Project-HAMi / HAMi

Heterogeneous AI Computing Virtualization Middleware
http://project-hami.io/
Apache License 2.0
874 stars 183 forks source link

volcano schedule #453

Open Japson0 opened 2 months ago

Japson0 commented 2 months ago

I have a question, I saw that HAMI supports volcano, I read other issues, it only states that you can use the volcanic-device-plugin to support it. So I would like to ask can volcano match the HAMI plugin. Since pure vgpu doesn't quite fit our scenario, I wanted to use some of volcano's features and some of HAMI's features. For example, the number of the GPU. It looks like the following configuration

apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: test-job1
  namespace: emlp
  annotations:
    nvidia.com/use-gpuuuid: "GPU-123456"
spec:
  minAvailable: 1
  schedulerName: volcano
  policies:
    - event: PodEvicted
      action: RestartJob
  plugins:
    ssh: []
    env: []
    svc: []
  maxRetry: 5
  tasks:
    - replicas: 1
      name: "default-nginx"
      template:
        metadata:
          name: test
        spec:
          containers:
            - image: nvidia/cuda:12.4.1-base-ubuntu22.04
              imagePullPolicy: IfNotPresent
              name: llama
              securityContext: 
               privileged: true 
              command: ["sleep"]
              args: ["100000"]
              resources:
                limits:
                  nvidia.com/gpu: 1
                  cpu: 3
          restartPolicy: OnFailure
archlitchi commented 2 months ago

the 'specify uuid' of vgpu is not supported on volcano-scheduler yet, we plan to implement that next version

Japson0 commented 2 months ago

the 'specify uuid' of vgpu is not supported on volcano-scheduler yet, we plan to implement that next version

Thanks for taking the time to answer my w question

Japson0 commented 2 months ago

@archlitchi Please tell me which hami attributes support volcano schedule, or where I can look it up. thanks