Ray Autoscaler integrate with Ray K8s Operator

[Qstar]

Why are these changes needed?

1.fix the problem falling behind in terms of scaling features; autoscaling not up to par with other Clouds

Main modifications:

• Add Operator node provider.
• Update logic of Ray Operator.

Related issue number

Checks

• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[zhe-thoughts]

Thanks for the contribution @Qstar. Is this on top of #1 ? cc @raulchen @caitengwei

[akanso]

Thanks for the contribution @Qstar. Is this on top of #1 ? cc @raulchen @caitengwei

@zhe-thoughts we based #1 (shown here: https://github.com/akanso/ray-contrib/tree/design-1b/msft-operator/ray-operator) on top of the schema in #2 , and then we modified it to use pod template instead of extensions as per the discussions in this google doc

[Qstar]

Thanks for the contribution @Qstar. Is this on top of #1 ? cc @raulchen @caitengwei

This PR(#2) is for integrating Autoscaler with Operator( already in the existing repo), and the PR(#1) makes modifications based on PR(#2) (Please correct me if I'm wrong @akanso ).

[caitengwei]

@zhe-thoughts we plan to merge #2 first, because we're using the autoscaler & operator integration in our internal scenario. And we'll collaborate with @akanso based on #1 to migrate our implementation and autoscaler integration. Is that ok to you?

[zhe-thoughts]

Thanks for explaining this @caitengwei @Qstar @akanso. The plan sounds good to me. @ericl @edoakes thoughts?