Skip to content

Commit

Permalink
chore(helm): mark GPU node as no CPU resource available (#474)
Browse files Browse the repository at this point in the history
Because

- In current setup, GPU node has CPU resources available, causing pure
CPU model will sometime be scheduled onto GPU node, which is not
desirable, we want CPU models to only be scheduled on CPU node.

This commit

- mark GPU node as no CPU resource available for scheduling
  • Loading branch information
heiruwu authored Apr 10, 2024
1 parent 84093ee commit ceb0531
Show file tree
Hide file tree
Showing 2 changed files with 5 additions and 2 deletions.
3 changes: 3 additions & 0 deletions charts/core/templates/ray-service/ray-service.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -114,6 +114,9 @@ spec:
maxReplicas: {{ $workerGroupSpecs.maxReplicas }}
groupName: {{ $workerGroupSpecs.groupName }}
rayStartParams:
{{- if $workerGroupSpecs.gpuWorkerGroup.enabled }}
num-cpus: "0"
{{- end }}
{{- if $workerGroupSpecs.gpuWorkerGroup.customResource }}
resources: {{ $workerGroupSpecs.gpuWorkerGroup.customResource }}
{{- end }}
Expand Down
4 changes: 2 additions & 2 deletions charts/core/values.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -687,11 +687,11 @@ rayService:
headGroupSpec:
resources:
limits:
cpu: "0"
cpu: "2"
memory: "4Gi"
nvidia.com/gpu: 0
requests:
cpu: "0"
cpu: "2"
memory: "4Gi"
nvidia.com/gpu: 0
affinity: {}
Expand Down

0 comments on commit ceb0531

Please sign in to comment.