be related to my changes. I am able to run the tests locally with success.

[nwangtw]

I think the test failure might not be related to my changes. I am able to run the tests locally with success.

//heron/schedulers/tests/java:KubernetesSchedulerTest FAILED in 1.2s
FAILED org.apache.heron.scheduler.kubernetes.KubernetesSchedulerTest.testGetJobLinks (0.0s)

Unit tests should be normally stable. Let me trigger another build and see.

[nicknezis]

Looks like it failed again. I'll try to figure out why. Thanks!

[nicknezis]

Is this PR ok to be accepted?

[joshfischer1108]

@nicknezis LGTM. I'll wait a bit to see if anyone else has feedback.