Add inftyai-scheduler support and config updates

[carlory]

What this PR does / why we need it

Add inftyai-scheduler support and config updates

Test it with a custom scheduler name.

(base) ➜  llmaz git:(scheduler-plugins) kubectl get cm llmaz-global-config demo-inftyai-scheduler-scheduler-config -oyaml | grep sched
    config.data: 'scheduler-name: demo-inftyai-scheduler'
    scheduler-config.yaml: |
      apiVersion: kubescheduler.config.k8s.io/v1
        resourceName: demo-inftyai-scheduler
      - schedulerName: demo-inftyai-scheduler
    name: demo-inftyai-scheduler-scheduler-config
(base) ➜  llmaz git:(scheduler-plugins) kubectl get po qwen2-0--5b-0 -oyaml | grep sched
  schedulerName: demo-inftyai-scheduler

Which issue(s) this PR fixes

Fixes #

Special notes for your reviewer

Does this PR introduce a user-facing change?

 Add inftyai-scheduler support and config updates

[kerthcet]

Better not to add too much logic to the makefile command, what's the advantages over the current one?

[carlory]

Because the default value is default-scheduler in config/default/configmap.yaml and it can not be easily changed by user via helm install --set command.

I can not edit chart/template/global-config.yaml because it is generated by helmify. I want users to be able to easily use the chart without extra configuration.

[kerthcet]

But it's not a general solution, we have a lot of configurations.

[kerthcet]

I'm ok to let user configure the configmaps via global.values.yaml. It's also not difficult.

[carlory]

Can we switch the default scheduler to inftyai-scheduler in config/default/configmap.yaml? And add make install-inftyai-scheduler and make uninstall-inftyai-scheduler to the Makefile.

install-inftyai-scheduler:
	kubectl apply --server-side -k config/inftyai-scheduler

.PHONY: uninstall-prometheus
uninstall-inftyai-scheduler:
	kubectl delete -k config/inftyai-scheduler

[carlory]

If not, we can not enable the inftyai-scheduler in chart by default due to the immutability of the configmap which is generated by helmify.

[carlory]

I'm ok to let user configure the configmaps via global.values.yaml. It's also not difficult.

This pr has supported it.

[kerthcet]

What I mean is we may have several configurations in globalconfig, so this is not a general solution, I would suggest revert to the original configurations, and people can modify the chart/values.global.yaml before installation/upgrade, does that workaround?

[carlory]

It works. Howerver, if an user want to use the scheduler, he have to do the following:

1. Get values from chart via helm show values oci://registry-1.docker.io/inftyai/llmaz > /t mp/values.yaml

2. Modify the values.yaml to enable the scheduler like this:

kube-scheduler:
  enabled: true

globalConfig:
  configData: |-
    scheduler-name: inftyai-scheduler
    # init-container-image: inftyai/model-loader:v0.0.10

3. Install the chart with the modified values.yaml.

[carlory]

We can not add this config to values.global.yaml in codebase because make helm-package appends (not merge) the file to values.yaml.

[carlory]

Another drawback is that users need to provide the whole content of the configData when they want to change only one field because the globalConfig.configData is a string, not a map.

[kerthcet]

let's merge for now and optimize it later.
/lgtm
/approve
/kind cleanup

Thanks @carlory