I am running an EKS 1.29 cluster with managed nodegroups along with cluster-autoscaler.

The nodegroup that I use to run my k8s indexed jobs scales down to zero, when there are no jobs. One node can accomodate upto 4 jobs, enforced using resources.requests of Pod spec.

I have to run a job with parallelism: 100, completions: 100 and completionMode: Indexed with an added complexity. I want Pods to be scheduled according to job-completion-index values. In this case, Pods with job-completion-index: 0-3 should be scheduled on same node, Pods with job-completion-index: 4-7 should be scheduled on same node and so on.

How can I do this?