What it does

kube-batch is a batch scheduler for Kubernetes, designed for high-performance workloads like AI/ML, Big Data, and HPC. It sits alongside the default scheduler to handle gang scheduling, queueing, and resource sharing for jobs that need to run as coordinated groups rather than scattered pods.

The interesting bit

The project is officially retired by Kubernetes, yet its DNA lives on: both Kubeflow and Volcano — two of the most active batch-scheduling ecosystems in Kubernetes — started here. It’s a rare case of a project that succeeded itself into obsolescence.

Key highlights

• Born from “a decade and a half of experience” running batch workloads at scale (the README’s claim, not independently verifiable)
• Adopted by Baidu, TuSimple, Vivo, and others before retirement
• Explicitly scoped: handles core scheduling mechanisms, leaves “out-of-scope” parts to other projects
• Community channel still points to active Kubernetes SIG Scheduling

Caveats

• Repository is retired/archived; no ongoing development
• README is thin on technical specifics — no benchmarks, no detailed feature list, no comparison to default scheduler
• Architecture diagram exists but isn’t explained in text

Verdict

Worth studying if you’re building a Kubernetes scheduler or tracing the lineage of Volcano/Kubeflow. Skip it if you need something production-ready today — use Volcano instead.