CNCF Atlas

English

日本語

English

日本語

Appearance

Getting Started ​

Verified against the install manifest on master (latest release v1.15.0). Commands assume a working Kubernetes cluster and kubectl with cluster-admin.

Prerequisites ​

  • A Kubernetes cluster (the YAML install works on both x86_64 and arm64).
  • kubectl configured against that cluster.
  • For Kubernetes v1.17 and above, the recommended CRDs are used automatically by the manifest below.

Install ​

Apply the development manifest to an existing cluster:

bash
kubectl apply -f https://raw.githubusercontent.com/volcano-sh/volcano/master/installer/volcano-development.yaml

Or install the official release with Helm:

bash
helm repo add volcano-sh https://volcano-sh.github.io/helm-charts
helm install volcano volcano-sh/volcano -n volcano-system --create-namespace

A first working setup ​

  1. Confirm the three control-plane pods are running in volcano-system.

    bash
    kubectl get pods -n volcano-system

Expected output is one running pod each for the scheduler, controllers, and admission webhook:

text
NAME                                   READY   STATUS    RESTARTS   AGE
volcano-admission-5bd5756f79-dnr4l     1/1     Running   0          96s
volcano-controllers-687948d9c8-nw4b4   1/1     Running   0          96s
volcano-scheduler-94998fc64-4z8kh      1/1     Running   0          96s

  1. Submit a VolcanoJob. The key fields are schedulerName: volcano and minAvailable, which is the gang size. This example asks for 6 replicas but will only start once at least 3 pods can be placed together.

    yaml
    apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: test-job
spec:
  minAvailable: 3
  schedulerName: volcano
  queue: default
  maxRetry: 5
  tasks:
    - replicas: 6
      name: "default-nginx"
      template:
        spec:
          containers:
            - image: nginx
              imagePullPolicy: IfNotPresent
              name: nginx
              resources:
                requests:
                  cpu: "1"
          restartPolicy: OnFailure

  2. Apply it.

    bash
    kubectl apply -f job.yaml

Verify it works ​

Check that the job created a PodGroup and that its pods were scheduled by Volcano:

bash
kubectl get podgroups
kubectl get pods -l volcano.sh/job-name=test-job

A healthy job shows its PodGroup reaching the Running phase once minAvailable pods are placed. If fewer than minAvailable pods can fit, the gang stays pending rather than starting partially.

Where to go next ​

For queues and fair-share, hierarchical quota, topology- and NUMA-aware scheduling, GPU/NPU device sharing, and enabling preemption or reclaim actions, see the official documentation at https://volcano.sh/en/. Production concerns such as high availability of the scheduler, webhook certificates, and tuning the schedule period are covered there.