operator-sdk demo

参考文档

Golang Based Operator Tutorial

构建一个 demo project

前提

  • 装置 operator-sdk operator-sdk release
  • 有一个可拜访的 v1.11.3+ 版本的 kubernetes 集群（若应用 apiextensions.k8s.io/v1 CRDs，需 v1.16.0+）
  • 以 admin 身份拜访 kubernetes 集群

初始化我的项目

env:

$ operator-sdk versionoperator-sdk version: "v1.3.0", commit: "1abf57985b43bf6a59dcd18147b3c574fa57d3f6", kubernetes version: "1.19.4", go version: "go1.15.5", GOOS: "linux", GOARCH: "amd64"$ mkdir -p ${HOME}/golang_codes/operator_sdk_app/demo-operator$ cd ${HOME}/golang_codes/operator_sdk_app/demo-operator

init：

$ operator-sdk init --domain=example.com --repo=github.com/example-inc/memcached-operatorWriting scaffold for you to edit...Get controller runtime:$ go get sigs.k8s.io/controller-runtime@v0.7.0go: downloading ...Update go.mod:$ go mod tidygo: downloading ...Running make:$ makego: creating new go.mod: module tmpDownloading sigs.k8s.io/controller-tools/cmd/controller-gen@v0.4.1go: downloading ...go fmt ./...go vet ./...go build -o bin/manager main.go

理解我的项目目录构造

The --repo=<path> flag is required when creating a project outside of $GOPATH/src, as scaffolded files require a valid module path.

下面这段话没懂 =。=

a. Manager:

主程序 main.go 初始化&运行 Manager。 Manager 为 custom resource API defintions 注册 _Scheme_、设置并运行 controller,webhooks。 Kubebuilder entrypoint doc Manager 能够限度所有 controller 监听资源的 namespace：

mgr, err := ctrl.NewManager(cfg, manager.Options{Namespace: namespace})//默认监听 operator 所在的 namespace//若须要监听所有 namespace，设置为空：mgr, err := ctrl.NewManager(cfg, manager.Options{Namespace: ""})//监听一组特定的 namespace：var namespaces []string // List of Namespaces// Create a new Cmd to provide shared dependencies and start componentsmgr, err := ctrl.NewManager(cfg, manager.Options{ NewCache: cache.MultiNamespacedCacheBuilder(namespaces),})

b. Operator scope:

namespace-scope vs cluster-scope：operator scope

create api & controller:

a. 多组 API：

$ operator-sdk edit --multigroup=true

_PROJECT _文件更新为：

domain: example.comlayout: go.kubebuilder.io/v3multigroup: true...

理解 multi-group migration

b. 创立：

$ operator-sdk create api --group=cache --version=v1alpha1 --kind=Memcached --resource=true --controller=trueWriting scaffold for you to edit...api/v1alpha1/memcached_types.gocontrollers/memcached_controller.goRunning make:$ make/root/golang_codes/operator_sdk_app/demo-operator/bin/controller-gen object:headerFile="hack/boilerplate.go.txt" paths="./..."go fmt ./...go vet ./...go build -o bin/manager main.go

理解 CRD API 标准
理解 kubebuilder api
理解 kubebuilder controller

空虚逻辑

批改 CR API:

批改 CR 类型 (*_types.go)。 给 CR 定义增加正文：_ +kubebuilder:subresource:status_

Add the +kubebuilder:subresource:status marker to add a status subresource to the CRD manifest so that the controller can update the CR status without changing the rest of the CR object.

status 子资源？？ 同步更新 deepcopy 接口实现：_ make generate _

生成 CRD 清单：

CR API 定义好之后，生成 CRD 清单：_ make manifests _

OpenAPI 校验：

OpenAPIv3 schemas are added to CRD manifests in the spec.validation block when the manifests are generated. This validation block allows Kubernetes to validate the properties in a Memcached Custom Resource when it is created or updated. Markers (annotations) are available to configure validations for your API. These markers will always have a +kubebuilder:validation prefix. Usage of markers in API code is discussed in the kubebuilder CRD generation and marker documentation. A full list of OpenAPIv3 validation markers can be found here. To learn more about OpenAPI v3.0 validation schemas in CRDs, refer to the Kubernetes Documentation.

实现控制器：

reconciliation logic.

控制器逻辑实现

a. 控制器监听资源：

b. 协调循环：

c. 定义权限&生成 RBAC 清单：

部署运行

创立资源：

make install

配置测试环境：

a. 集群外

暂未 try

b. 集群内

make docker-build IMG=my/memcached-operator:v0.0.1

部署 operator：

make deploy IMG=my/memcached-operator:v0.0.1
留神： 以上步骤中未蕴含 webhook 准入校验的内容，可临时 敞开准入控制器 ，敞开时批改 /etc/kubernetes/manifests/kube-apiserver.yaml 文件后，需重启 kubelet 失效。

部署实现之后，查看：

$ kubectl -n demo-operator-system get cmNAME                           DATA   AGEdemo-operator-manager-config   1      4h6mf1c5ece8.example.com           0      4h5m$ kubectl -n demo-operator-system get podsNAME                                                READY   STATUS    RESTARTS   AGEdemo-operator-controller-manager-57d894fbfc-gjv9b   2/2     Running   0          4h6m

pod 内运行了两个容器：kube-rbac-proxy, manager

OLM 部署 operator：

todo..

一些问题：

1. 构建镜像时执行 go mod download 失败

(1) 批改 Dockerfile，增加环境变量：

ENV GO111MODULE=on  GOPROXY=https://goproxy.cn,direct

(2) 若在虚机上 docker build，增加参数使共享节点网络：

dokcer build --network=host -t ...

2. 反复构建镜像之后产生 <none>:<none> 的镜像

# 查看$ docker images -a | grep none# 删除"坏"的 <none>:<none> 镜像#docker rmi $(docker images -f "dangling=true" -q)# 清理 dangling 镜像$ docker image prune# 换个版本从新构建docker build -t ...

3. 防火墙导致拉取镜像失败

镜像 gcr.io/distroless/static:nonroot
没找到可用的 http_proxy (￣▽￣)"
应用代替镜像：gengweifeng/gcr-io-distroless-static-nonroot - Docker Hub

4. 在集群中构建镜像&部署时，pod 拉取镜像失败

首先确认 imagePullPolicy: IfNotPresent
(1) 若在 kubernetes 集群：
本地构建镜像，导致仅集群中某一个节点本地有该镜像，其余节点无。拷贝镜像，或限度 pod 节点亲和性解决。
(2) 若在 minikube：
构建镜像前执行：eval $(minikube docker-env)