Schedule AI Workloads at Scale

$ kubectl get queues
NAME              PRIORITY   PARENT          CHILDREN
department-a      medium                     team-research-a,team-inference
department-b      medium                     team-training,team-research-b
team-inference    high       department-a
team-research-a   medium     department-a
team-training     medium     department-b
team-research-b   low        department-b

$ kubectl describe queue team-research-a
Name:         team-research-a
Spec:
  parentQueue: department-a
  resources:
    cpu: { quota: 64, limit: 128, overQuotaWeight: 1 }
    gpu: { quota: 8,  limit: 16,  overQuotaWeight: 1 }
Status:
  allocated:
    cpu: 52
    gpu: 14

$ kubectl get schedulingshards
NAME      AGE
default   12d