Adoption & Ecosystem
Who uses it
The organisations below are listed as production users in the Volcano community adopters file. Use cases are not individually detailed there beyond the production designation.
| Organisation | Use case | Source |
|---|---|---|
| Huawei Cloud | Production batch scheduling | adopters.md |
| Tencent | Production batch scheduling | adopters.md |
| Baidu | Production batch scheduling | adopters.md |
| IQIYI | Production batch scheduling | adopters.md |
| Xiaohongshu | Production batch scheduling | adopters.md |
| DiDi | Production batch scheduling | adopters.md |
| iFlytek | Production batch scheduling | adopters.md |
| Kingsoft Cloud | Production batch scheduling | adopters.md |
| Zoom | Production batch scheduling | adopters.md |
The list also names Bosszhipin, Ruitian Capital, and Momenta among production users.
Adoption signals
Measured from the GitHub API on 2026-06-25:
- Stars: 5,699; forks: 1,415; open issues: 657.
- Contributors: the GitHub contributors API runs to roughly 447 (anonymous included).
- The CNCF incubation blog reports growth from 70+ to 350+ contributors and 5 to 50+ organisations between Sandbox acceptance and Incubation.
Ecosystem
Volcano integrates with batch and AI frameworks including Spark, Flink, Ray, PyTorch, TensorFlow, MindSpore, PaddlePaddle, Kubeflow, MPI/Horovod, Argo, and KubeGene, per the README. Apache Spark adopted Volcano as a built-in batch scheduler on Kubernetes (CNCF Spark blog). It supports heterogeneous devices such as GPU and NPU. For multi-cluster scheduling there is a separate volcano-sh/volcano-global federation scheduler.
Alternatives
| Alternative | Differs by |
|---|---|
kube-scheduler (default) | Schedules one pod at a time with no gang, queue, or fair-share. Volcano replaces or complements it for batch workloads. |
kube-batch | The predecessor that Volcano's scheduler is built on; now largely inactive, with Volcano as its successor. |
| Apache YuniKorn (CNCF) | Also a batch/queue scheduler for Kubernetes, centred on hierarchical queues. Volcano differentiates with the VolcanoJob CRD lifecycle and a wider plugin set (NUMA, device, topology). |
| Kueue (Kubernetes SIG) | Focuses on job queueing and quota and defers placement to a scheduler. Volcano reaches down into the scheduling algorithm itself. |
| YARN and traditional HPC schedulers | Not Kubernetes-native; require a separate resource manager. |
Pick Volcano when the workload is gang-shaped and Kubernetes-native; pick Kueue when you only need queueing on top of an existing scheduler.