Why it matters

Kueue is emerging K8s AI standard. Understanding shapes K8s AI ops.

Advertisement

The architecture

ResourceFlavor: GPU types.

ClusterQueue: quota.

LocalQueue: per-namespace.

Workload: gang-scheduled job.

Kueue architectureWorkloadK8s jobLocalQueuenamespace queueClusterQueuequota + policyKueue + JobSet + PyTorch Job for distributed training on K8s
Kueue.
Advertisement

How it works end to end

SIG-K8s project.

Integrates with JobSet, PyTorch Job.

Cohorts: share quota across teams.