KServe ModelMesh部署时出现seldonio/mlserver镜像拉取失败问题

该问题不属于Docker与K8s的核心配置异常，本质是容器镜像拉取环节的网络超时故障。

部署过程输出的事件日志如下：

Events:
  Type     Reason     Age    From               Message
  ----     ------     ----   ----               -------
  Normal   Scheduled  4m20s  default-scheduler  Successfully assigned modelmesh-serving/modelmesh-serving-mlserver-0.x-77cc8fd548-xdgvr to minikube
  Normal   Pulling    4m18s  kubelet            Pulling image "kserve/modelmesh:v0.9.0-rc0"
  Normal   Pulled     3m18s  kubelet            Successfully pulled image "kserve/modelmesh:v0.9.0-rc0" in 59.419620166s
  Normal   Created    3m18s  kubelet            Created container mm
  Normal   Started    3m17s  kubelet            Started container mm
  Normal   Pulling    3m17s  kubelet            Pulling image "seldonio/mlserver:0.5.2"
  Warning  Failed     68s    kubelet            Failed to pull image "seldonio/mlserver:0.5.2": rpc error: code = Unknown desc = context deadline exceeded
  Warning  Failed     68s    kubelet            Error: ErrImagePull
  Normal   Pulling    68s    kubelet            Pulling image "kserve/modelmesh-runtime-adapter:v0.9.0-rc0"

从日志可明确判断故障特征：

• 第一个基础镜像kserve/modelmesh:v0.9.0-rc0拉取耗时接近1分钟，已经达到kubelet默认拉取超时阈值的临界值，说明当前公网镜像拉取链路质量很差
• 拉取seldonio/mlserver:0.5.2时直接触发context deadline exceeded超时错误，属于典型的公网镜像仓库连通性差、带宽不足导致的拉取失败，和Docker、K8s核心组件配置错误无关
• 后续拉取kserve/modelmesh-runtime-adapter:v0.9.0-rc0是kubelet按顺序拉取Pod内其他容器镜像的正常行为，不属于额外异常

• 验证网络连通性：执行minikube ssh进入集群节点，手动运行docker pull seldonio/mlserver:0.5.2，确认手动拉取是否同样超时，锁定网络故障根因
• 配置镜像加速：给minikube节点内的Docker配置国内镜像加速源，重启Docker服务后重新触发Pod部署
• 本地导入镜像：在网络通畅的环境下提前拉取所需版本的镜像，通过minikube image load <镜像名:标签>命令将本地镜像直接导入minikube节点，跳过公网拉取环节
• 调整超时阈值：修改kubelet的--image-pull-progress-deadline启动参数，将默认1分钟的拉取超时时间适当调大，给大体积镜像拉取预留充足时间

内容的提问来源于stack exchange，提问作者bonijad383