kubernetes-sigs/kueue

Kueue (part of kubernetes-sigs) has established itself as the standard for 'Job Queueing' in the Kubernetes ecosystem. Unlike older projects like Volcano or Yunikorn, which often replace the core Kubernetes scheduler, Kueue works *with* the default scheduler by controlling job admission. This 'K8s-native' approach makes it significantly easier to adopt and maintain. With over 2,400 stars and 570+ forks, it has high velocity and institutional backing (specifically from Google and the wider K8s SIG-Scheduling community). Its defensibility stems from its status as a SIG-sponsored project, which creates a massive gravity well for contributions and integrations (e.g., JobSet, Ray, MPIJob, and PyTorchJob all have native Kueue support). While cloud providers like GCP are already integrating Kueue into their managed offerings (e.g., GKE Job Management), this is a form of adoption rather than displacement. The risk from frontier labs is low because this is plumbing they would rather use (to manage their own massive training clusters) than reinvent. The primary threat would be a fundamental shift in how Kubernetes handles batch workloads at the core level, which is unlikely given the current trajectory.