Update katib manifests

[texasmichelle]

Following the public docs, I ran into the error reported here, which was patched in kubeflow/katib.

Related changes need to be made in this repo. What is the logic behind storing manifests in both places? How have we kept them in sync in the past and which reflects the source of truth?

[texasmichelle]

After applying the patch, I'm still not seeing any metrics in the UI. @gaocegege @YujiOshima Can you help identify the problem? This is from the v0.3-branch.

[texasmichelle]

Definitions for workerConfigMap in kubeflow/kubeflow & kubeflow/katib are also out of sync.

These discrepancies are preventing me from getting any of the examples to run.

[texasmichelle]

Also, I'm seeing some objects created in the kubeflow namespace and others in katib - is that intended behavior for our default install?

[richardsliu]

@YujiOshima Can you look into this? Thanks.

[richardsliu]

Looks like we have CRD duplication across other components as well, for example tf-operator.

I think we need to add an e2etest for katib in kubeflow/kubeflow, since we are installing katib by default. The process should be:

1. Make your changes in kubeflow/katib;
2. Push Docker images (automatically done after post submits);
3. Upgrade the manifest on kubeflow/kubeflow:
   • Update the image tag on katib componets
   • If needed, also update the libsonnet files
   • Make sure the e2e test passes.

That way we can at least make sure that the main repo components work out of the box.

[richardsliu]

Also for the examples, we should change all the namespaces to kubeflow instead of katib, since the former is used in all the public documentations.

[YujiOshima]

In v0.3-branch, the image of studyjob-controller is katib/studyjob-controller.
https://github.com/kubeflow/kubeflow/blob/v0.3-branch/kubeflow/katib/prototypes/all.jsonnet#L15

This is the latest code version of Katib.
It occurs mismatch API between other components since others using gcr.io/kubeflow-images-public/katib/ ... :v0.1.2-alpha-45-g3dce496 version.

Please note the {{.StudyId}}, {{TrialId}}, and {{.WorkerId}} should not be capitalize in this version.

[texasmichelle]

Is the fix for v0.3-branch to revert the use of the untagged image to katib/studyjob-controller:v0.1.2-alpha-45-g3dce496 or to use a newly generated tag consistently for every image in all.jsonnet?

[texasmichelle]

Created katib-245 for changing katib namespace -> kubeflow in all manifests, not just examples.

Created website-293 for the corresponding docs update.