Adjusting the HAMi scheduling policy through config_patch

When creating or updating a Kubernetes cluster, you can add HAMi configuration in spec.accelerator_virtualization.config_patch through YAML or API. Neutree supports configuring only devicePlugin, scheduler, and global in config_patch.

Configuration example

apiVersion: v1
kind: Cluster
metadata:
  name: <cluster-name>
  workspace: <workspace-name>
spec:
  type: kubernetes
  version: "v1.1.0"
  image_registry: <image-registry>
  accelerator_virtualization:
    enabled: true
    config_patch:
      scheduler:
        defaultSchedulerPolicy:
          nodeSchedulerPolicy: binpack
          gpuSchedulerPolicy: topology-aware
  config:
    kubernetes_config:
      kubeconfig: <kubeconfig>

Operation example

To update an existing cluster, refer to the following example.

First, export the existing cluster configuration by using the CLI, and then add accelerator_virtualization.config_patch to the exported YAML:

neutree-cli-<arch> get Cluster <cluster-name> \
  -w <workspace-name> \
  -o yaml \
  --api-key <api_key> \
  --server-url <server_url> \
  > hami-scheduler-policy.yaml

Edit hami-scheduler-policy.yaml and add or adjust the following fields:

spec:
  accelerator_virtualization:
    enabled: true
    config_patch:
      scheduler:
        defaultSchedulerPolicy:
          nodeSchedulerPolicy: binpack
          gpuSchedulerPolicy: topology-aware

Save the file and run:

neutree-cli-<arch> apply -f hami-scheduler-policy.yaml \
  --api-key <api_key> \
  --server-url <server_url>

View the updated cluster configuration:

neutree-cli-<arch> get Cluster <cluster-name> \
  -w <workspace-name> \
  -o yaml \
  --api-key <api_key> \
  --server-url <server_url>