CraneSched

A distributed scheduling system for HPC and AI workloads — built for performance, scale, and simplicity.

Get started Try the demo GitHub

---

Why CraneSched?

• Performance

  ---

  Over 100k scheduling decisions per second with fast job–resource matching.

• Scalability

  ---

  Proven design for million-core clusters and large-scale deployments.

• Usability

  ---

  Clean, consistent CLI for users and admins (cbatch, cqueue, crun, calloc, cinfo…).

• Security

  ---

  RBAC and encrypted communication out of the box.

• Resilience

  ---

  Automatic job recovery, no single point of failure, fast state restoration.

• Open Source

  ---

  Community-driven and extensible with a pluggable architecture.

---

Quick Start

• Deploy Backend (Rocky Linux 9)

  ---

  Recommended for production.

  Open guide →

• Configure Cluster

  ---

  Database, partitions, nodes, and policies.

  Database • Config

• Deploy Frontend

  ---

  User tools and services (CLI, cfored, cplugind).

  Open guide →

• Run Your First Job

  ---

  Batch: cbatch • Interactive: crun, calloc

---

Architecture

CraneSched introduces a Resource Manager to support both HPC and AI workloads:

• HPC jobs: the Cgroup Manager allocates resources and provides cgroup-based isolation.
• AI jobs: the Container Manager uses Kubernetes for resource allocation and container lifecycle management.

---

CLI Reference

• User commands: cbatch, cqueue, crun, calloc, cinfo
• Admin commands: cacct, cacctmgr, ceff, ccontrol, ccancel
• Exit codes: reference

---

Links

• Demo cluster: https://hpc.pku.edu.cn/demo/cranesched
• Backend: https://github.com/PKUHPC/CraneSched
• Frontend: https://github.com/PKUHPC/CraneSched-FrontEnd

---

License

CraneSched is dual-licensed under AGPLv3 and a commercial license. See LICENSE or contact mayinping@pku.edu.cn for commercial licensing.