fix: Only consider PodGroup running if at least minimum pods in running state

[jeffreyftang]

No description provided.

[linear]

PUX-912 Volcano considers PodGroups as Running prematurely

The move to the Running phase only checks for successful allocation of resources to the PodGroup:

// If there're enough allocated resource, it's running
if int32(allocated) >= jobInfo.PodGroup.Spec.MinMember {
	status.Phase = scheduling.PodGroupRunning
}

This doesn't account for errors or delays (e.g., large image pulls) that could occur between pod allocation and pod startup. Since we rely on the Running status of the PodGroup as a signal that the engine is ready for use, this can lead to timeouts and other bad outcomes in some cases.

This seems to just be a bug in Volcano.

[tgaddair]

Nice! LGTM! Could be a good one to upstream.