导语：重启服务器之后，节点连接pod报错Error from server: error dialing backend: dial tcp 172.16.20.88:10250

现象是pod都无法进入，查看报错日志就是提示Error from server: error dialing backend: dial tcp 172.16.20.88:10250。

查看节点状态 kubectl get nodes发现节点是notready

登录到节点查看kubelet服务 发现kubelet报错，怀疑是修改了/etc/docker/daemon.json所影响。证实的确是修改/etc/docker/daemon.json导致

故障原因：调整了kubesphere安装的k8s的docker配置。

原配置

{
  "log-opts": {
    "max-size": "5m",
    "max-file":"3"
  },
  "exec-opts": ["native.cgroupdriver=systemd"],
    "default-shm-size": "2G",
"insecure-registries": ["harbor.deepwise.com","10.10.3.5","172.28.3.5"] ,
    "runtimes": {
        "nvidia": {
            "path": "nvidia-container-runtime",
            "runtimeArgs": []
        }
    }
}

故障配置

{
    "default-runtime": "nvidia",
    "runtimes": {
        "nvidia": {
            "path": "/usr/bin/nvidia-container-runtime",
            "runtimeArgs": []
        }
    }
}

调整后的配置

{
  "log-opts": {
    "max-size": "5m",
    "max-file":"3"
  },
  "exec-opts": ["native.cgroupdriver=systemd"],
"insecure-registries": ["harbor.deepwise.com","10.10.3.5","172.28.3.5"] ,
    "default-runtime": "nvidia",
    "runtimes": {
        "nvidia": {
            "path": "/usr/bin/nvidia-container-runtime",
            "runtimeArgs": []
        }
    }
}