Support TrainJob ResourcePerNode in CoScheduling plugin

[tenzen-y]

What you would like to be added?

I would like to consider the TrainJob .spec.trainer.resourcePerNode when calculating total resource requests for PodGroup in coscheduling plugin.

https://github.com/tenzen-y/trainer/blob/b36fe46dd46fcac5754ab5ed4bbc552ac5bb8d5a/pkg/runtime/framework/plugins/coscheduling/coscheduling.go#L122-L132

Why is this needed?

In the current coscheduling implementation, the PodGroup does not work well or causes a deadlock when they specify the .spec.trainer.resourcePerNode in the TrainJob.

Love this feature?

Give it a 👍 We prioritize the features with most 👍

[tenzen-y]

/assign

[tenzen-y]

/area controller

[Harshal292004]

@tenzen-y are you working on this ?

[Harshal292004]

/assign

[Harshal292004]

@tenzen-y May you please look into the PR I raised for this issue.

[tenzen-y]

This is blocked by others.

/unassign @Harshal292004