英伟达(NVIDIA)发布了名为 NVIDIA Kubernetes Engine (NVKE) 的预览版服务,旨在让用户在自有 GPU 基础设施上运行托管 Kubernetes 集群 1。该工具作为管理平面,通过 NVIDIA Infra Controller (NICo) 配置机器并创建独立的工作负载集群,支持集群生命周期管理及节点身份认证 1。每个创建的集群都拥有独立的控制平面、etcd 和节点池,且必须使用 NICo 作为唯一兼容的基础设施提供商 1。
由于目前处于 Preview 阶段,官方明确指出不建议将 NVKE 用于生产工作负载,且其 API 和行为可能发生变更 1。在本地快速入门时,用户需要一台支持 Docker 和虚拟 TPM 的 Linux 主机或虚拟机,硬件要求至少包含 6 CPU、12 GiB 内存和 100 GiB 磁盘 1。安全机制方面,节点通过 SPIFFE/SPIRE 利用 TPM 进行身份验证,同时 etcd 数据加密可用 1。
需要注意的是,NVKE 当前存在若干局限性:API 认证默认关闭并使用 HTTP 协议,管理集群不支持自我升级、扩展或替换 1。该项目采用 Apache License 2.0 许可证,版权年份标注为 2026 1。
NVIDIA has introduced a preview version of the NVIDIA Kubernetes® Engine (NVKE), a service designed to enable users to operate managed Kubernetes clusters on their own GPU infrastructure. 1 The tool functions as a management plane that configures machines via the NVIDIA Infra Controller (NICo) and creates independent workload clusters, each featuring its own control plane, etcd database, and node pool. 1 Currently in the preview stage, NVKE is not recommended for production environments due to potential changes in its API and behavior. 1
To function correctly, NVKE requires the NVIDIA Infra Controller (NICo) as the sole compatible infrastructure provider. 1 For local deployment, the system demands a Linux host or virtual machine equipped with at least six CPUs, 12 GiB of memory, and 100 GiB of disk space, along with support for Docker and virtual TPMs. 1 Security mechanisms include node authentication via SPIFFE/SPIRE using TPMs, although etcd data encryption remains an optional feature. 1
The current iteration of NVKE presents several operational limitations, including default-disabled API authentication over HTTP and a lack of support for self-upgrades, scaling, or replacement of the management cluster. 1 Released under the Apache License 2.0 with a 2026 copyright notice, the project remains experimental for critical workflows. 1
评论
还没有评论,欢迎留下第一条。