Alauda AI

Introduction

TOC

VolcanoDocumentation

Volcano

Alauda support for Volcano packages the Volcano CNCF project as a Cluster Plugin for Alauda Container Platform. Volcano is a cloud-native batch system for Kubernetes that provides scheduling and resource management for machine learning, big data, and HPC workloads.

Main components and capabilities include:

  • Volcano scheduler: A batch-oriented scheduler that complements the default Kubernetes scheduler. It supports gang scheduling, fair-share, binpack, and other policies that are required by distributed training and batch workloads.
  • VolcanoJob (batch.volcano.sh/v1alpha1): A job API that groups tasks with different roles (such as master and worker) and schedules them as a single unit, so a multi-pod training job either starts together or not at all.
  • PodGroup (scheduling.volcano.sh/v1beta1): A grouping primitive used by the scheduler to track quota, queue, and gang-scheduling constraints across pods that belong to the same logical job.
  • Queue (scheduling.volcano.sh/v1beta1): A resource queue that organizes jobs by priority, weight, and capability, enabling tenant-aware sharing of cluster resources.
  • JobFlow (flow.volcano.sh/v1alpha1): Declarative orchestration of multiple VolcanoJobs with dependencies, used for multi-stage pipelines such as preprocessing followed by training.

For installation on the platform, see Install Volcano.

Documentation

Volcano upstream documentation and related resources:

  • Volcano Documentation: https://volcano.sh/ — Official documentation covering concepts, scheduler plugins, and usage guides.
  • Volcano GitHub: https://github.com/volcano-sh/volcano — Source code, API reference, and examples for the Volcano project.
  • Kubeflow integration: When using Kubeflow Trainer with the Volcano scheduler, install Alauda support for Volcano before deploying the training component, and enable the Volcano scheduler option in the training component's configuration.