From 8c501051d3118b7f6657ac66ca008c26e4ec6186 Mon Sep 17 00:00:00 2001
From: shenlan <manbuzhe2009@qq.com>
Date: Fri, 27 Jun 2025 12:15:47 +0800
Subject: [PATCH] Support sealos GPU runtime

---
 docs/gpu-containerd.md | 52 ++++++++++++++++++++++++++++++++++++++++++
 scripts/gpu-k8s.sh     | 11 +++++++--
 2 files changed, 61 insertions(+), 2 deletions(-)
 create mode 100644 docs/gpu-containerd.md

diff --git a/docs/gpu-containerd.md b/docs/gpu-containerd.md
new file mode 100644
index 0000000..d186c1d
--- /dev/null
+++ b/docs/gpu-containerd.md
@@ -0,0 +1,52 @@
+# 在 Ubuntu 上安装 NVIDIA 驱动和 nvidia-container-toolkit
+
+以下步骤演示如何在主机安装 NVIDIA 驱动及 nvidia-container-toolkit，并将 containerd 配置为能够使用 GPU。末尾还补充了 sealos 自带 containerd 的配置方式。
+
+## 1. 安装 NVIDIA 驱动和 nvidia-container-toolkit
+
+```bash
+# 添加 NVIDIA 容器工具箱仓库
+sudo apt-get update
+sudo apt-get install -y ca-certificates curl gnupg
+curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | \
+  sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
+
+distribution=$(awk -F= '/^ID=/{print $2}' /etc/os-release)$(awk -F= '/^VERSION_ID=/{print $2}' /etc/os-release)
+curl -s -L "https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list" | \
+  sed 's#^deb #deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] #' | \
+  sudo tee /etc/apt/sources.list.d/nvidia-docker.list
+
+sudo apt-get update
+sudo apt-get install -y nvidia-driver-535 nvidia-container-toolkit
+```
+
+安装完成后可通过 `nvidia-smi` 验证驱动是否正常：
+
+```bash
+nvidia-smi
+```
+
+## 2. 配置 containerd 使用 GPU
+
+使用 `nvidia-ctk` 工具可以快速生成配置并设置为默认运行时：
+
+```bash
+sudo nvidia-ctk runtime configure --runtime=containerd --set-as-default
+sudo systemctl restart containerd
+```
+
+以上命令会在 `/etc/containerd/config.toml` 中新增 `nvidia` 运行时，使 kubelet 或其他工具可以直接调度 GPU 容器。
+
+## 3. sealos-containerd 支持
+
+若主机通过 [sealos](https://github.com/labring/sealos) 部署集群，其内置的 containerd 服务名通常为 `sealos-containerd`，配置文件位于 sealos 数据目录，例如：
+`/var/lib/sealos/data/default/rootfs/etc/containerd/config.toml`。可按以下方式配置：
+
+```bash
+sudo nvidia-ctk runtime configure \
+  --config /var/lib/sealos/data/default/rootfs/etc/containerd/config.toml \
+  --set-as-default
+sudo systemctl restart sealos-containerd
+```
+
+完成后即可在 sealos 集群内运行需要 GPU 的容器或 Pod。
diff --git a/scripts/gpu-k8s.sh b/scripts/gpu-k8s.sh
index 64e521a..043d115 100644
--- a/scripts/gpu-k8s.sh
+++ b/scripts/gpu-k8s.sh
@@ -118,8 +118,15 @@ install_nvidia() {
     sudo apt-get update -y
     sudo apt-get install -y ${NVIDIA_DRIVER_VERSION} nvidia-container-toolkit
   fi
-  sudo nvidia-ctk runtime configure --runtime=containerd --set-as-default
-  sudo systemctl restart containerd
+  if [ "$DEPLOY_MODE" = "sealos" ]; then
+    sudo nvidia-ctk runtime configure \
+      --config /var/lib/sealos/data/default/rootfs/etc/containerd/config.toml \
+      --set-as-default
+    sudo systemctl restart sealos-containerd
+  else
+    sudo nvidia-ctk runtime configure --runtime=containerd --set-as-default
+    sudo systemctl restart containerd
+  fi
   if ! command -v nvidia-smi >/dev/null; then echo "❌ nvidia-smi 未找到"; exit 1; fi
   nvidia-smi || { echo "❌ NVIDIA 驱动有问题"; exit 1; }
 }