fix(cli): dynamic creation of nvidia runtimeclass

2026-03-02 11:20:00 +08:00
2 changed files with 5 additions and 2 deletions
--- a/infrastructure/gpu/.olares/config/gpu/hami/templates/device-plugin/runtime-class.yaml
+++ b/infrastructure/gpu/.olares/config/gpu/hami/templates/device-plugin/runtime-class.yaml
@@ -1,4 +1,6 @@
 {{- if and .Values.devicePlugin.createRuntimeClass .Values.devicePlugin.runtimeClassName }}
+{{- $existingRuntimeClass := lookup "node.k8s.io/v1" "RuntimeClass" "" .Values.devicePlugin.runtimeClassName }}
+{{- if not $existingRuntimeClass }}
 apiVersion: node.k8s.io/v1
 kind: RuntimeClass
 metadata:
@@ -7,3 +9,4 @@ metadata:
    helm.sh/hook: pre-install,pre-upgrade
 handler: nvidia
 {{- end }}
+{{- end }}
--- a/infrastructure/gpu/.olares/config/gpu/hami/values.yaml
+++ b/infrastructure/gpu/.olares/config/gpu/hami/values.yaml
@@ -151,9 +151,9 @@ devicePlugin:
  deviceMemoryScaling: 1
  deviceCoreScaling: 1
  # The runtime class name to be used by the device plugin, and added to the pod.spec.runtimeClassName of applications utilizing NVIDIA GPUs
-  runtimeClassName: ""
+  runtimeClassName: "nvidia"
  # Whether to create runtime class, name comes from runtimeClassName when it is set
-  createRuntimeClass: false
+  createRuntimeClass: true
  migStrategy: "none"
  disablecorelimit: "false"
  passDeviceSpecsEnabled: false