增强友好性

app/k8sManager/k8s_utils_linuxos_ubuntu.py

@@ -296,6 +296,7 @@ def create_or_update_statefulset(apps_v1, v1, namespace_name, statefulset_name,
                 "kind": "StatefulSet",
                 "metadata": {"name": statefulset_name, "namespace": namespace_name},
                 "spec": {
+                    "runtimeClassName": "nvidia",  # 指定使用 nvidia runtime"
                     "replicas": replicas,
                     "selector": {"matchLabels": source_selflabel},
                     "serviceName": statefulset_name,

app/k8sManager/multiple_clusters.py

@@ -207,6 +207,7 @@ async def new_cluster_install(params):
             # "files/nfs-provisioner-deploy.yaml":"/opt/nfs-provisioner-deploy.yaml",
             "files/nfs-rbac.yaml": "/opt/nfs-rbac.yaml",
             "files/config.toml": "/opt/config.toml",
+            "files/runtimeclass-nvidia.yaml": "/opt/runtimeclass-nvidia.yaml",
             "files/nvidia-device-plugin.yml": "/opt/nvidia-device-plugin.yml",
             "files/libnvidia-container-tools_1.17.8-1_amd64.deb": "/opt/libnvidia-container-tools_1.17.8-1_amd64.deb",
             "files/libnvidia-container1_1.17.8-1_amd64.deb": "/opt/libnvidia-container1_1.17.8-1_amd64.deb",

files/config.toml

@@ -83,7 +83,7 @@ version = 2
       setup_serially = false
 
     [plugins."io.containerd.grpc.v1.cri".containerd]
-      default_runtime_name = "nvidia"
+      default_runtime_name = "runc"
       disable_snapshot_annotations = true
       discard_unpacked_layers = false
       ignore_blockio_not_enabled_errors = false

files/nvidia-device-plugin.yml

@@ -28,13 +28,14 @@ spec:
       labels:
         name: nvidia-device-plugin-ds
     spec:
+      runtimeClassName: nvidia  # 指定使用 nvidia runtime
       tolerations:
       - key: nvidia.com/gpu
         operator: Exists
         effect: NoSchedule
       priorityClassName: "system-node-critical"
       containers:
-      - image: nvcr.io/nvidia/k8s-device-plugin:v0.13.0
+      - image: nvcr.io/nvidia/k8s-device-plugin:v0.16.1
         name: nvidia-device-plugin-ctr
         securityContext:
           allowPrivilegeEscalation: false

files/runtimeclass-nvidia.yaml

@@ -0,0 +1,6 @@
+# runtimeclass-nvidia.yaml
+apiVersion: node.k8s.io/v1
+kind: RuntimeClass
+metadata:
+  name: nvidia  # 与 Pod 中引用的名称一致
+handler: nvidia  # 对应 containerd 配置中的 runtime 名称

script/k8s_install.sh

@@ -544,10 +544,12 @@ if [ "$1" == "master" ]; then
         # kubectl apply -f /opt/ingress-nginx-controller.yaml || log_error "本地安装ingress-nginx-controller插件失败"
         log_info "正在安装nfs-client-provisioner插件"
         aptitude -y install nfs-kernel-server nfs-common=1:1.3.4-2.5ubuntu3.7
-        # log_info "正在安装MetricsServer插件"
+        log_info "正在安装MetricsServer插件"
-        # kubectl apply -f /opt/components.yaml || log_error "本地安装MetricsServer插件失败"
+        kubectl apply -f /opt/components.yaml || log_error "本地安装MetricsServer插件失败"
-        # log_info "正在安装GPU模式必要插件"
+        log_info "正在安装GPU模式必要插件runtimeclass-nvidia.yaml"
-        # kubectl apply -f /opt/nvidia-device-plugin.yml || log_error "本地安装GPU模式必要插件失败"
+        kubectl apply -f /opt/runtimeclass-nvidia.yaml || log_error "本地安装GPU模式必要插件runtimeclass-nvidia.yaml失败"
+        log_info "正在安装GPU模式必要插件nvidia-device-plugin.yml"
+        kubectl apply -f /opt/nvidia-device-plugin.yml || log_error "本地安装GPU模式必要插件失败"
         if [ $? -ne 0 ]; then
             echo "NFS 服务器端安装失败,请检查网络连接或软件源。"
             exit 1