DockerDesktop

Sun, 28 May 2023 15:45:09 +0800

DockerDesktop#

在设置中打开kubernetes的支持即可。

Minikube

Sun, 28 May 2023 15:44:15 +0800

Minikube#

Kind

Sun, 28 May 2023 15:44:03 +0800

Kind#

kind是一款依赖docker或podman来运行单机版kubernetes的工具。

kind官网：https://kind.sigs.k8s.io/

安装#

go install sigs.k8s.io/kind@latest

命令#

# 创建集群
$ kind create cluster

# 删除集群
$ kind delete cluster

Colima

Sun, 28 May 2023 15:46:09 +0800

Colima#

Colima 是MacOS 及Linux上的容器/Kubernetes安装器。由于支持M1及Intel的MacOS，且自带docker环境，故可以是M1芯片的MacOS的首选。

Colima基于 K3S 实现的Kubernetes

Github连接：https://github.com/abiosoft/colima

安装#

# Homebrew
$ brew install colima

使用#

# 只启动Docker容器
$ colima start

# 启动Kubernetes
$ colima start --kubernetes

# 在只启动docker的colima环境增加kubernetes
$ colima kubernetes start

# 停止colima服务
$ colima stop

# 删除colima服务
$ colima delete

K3S

Fri, 12 Sep 2025 15:46:09 +0800

K3S#

新建集群#

curl -sfL https://rancher-mirror.rancher.cn/k3s/k3s-install.sh | \
     INSTALL_K3S_MIRROR=cn sh -s - \
     --system-default-registry "registry.cn-hangzhou.aliyuncs.com" \
     --write-kubeconfig ~/.kube/config \
     --write-kubeconfig-mode 644 \
     --kube-proxy-arg proxy-mode=ipvs \
     --docker \
     --disable servicelb,traefik,local-storage,metrics-server

# 指定版本的环境变量
INSTALL_K3S_VERSION=v1.35.0+k3s1

# 默认运行时改为nvidia
--default-runtime=nvidia

节点加入集群#

# 主节点
cat /var/lib/rancher/k3s/server/node-token

# 子节点
curl -sfL http://rancher-mirror.rancher.cn/k3s/k3s-install.sh | \
    INSTALL_K3S_MIRROR=cn K3S_URL=https://192.168.95.3:6443 \
    K3S_TOKEN=K1019f852595b9a786009f19fce810ec01cade498313d4feba303c455f46901e450::server:8473e205d08a849a77648aad8a2cf5a1  \
    sh -

配置nerdctl#

# 获取nerdctl
https://github.com/containerd/nerdctl/releases/download/v2.1.6/nerdctl-2.1.6-linux-amd64.tar.gz

调整配置: /etc/nerdctl/nerdctl.toml

mkdir -p /etc/nerdctl
tee /etc/nerdctl/nerdctl.toml > /dev/null <<EOF
address        = "unix:///run/k3s/containerd/containerd.sock"
namespace      = "k8s.io"
EOF

卸载k3s#

k3s-uninstall.sh

k3s配置调度器#

# /etc/systemd/system/k3s.service
    --kube-scheduler-arg config=/etc/kubernetes/scheduler-plugins.yaml

# /etc/kubernetes/scheduler-plugins.yaml
apiVersion: kubescheduler.config.k8s.io/v1
kind: KubeSchedulerConfiguration
clientConnection:
  kubeconfig: /etc/rancher/k3s/k3s.yaml
  qps: 100
  burst: 150
extenders:
  - urlPrefix: "http://10.43.100.176:8888/scheduler"
    filterVerb: "predicates"
    prioritizeVerb: "priorities"
    bindVerb: "bind"
    weight: 100
    enableHTTPS: false
    nodeCacheCapable: false
    ignorable: false
    httpTimeout: 60s
    managedResources:
    - name: vgpu.jdos.io/vcuda-core
    - name: nvidia.com/gpu
    - name: huawei.com/Ascend310P
    - name: cambricon.com/mlu
    - name: huawei.com/Ascend910
    - name: huawei.com/npu-core
    - name: hygon.com/dcu

k3s - containerd 配置国内源#

# 国内镜像加速
cat > /etc/rancher/k3s/registries.yaml <<EOF
mirrors:
  docker.io:
    endpoint:
    - https://docker.1panel.live
    - https://docker.1ms.run
    - https://dytt.online
    - https://lispy.org
    - https://docker-0.unsee.tech
    - https://docker.xiaogenban1993.com
    - https://666860.xyz
    - https://hub.rat.dev
    - https://docker.m.daocloud.io
    - https://demo.52013120.xyz
    - https://proxy.vvvv.ee
EOF

安装单机简易版 on 安装K8S | 君自强