Kubernetes中自定义Controller

作者：乔克
公众号：运维开发故事
博客：https://www.coolops.cn（COOLOPS）
知乎：乔克叔叔

大家好，我是乔克。

在Kubernetes中，Pod是最小的调度单元，它由各种各样的Controller管理，比如ReplicaSet Controller，Deployment Controller等。

Kubernetes内置了许多Controller，这些Controller能满足80%的业务需求，但是企业里也难免需要自定义Controller来适配自己的业务需求。

网上自定义Controller的文章很多，基本都差不多。俗话说：光说不练假把式，本篇文章主要是自己的一个实践归档总结，如果对你有帮助，可以一键三连！

本文主要从以下几个方面进行介绍，其中包括理论部分和具体实践部分。

![image.png](https://img-blog.csdnimg.cn/img_convert/1b43112054476c9ea3c54b85175ae0e8.png#clientId=u107ea501-9175-4&from=paste&height=285&id=ue21f6f98&margin=[object Object]&name=image.png&originHeight=285&originWidth=655&originalType=binary&ratio=1&size=16219&status=done&style=none&taskId=uff246c85-adb4-46cf-b7c2-4dc1486147a&width=655)

Controller的实现逻辑

当我们向kube-apiserver提出创建一个Deployment需求的时候，首先是会把这个需求存储到Etcd中，如果这时候没有Controller的话，这条数据仅仅是存在Etcd中，并没有产生实际的作用。

所以就有了Deployment Controller，它实时监听kube-apiserver中的Deployment对象，如果对象有增加、删除、修改等变化，它就会做出相应的相应处理，如下：

// pkg/controller/deployment/deployment_controller.go 121行
.....dInformer.Informer().AddEventHandler(cache.ResourceEventHandlerFuncs{AddFunc:    dc.addDeployment,UpdateFunc: dc.updateDeployment,// This will enter the sync loop and no-op, because the deployment has been deleted from the store.DeleteFunc: dc.deleteDeployment,})
......

其实现的逻辑图如下（图片来自网络）：
![image.png](https://img-blog.csdnimg.cn/img_convert/fe8457e45169369dc9dd5cb9ae1909d9.png#clientId=u4c6ad41c-bed8-4&from=paste&id=u34aa9cbf&margin=[object Object]&name=image.png&originHeight=540&originWidth=720&originalType=url&ratio=1&size=268570&status=done&style=none&taskId=u74751601-5ed2-474e-a24a-aea6aa7f3c7)

可以看到图的上半部分都由client-go实现了，下半部分才是我们具体需要去处理的。

client-go主要包含Reflector、Informer、Indexer三个组件。

Reflector会List&Watch kube-apiserver中的特定资源，然后会把变化的资源放入Delta FIFO队列中。
Informer会从Delta FIFO队列中拿取对象交给相应的HandleDeltas。
Indexer会将对象存储到缓存中。

上面部分不需要我们去开发，我们主要关注下半部分。

当把数据交给Informer的回调函数HandleDeltas后，Distribute会将资源对象分发到具体的处理函数，这些处理函数通过一系列判断过后，把满足需求的对象放入Workqueue中，然后再进行后续的处理。

code-generator介绍

上一节说到我们只需要去实现具体的业务需求，这是为什么呢？主要是因为kubernetes为我们提供了code-generator【1】这样的代码生成器工具，可以通过它自动生成客户端访问的一些代码，比如Informer、ClientSet等。

code-generator提供了以下工具为Kubernetes中的资源生成代码：

deepcopy-gen：生成深度拷贝方法，为每个 T 类型生成 func (t* T) DeepCopy() *T 方法，API 类型都需要实现深拷贝
client-gen：为资源生成标准的 clientset
informer-gen：生成 informer，提供事件机制来响应资源的事件
lister-gen：生成 Lister**，**为 get 和 list 请求提供只读缓存层（通过 indexer 获取）

如果需要自动生成，就需要在代码中加入对应格式的配置，如下：
![image.png](https://img-blog.csdnimg.cn/img_convert/415f7ab956b09d6b265ade29f2a5697c.png#clientId=u4c6ad41c-bed8-4&from=paste&height=409&id=uea586526&margin=[object Object]&name=image.png&originHeight=409&originWidth=740&originalType=binary&ratio=1&size=57174&status=done&style=none&taskId=ue44d5792-f763-426e-a569-ddb7647649b&width=740)

其中：

// +genclient表示需要创建client
// +k8s:deepcopy-gen:interfaces=k8s.io/apimachinery/pkg/runtime.Object表示在需要实现k8s.io/apimachinery/pkg/runtime.Object这个接口

除此还有更多的用法，可以参考Kubernetes Deep Dive: Code Generation for CustomResources【2】进行学习。

CRD介绍

CRD全称CustomResourceDefinition，中文简称自定义资源，上面说的Controller主要就是用来管理自定义的资源。

我们可以通过下面命令来查看当前集群中使用了哪些CRD，如下：

# kubectl get crd
NAME                                                 CREATED AT
ackalertrules.alert.alibabacloud.com                 2021-06-15T02:19:59Z
alertmanagers.monitoring.coreos.com                  2019-12-12T12:50:00Z
aliyunlogconfigs.log.alibabacloud.com                2019-12-02T10:15:02Z
apmservers.apm.k8s.elastic.co                        2020-09-14T01:52:53Z
batchreleases.alicloud.com                           2019-12-02T10:15:53Z
beats.beat.k8s.elastic.co                            2020-09-14T01:52:53Z
chaosblades.chaosblade.io                            2021-06-15T02:30:54Z
elasticsearches.elasticsearch.k8s.elastic.co         2020-09-14T01:52:53Z
enterprisesearches.enterprisesearch.k8s.elastic.co   2020-09-14T01:52:53Z
globaljobs.jobs.aliyun.com                           2020-04-26T14:40:53Z
kibanas.kibana.k8s.elastic.co                        2020-09-14T01:52:54Z
prometheuses.monitoring.coreos.com                   2019-12-12T12:50:01Z
prometheusrules.monitoring.coreos.com                2019-12-12T12:50:02Z
servicemonitors.monitoring.coreos.com                2019-12-12T12:50:03Z

但是仅仅是创建一个CRD对象是不够的，因为它是静态的，创建过后仅仅是保存在Etcd中，如果需要其有意义，就需要Controller配合。

创建CRD的例子如下：

apiVersion: apiextensions.k8s.io/v1
kind: CustomResourceDefinition
metadata:# name 必须匹配下面的spec字段：<plural>.<group>name: students.coolops.io
spec:# group 名用于 REST API 中的定义：/apis/<group>/<version>group: coolops.io# 列出自定义资源的所有 API 版本versions:- name: v1    # 版本名称，比如 v1、v1beta1served: true    # 是否开启通过 REST APIs 访问 `/apis/<group>/<version>/...`storage: true   # 必须将一个且只有一个版本标记为存储版本schema:         # 定义自定义对象的声明规范openAPIV3Schema:type: objectproperties:spec:type: objectproperties:name:type: stringschool:type: stringscope: Namespaced    # 定义作用范围：Namespaced（命名空间级别）或者 Cluster（整个集群）names:plural: students   # plural 名字用于 REST API 中的定义：/apis/<group>/<version>/<plural>shortNames:        # shortNames 相当于缩写形式- stukind: Student      # kind 是 sigular 的一个驼峰形式定义，在资源清单中会使用 singular: student  # singular 名称用于 CLI 操作或显示的一个别名

具体演示

本来准备根据官方的demo【3】进行讲解，但是感觉有点敷衍，而且这类教程网上一大堆，所以就准备自己实现一个数据库管理的一个Controller。

因为是演示怎么开发Controller，所以功能不会复杂，主要的功能是：

创建数据库实例
删除数据库实例
更新数据库实例

开发环境说明

本次实验环境如下：

软件	版本
kubernetes	v1.22.3
go	1.17.3
操作系统	CentOS 7.6

创建CRD

CRD是基础，Controller主要是为CRD服务的，所以我们要先定义好CRD资源，便于开发。

apiVersion: apiextensions.k8s.io/v1
kind: CustomResourceDefinition
metadata:name: databasemanagers.coolops.cn
spec:group: coolops.cnversions:- name: v1alpha1served: truestorage: trueschema:openAPIV3Schema:type: objectproperties:spec:type: objectproperties:deploymentName:type: strinreplicas:type: integerminimum: 1maximum: 10dbtype:type: stringstatus:type: objectproperties:availableReplicas:type: integernames:kind: DatabaseManagerplural: databasemanagerssingular: databasemanagershortNames:- dmscope: Namespaced

创建CRD，检验是否能创建成功。

# kubectl apply -f crd.yaml
customresourcedefinition.apiextensions.k8s.io/databasemanagers.coolops.cn created
# kubectl get crd | grep databasemanagers
databasemanagers.coolops.cn                           2021-11-22T02:31:29Z

自定义一个测试用例，如下：

apiVersion: coolops.cn/v1alpha1
kind: DatabaseManager
metadata:name: example-mysql
spec:dbtype: "mysql"deploymentName: "example-mysql"replicas: 1

创建后进行查看：

# kubectl apply -f example-mysql.yaml
databasemanager.coolops.cn/example-mysql created
# kubectl get dm
NAME            AGE
example-mysql   9s

不过现在仅仅是创建了一个静态数据，并没有任何实际的应用，下面来编写Controller来管理这个CRD。

开发Controller

项目地址：https://gitee.com/coolops/database-manager-controller

自动生成代码

1、创建项目目录database-manager-controller，并进行go mod 初始化

# mkdir database-manager-controller
# cd database-manager-controller
# go mod init

2、创建源码包目录pkg/apis/databasemanager

# mkdir pkg/apis/databasemanager -p
# cd pkg/apis/databasemanager

3、在pkg/apis/databasemanager目录下创建register.go文件，并写入一下内容

package databasemanager// GroupName is the group for database manager
const (GroupName = "coolops.cn"
)

4、在pkg/apis/databasemanager目录下创建v1alpha1目录，进行版本管理

# mkdir v1alpha1
# cd v1alpha1

5、在v1alpha1目录下创建doc.go文件，并写入以下内容

// +k8s:deepcopy-gen=package
// +groupName=coolops.cn// Package v1alpha1 is the v1alpha1 version of the API
package v1alpha1

其中// +k8s:deepcopy-gen=package和// +groupName=coolops.cn都是为了自动生成代码而写的配置。

6、在v1alpha1目录下创建type.go文件，并写入以下内容

package v1alpha1import metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"// +genclient
// +k8s:deepcopy-gen:interfaces=k8s.io/apimachinery/pkg/runtime.Objecttype DatabaseManager struct {metav1.TypeMeta   `json:",inline"`metav1.ObjectMeta `json:"metadata,omitempty"`Spec              DatabaseManagerSpec   `json:"spec"`Status            DatabaseManagerStatus `json:"status"`
}// DatabaseManagerSpec 期望状态
type DatabaseManagerSpec struct {DeploymentName string `json:"deploymentName"`Replicas       *int32 `json:"replicas"`Dbtype         string `json:"dbtype"`
}// DatabaseManagerStatus 当前状态
type DatabaseManagerStatus struct {AvailableReplicas int32 `json:"availableReplicas"`
}// +k8s:deepcopy-gen:interfaces=k8s.io/apimachinery/pkg/runtime.Object// DatabaseManagerList is a list of DatabaseManagerList resources
type DatabaseManagerList struct {metav1.TypeMeta `json:",inline"`metav1.ListMeta `json:"metadata"`Items []DatabaseManager `json:"items"`
}

type.go主要定义我们的资源类型。

7、在v1alpha1目录下创建register.go文件，并写入以下内容

package v1alpha1import (dbcontroller "database-manager-controller/pkg/apis/databasemanager"metav1 "k8s.io/apimachinery/pkg/apis/meta/v1""k8s.io/apimachinery/pkg/runtime""k8s.io/apimachinery/pkg/runtime/schema"
)// SchemeGroupVersion is group version used to register these objects
var SchemeGroupVersion = schema.GroupVersion{Group: dbcontroller.GroupName, Version: dbcontroller.Version}// Kind takes an unqualified kind and returns back a Group qualified GroupKind
func Kind(kind string) schema.GroupKind {return SchemeGroupVersion.WithKind(kind).GroupKind()
}// Resource takes an unqualified resource and returns a Group qualified GroupResource
func Resource(resource string) schema.GroupResource {return SchemeGroupVersion.WithResource(resource).GroupResource()
}var (// SchemeBuilder initializes a scheme builderSchemeBuilder = runtime.NewSchemeBuilder(addKnownTypes)// AddToScheme is a global function that registers this API group & version to a schemeAddToScheme = SchemeBuilder.AddToScheme
)// Adds the list of known types to Scheme.
func addKnownTypes(scheme *runtime.Scheme) error {scheme.AddKnownTypes(SchemeGroupVersion,&DatabaseManager{},&DatabaseManagerList{},)metav1.AddToGroupVersion(scheme, SchemeGroupVersion)return nil
}

register.go的作用是通过addKnownTypes方法使得client可以知道DatabaseManager类型的API对象。

至此，自动生成代码的准备工作完成了，目前的代码目录结构如下：

# tree .
.
├── artifacts
│   └── database-manager
│       ├── crd.yaml
│       └── example-mysql.yaml
├── go.mod
├── go.sum
├── LICENSE
├── pkg
│   └── apis
│       └── databasemanager
│           ├── register.go
│           └── v1alpha1
│               ├── doc.go
│               ├── register.go
│               └── type.go

接下里就使用code-generator进行代码自动生成了。

8、创建生成代码的脚本

以下代码主要参考sample-controller【3】

（1）在项目根目录下，创建hack目录，代码生成的脚本配置在该目录下

# mkdir hack && cd hack

（2）创建tools.go文件，添加 code-generator 依赖

//go:build tools
// +build tools// This package imports things required by build scripts, to force `go mod` to see them as dependencies
package toolsimport _ "k8s.io/code-generator"

（3）创建update-codegen.sh文件，用来生成代码

#!/usr/bin/env bashset -o errexit
set -o nounset
set -o pipefailSCRIPT_ROOT=$(dirname "${BASH_SOURCE[0]}")/..
CODEGEN_PKG=${CODEGEN_PKG:-$(cd "${SCRIPT_ROOT}"; ls -d -1 ./vendor/k8s.io/code-generator 2>/dev/null || echo ../code-generator)}# generate the code with:
# --output-base    because this script should also be able to run inside the vendor dir of
#                  k8s.io/kubernetes. The output-base is needed for the generators to output into the vendor dir
#                  instead of the $GOPATH directly. For normal projects this can be dropped.
bash "${CODEGEN_PKG}"/generate-groups.sh "deepcopy,client,informer,lister" \database-manager-controller/pkg/client  database-manager-controller/pkg/apis \databasemanager:v1alpha1 \--output-base "$(dirname "${BASH_SOURCE[0]}")/../.." \--go-header-file "${SCRIPT_ROOT}"/hack/boilerplate.go.txt# To use your own boilerplate text append:
#   --go-header-file "${SCRIPT_ROOT}"/hack/custom-boilerplate.go.txt

其中以下代码段根据实际情况进行修改。

bash "${CODEGEN_PKG}"/generate-groups.sh "deepcopy,client,informer,lister" \database-manager-controller/pkg/client  database-manager-controller/pkg/apis \databasemanager:v1alpha1 \--output-base "$(dirname "${BASH_SOURCE[0]}")/../.." \--go-header-file "${SCRIPT_ROOT}"/hack/boilerplate.go.txt

（4）创建verify-codegen.sh文件，主要用于校验生成的代码是否为最新的

#!/usr/bin/env bashset -o errexit
set -o nounset
set -o pipefailSCRIPT_ROOT=$(dirname "${BASH_SOURCE[0]}")/..DIFFROOT="${SCRIPT_ROOT}/pkg"
TMP_DIFFROOT="${SCRIPT_ROOT}/_tmp/pkg"
_tmp="${SCRIPT_ROOT}/_tmp"cleanup() {rm -rf "${_tmp}"
}
trap "cleanup" EXIT SIGINTcleanupmkdir -p "${TMP_DIFFROOT}"
cp -a "${DIFFROOT}"/* "${TMP_DIFFROOT}""${SCRIPT_ROOT}/hack/update-codegen.sh"
echo "diffing ${DIFFROOT} against freshly generated codegen"
ret=0
diff -Naupr "${DIFFROOT}" "${TMP_DIFFROOT}" || ret=$?
cp -a "${TMP_DIFFROOT}"/* "${DIFFROOT}"
if [[ $ret -eq 0 ]]
thenecho "${DIFFROOT} up to date."
elseecho "${DIFFROOT} is out of date. Please run hack/update-codegen.sh"exit 1
fi

（5）创建boilerplate.go.txt，主要用于为代码添加开源协议

/*
Copyright The Kubernetes Authors.Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License athttp://www.apache.org/licenses/LICENSE-2.0Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

（6）配置go vendor依赖目录

从update-codegen.sh脚本可以看到该代码生成脚本是利用vendor目录下的依赖进行的，我们项目本身没有配置，执行以下命令进行创建。

# go mod vendor

（7）在项目根目录下执行脚本生成代码

# chmod +x hack/update-codegen.sh
# ./hack/update-codegen.sh
Generating deepcopy funcs
Generating clientset for databasemanager:v1alpha1 at database-manager-controller/pkg/client/clientset
Generating listers for databasemanager:v1alpha1 at database-manager-controller/pkg/client/listers
Generating informers for databasemanager:v1alpha1 at database-manager-controller/pkg/client/informers

然后新的目录结构如下：

# tree pkg/
pkg/
├── apis
│   └── databasemanager
│       ├── register.go
│       └── v1alpha1
│           ├── doc.go
│           ├── register.go
│           ├── type.go
│           └── zz_generated.deepcopy.go
└── client├── clientset│   └── versioned│       ├── clientset.go│       ├── doc.go│       ├── fake│       │   ├── clientset_generated.go│       │   ├── doc.go│       │   └── register.go│       ├── scheme│       │   ├── doc.go│       │   └── register.go│       └── typed│           └── databasemanager│               └── v1alpha1│                   ├── databasemanager_client.go│                   ├── databasemanager.go│                   ├── doc.go│                   ├── fake│                   │   ├── doc.go│                   │   ├── fake_databasemanager_client.go│                   │   └── fake_databasemanager.go│                   └── generated_expansion.go├── informers│   └── externalversions│       ├── databasemanager│       │   ├── interface.go│       │   └── v1alpha1│       │       ├── databasemanager.go│       │       └── interface.go│       ├── factory.go│       ├── generic.go│       └── internalinterfaces│           └── factory_interfaces.go└── listers└── databasemanager└── v1alpha1├── databasemanager.go└── expansion_generated.go

Controller开发

上面已经完成了自动代码的生成，生成了informer、lister、clientset的代码，下面就开始编写真正的Controller功能了。

我们需要实现的功能是：

创建数据库实例
更新数据库实例
删除数据库实例

（1）在代码根目录创建controller.go文件，编写如下内容

package mainimport ("context"dbmanagerv1 "database-manager-controller/pkg/apis/databasemanager/v1alpha1"clientset "database-manager-controller/pkg/client/clientset/versioned"dbmanagerscheme "database-manager-controller/pkg/client/clientset/versioned/scheme"informers "database-manager-controller/pkg/client/informers/externalversions/databasemanager/v1alpha1"listers "database-manager-controller/pkg/client/listers/databasemanager/v1alpha1""fmt""github.com/golang/glog"appsv1 "k8s.io/api/apps/v1"corev1 "k8s.io/api/core/v1""k8s.io/apimachinery/pkg/api/errors"metav1 "k8s.io/apimachinery/pkg/apis/meta/v1""k8s.io/apimachinery/pkg/util/runtime"utilruntime "k8s.io/apimachinery/pkg/util/runtime""k8s.io/apimachinery/pkg/util/wait"appsinformers "k8s.io/client-go/informers/apps/v1""k8s.io/client-go/kubernetes""k8s.io/client-go/kubernetes/scheme"typedcorev1 "k8s.io/client-go/kubernetes/typed/core/v1"appslisters "k8s.io/client-go/listers/apps/v1""k8s.io/client-go/tools/cache""k8s.io/client-go/tools/record""k8s.io/client-go/util/workqueue""k8s.io/klog/v2""time"
)const controllerAgentName = "database-manager-controller"const (// SuccessSynced 用来表示事件被成功同步SuccessSynced = "Synced"// MessageResourceSynced 表示事件被触发时的消息信息MessageResourceSynced = "database manager synced successfully"MessageResourceExists = "Resource %q already exists and is not managed by DatabaseManager"ErrResourceExists     = "ErrResourceExists"
)type Controller struct {// kubeclientset 是kubernetes的clientsetkubeclientset kubernetes.Interface// dbmanagerclientset 是自己定义的API Group的clientsetdbmanagerclientset clientset.Interface// deploymentsLister list deployment 对象deploymentsLister appslisters.DeploymentLister// deploymentsSynced 同步deployment对象deploymentsSynced cache.InformerSynced// dbmanagerLister list databasemanager 对象dbmanagerLister listers.DatabaseManagerLister// dbmanagerSynced 同步DatabaseManager对象dbmanagerSynced cache.InformerSynced// workqueue 限速的队列workqueue workqueue.RateLimitingInterface// recorder 事件记录器recorder record.EventRecorder
}// NewController 初始化Controller
func NewController(kubeclientset kubernetes.Interface, dbmanagerclientset clientset.Interface,dbmanagerinformer informers.DatabaseManagerInformer, deploymentInformer appsinformers.DeploymentInformer) *Controller {utilruntime.Must(dbmanagerscheme.AddToScheme(scheme.Scheme))glog.V(4).Info("Create event broadcaster")// 创建eventBroadcastereventBroadcaster := record.NewBroadcaster()// 保存events到日志eventBroadcaster.StartLogging(glog.Infof)// 上报events到APIServereventBroadcaster.StartRecordingToSink(&typedcorev1.EventSinkImpl{Interface: kubeclientset.CoreV1().Events("")})recorder := eventBroadcaster.NewRecorder(scheme.Scheme, corev1.EventSource{Component: controllerAgentName})// 初始化Controllercontroller := &Controller{kubeclientset:      kubeclientset,dbmanagerclientset: dbmanagerclientset,deploymentsLister:  deploymentInformer.Lister(),deploymentsSynced:  deploymentInformer.Informer().HasSynced,dbmanagerLister:    dbmanagerinformer.Lister(),dbmanagerSynced:    dbmanagerinformer.Informer().HasSynced,workqueue:          workqueue.NewNamedRateLimitingQueue(workqueue.DefaultControllerRateLimiter(), "DatabaseManagers"),recorder:           recorder,}glog.Info("Start up event handlers")// 注册Event Handler,分别对于添加、更新、删除事件，具体的操作由事件对应的API将其加入队列中dbmanagerinformer.Informer().AddEventHandler(cache.ResourceEventHandlerFuncs{AddFunc: controller.enqueueDatabaseManager,UpdateFunc: func(oldObj, newObj interface{}) {oldDBManager := oldObj.(*dbmanagerv1.DatabaseManager)newDBManager := newObj.(*dbmanagerv1.DatabaseManager)if oldDBManager.ResourceVersion == newDBManager.ResourceVersion {return}controller.enqueueDatabaseManager(newObj)},DeleteFunc: controller.enqueueDatabaseManagerForDelete,})// 注册Deployment Event HandlerdeploymentInformer.Informer().AddEventHandler(cache.ResourceEventHandlerFuncs{AddFunc: controller.handleObject,UpdateFunc: func(old, new interface{}) {newDepl := new.(*appsv1.Deployment)oldDepl := old.(*appsv1.Deployment)if newDepl.ResourceVersion == oldDepl.ResourceVersion {// 如果没有改变，就返回return}controller.handleObject(new)},DeleteFunc: controller.handleObject,})return controller
}// Run 启动入口
func (c *Controller) Run(threadiness int, stopCh <-chan struct{}) error {defer utilruntime.HandleCrash()defer c.workqueue.ShuttingDown()glog.Info("start controller, cache sync")// 同步缓存数据if ok := cache.WaitForCacheSync(stopCh, c.dbmanagerSynced); !ok {return fmt.Errorf("failed to wait for caches to sync")}glog.Info("begin start worker thread")// 开启work线程for i := 0; i < threadiness; i++ {go wait.Until(c.runWorker, time.Second, stopCh)}glog.Info("worker thread started!!!!!!")<-stopChglog.Info("worker thread stopped!!!!!!")return nil
}// runWorker 是一个死循环，会一直调用processNextWorkItem从workqueue中取出数据
func (c *Controller) runWorker() {for c.processNextWorkItem() {}
}// processNextWorkItem 从workqueue中取出数据进行处理
func (c *Controller) processNextWorkItem() bool {obj, shutdown := c.workqueue.Get()if shutdown {return false}// We wrap this block in a func so we can defer c.workqueue.Done.err := func(obj interface{}) error {defer c.workqueue.Done(obj)var key stringvar ok boolif key, ok = obj.(string); !ok {c.workqueue.Forget(obj)runtime.HandleError(fmt.Errorf("expected string in workqueue but got %#v", obj))return nil}// 在syncHandler中处理业务if err := c.syncHandler(key); err != nil {return fmt.Errorf("error syncing '%s': %s", key, err.Error())}c.workqueue.Forget(obj)glog.Infof("Successfully synced '%s'", key)return nil}(obj)if err != nil {runtime.HandleError(err)return true}return true
}// syncHandler 处理业务Handler
func (c *Controller) syncHandler(key string) error {// 通过split得到namespace和namenamespace, name, err := cache.SplitMetaNamespaceKey(key)if err != nil {runtime.HandleError(fmt.Errorf("invalid resource key: %s", key))return nil}// 从缓存中取对象dbManager, err := c.dbmanagerLister.DatabaseManagers(namespace).Get(name)if err != nil {// 如果DatabaseManager对象被删除了，就会走到这里if errors.IsNotFound(err) {glog.Infof("DatabaseManager对象被删除，请在这里执行实际的删除业务: %s/%s ...", namespace, name)return nil}runtime.HandleError(fmt.Errorf("failed to list DatabaseManager by: %s/%s", namespace, name))return err}glog.Infof("这里是databasemanager对象的期望状态: %#v ...", dbManager)// 获取是否有deploymentNamedeploymentName := dbManager.Spec.DeploymentNameif deploymentName == "" {utilruntime.HandleError(fmt.Errorf("%s: deploymentName 不能为空", key))return nil}// 判断deployment是否在集群中存在deployment, err := c.deploymentsLister.Deployments(dbManager.Namespace).Get(deploymentName)if errors.IsNotFound(err) {// 如果没有找到，就创建deployment, err = c.kubeclientset.AppsV1().Deployments(dbManager.Namespace).Create(context.TODO(), newDeployment(dbManager), metav1.CreateOptions{})}// 如果Create 或者 Get 都出错，则返回if err != nil {return err}// 如果这个deployment不是由DatabaseManager控制，应该报告这个事件if !metav1.IsControlledBy(deployment, dbManager) {msg := fmt.Sprintf(MessageResourceExists, deployment.Name)c.recorder.Event(dbManager, corev1.EventTypeWarning, ErrResourceExists, msg)return fmt.Errorf("%s", msg)}// 如果replicas和期望的不等，则更新deploymentif dbManager.Spec.Replicas != nil && *dbManager.Spec.Replicas != *deployment.Spec.Replicas {klog.V(4).Infof("DatabaseManager %s replicas: %d, deployment replicas: %d", name, *dbManager.Spec.Replicas, *deployment.Spec.Replicas)deployment, err = c.kubeclientset.AppsV1().Deployments(dbManager.Namespace).Update(context.TODO(), newDeployment(dbManager), metav1.UpdateOptions{})}if err != nil {return err}// 更新状态err = c.updateDatabaseManagerStatus(dbManager, deployment)if err != nil {return err}glog.Infof("实际状态是从业务层面得到的，此处应该去的实际状态，与期望状态做对比，并根据差异做出响应(新增或者删除)")c.recorder.Event(dbManager, corev1.EventTypeNormal, SuccessSynced, MessageResourceSynced)return nil
}// updateDatabaseManagerStatus 更新DatabaseManager状态
func (c *Controller) updateDatabaseManagerStatus(dbmanager *dbmanagerv1.DatabaseManager, deployment *appsv1.Deployment) error {dbmanagerCopy := dbmanager.DeepCopy()dbmanagerCopy.Status.AvailableReplicas = deployment.Status.AvailableReplicas_, err := c.dbmanagerclientset.CoolopsV1alpha1().DatabaseManagers(dbmanager.Namespace).Update(context.TODO(), dbmanagerCopy, metav1.UpdateOptions{})return err
}func (c *Controller) handleObject(obj interface{}) {var object metav1.Objectvar ok boolif object, ok = obj.(metav1.Object); !ok {tombstone, ok := obj.(cache.DeletedFinalStateUnknown)if !ok {utilruntime.HandleError(fmt.Errorf("error decoding object, invalid type"))return}object, ok = tombstone.Obj.(metav1.Object)if !ok {utilruntime.HandleError(fmt.Errorf("error decoding object tombstone, invalid type"))return}klog.V(4).Infof("Recovered deleted object '%s' from tombstone", object.GetName())}klog.V(4).Infof("Processing object: %s", object.GetName())if ownerRef := metav1.GetControllerOf(object); ownerRef != nil {// 检查对象是否和DatabaseManager对象关联，如果不是就退出if ownerRef.Kind != "DatabaseManager" {return}dbmanage, err := c.dbmanagerLister.DatabaseManagers(object.GetNamespace()).Get(ownerRef.Name)if err != nil {klog.V(4).Infof("ignoring orphaned object '%s' of databaseManager '%s'", object.GetSelfLink(), ownerRef.Name)return}c.enqueueDatabaseManager(dbmanage)return}
}func newDeployment(dbmanager *dbmanagerv1.DatabaseManager) *appsv1.Deployment {var image stringvar name stringswitch dbmanager.Spec.Dbtype {case "mysql":image = "mysql:5.7"name = "mysql"case "mariadb":image = "mariadb:10.7.1"name = "mariadb"default:image = "mysql:5.7"name = "mysql"}labels := map[string]string{"app": dbmanager.Spec.Dbtype,}return &appsv1.Deployment{ObjectMeta: metav1.ObjectMeta{Namespace: dbmanager.Namespace,Name:      dbmanager.Name,OwnerReferences: []metav1.OwnerReference{*metav1.NewControllerRef(dbmanager, dbmanagerv1.SchemeGroupVersion.WithKind("DatabaseManager")),},},Spec: appsv1.DeploymentSpec{Replicas: dbmanager.Spec.Replicas,Selector: &metav1.LabelSelector{MatchLabels: labels},Template: corev1.PodTemplateSpec{ObjectMeta: metav1.ObjectMeta{Labels: labels},Spec: corev1.PodSpec{Containers: []corev1.Container{{Name:  name,Image: image,},},},},},}
}// 数据先放入缓存，再入队列
func (c *Controller) enqueueDatabaseManager(obj interface{}) {var key stringvar err error// 将对象放入缓存if key, err = cache.MetaNamespaceKeyFunc(obj); err != nil {runtime.HandleError(err)return}// 将key放入队列c.workqueue.AddRateLimited(key)
}// 删除操作
func (c *Controller) enqueueDatabaseManagerForDelete(obj interface{}) {var key stringvar err error// 从缓存中删除指定对象key, err = cache.DeletionHandlingMetaNamespaceKeyFunc(obj)if err != nil {runtime.HandleError(err)return}//再将key放入队列c.workqueue.AddRateLimited(key)
}

其主要逻辑和文章开头介绍的Controller实现逻辑一样，其中关键点在于：

在NewController方法中，定义了DatabaseManager和Deployment对象的Event Handler，除了同步缓存外，还将对应的Key放入queue中。
实际处理业务的方法是syncHandler，可以根据实际请求来编写代码以达到业务需求。

2、在项目根目录下创建main.go，编写入口函数

（1）编写处理系统信号量的Handler

这部分直接使用的demo中的代码【3】

（2）编写入口main函数

package mainimport ("flag""time"kubeinformers "k8s.io/client-go/informers""k8s.io/client-go/kubernetes""k8s.io/client-go/tools/clientcmd""k8s.io/klog/v2"clientset "database-manager-controller/pkg/client/clientset/versioned"informers "database-manager-controller/pkg/client/informers/externalversions""database-manager-controller/pkg/signals"
)var (masterURL  stringkubeconfig string
)func main() {// klog.InitFlags(nil)flag.Parse()// 设置处理系统信号的ChannelstopCh := signals.SetupSignalHandler()// 处理入参cfg, err := clientcmd.BuildConfigFromFlags(masterURL, kubeconfig)if err != nil {klog.Fatalf("Error building kubeconfig: %s", err.Error())}// 初始化kubeClientkubeClient, err := kubernetes.NewForConfig(cfg)if err != nil {klog.Fatalf("Error building kubernetes clientset: %s", err.Error())}// 初始化dbmanagerClientdbmanagerClient, err := clientset.NewForConfig(cfg)if err != nil {klog.Fatalf("Error building example clientset: %s", err.Error())}kubeInformerFactory := kubeinformers.NewSharedInformerFactory(kubeClient, time.Second*30)dbmanagerInformerFactory := informers.NewSharedInformerFactory(dbmanagerClient, time.Second*30)// 初始化controllercontroller := NewController(kubeClient, dbmanagerClient,dbmanagerInformerFactory.Coolops().V1alpha1().DatabaseManagers(), kubeInformerFactory.Apps().V1().Deployments())// notice that there is no need to run Start methods in a separate goroutine. (i.e. go kubeInformerFactory.Start(stopCh)// Start method is non-blocking and runs all registered informers in a dedicated goroutine.kubeInformerFactory.Start(stopCh)dbmanagerInformerFactory.Start(stopCh)if err = controller.Run(2, stopCh); err != nil {klog.Fatalf("Error running controller: %s", err.Error())}
}func init() {flag.StringVar(&kubeconfig, "kubeconfig", "", "Path to a kubeconfig. Only required if out-of-cluster.")flag.StringVar(&masterURL, "master", "", "The address of the Kubernetes API server. Overrides any value in kubeconfig. Only required if out-of-cluster.")
}

测试Controller

1、在项目目录下添加一个Makefile

build:echo "build database manager controller"CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build .

2、执行make build进行编译

# make build
echo "build database manager controller"
build database manager controller
CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build .

然后会输出database-manager-controller一个二进制文件。

3、运行controller

# chmod +x database-manager-controller
# ./database-manager-controller -kubeconfig=$HOME/.kube/config -alsologtostderr=true
I1123 09:52:41.595726   29173 controller.go:81] Start up event handlers
I1123 09:52:41.597448   29173 controller.go:120] start controller, cache sync
I1123 09:52:41.699716   29173 controller.go:125] begin start worker thread
I1123 09:52:41.699737   29173 controller.go:130] worker thread started!!!!!!

4、创建一个CRD测试用例，观察日志以及是否创建deployment

（1）测试样例如下

# cat example-mysql.yaml
apiVersion: coolops.cn/v1alpha1
kind: DatabaseManager
metadata:name: example-mysql
spec:dbtype: "mysql"deploymentName: "mysql"replicas: 1

（2）执行以下命令进行创建，观察日志

# kubectl apply -f example-mysql.yaml
databasemanager.coolops.cn/example-mysql created

可以看到对于的deployment和pod已经创建，不过由于Deployment的配置没有配置完全，mysql没有正常启动。
![image.png](https://img-blog.csdnimg.cn/img_convert/b0e8324086b0bc1cdd2ce410e1ca20a3.png#clientId=u107ea501-9175-4&from=paste&height=588&id=uf939c82e&margin=[object Object]&name=image.png&originHeight=588&originWidth=1598&originalType=binary&ratio=1&size=154914&status=done&style=none&taskId=u8f7a3184-00b9-48ae-b510-d81cd16226d&width=1598)

我们其实是可以看到Controller获取到了事件。
![image.png](https://img-blog.csdnimg.cn/img_convert/9e800f8af1b2ffac265eab11b791588f.png#clientId=u107ea501-9175-4&from=paste&height=1056&id=u2fcfd81d&margin=[object Object]&name=image.png&originHeight=1056&originWidth=2048&originalType=binary&ratio=1&size=433943&status=done&style=none&taskId=uac31feb6-ee60-4d54-b738-a94d8ef4c1b&width=2048)
如果我们删除对象，也可以从日志里正常看到响应。
![image.png](https://img-blog.csdnimg.cn/img_convert/fc70f8819f657dd444df5bb7c00bd804.png#clientId=u107ea501-9175-4&from=paste&height=480&id=u5a4900e0&margin=[object Object]&name=image.png&originHeight=480&originWidth=2048&originalType=binary&ratio=1&size=143406&status=done&style=none&taskId=u2dc883aa-2df4-414e-bcd1-ac6a4d03d7c&width=2048)

总结

上面就是自定义Controller的整个开发过程，相对来说还是比较简单，大部分东西社区都做好了，我们只需要套模子，然后实现自己的逻辑就行。

整个过程主要是参考sample-controller【3】，现在简单整理如下：

确定好目的，然后创建CRD，定义需要的对象
按规定编写代码，定义好CRD所需要的type，然后使用code-generator进行代码自动生成，生成需要的informer、lister、clientset。
编写Controller，实现具体的业务逻辑
编写完成后就是验证，看看是否符合预期，根据具体情况再做进一步的调整

引用

【1】 https://github.com/kubernetes/code-generator.git
【2】 https://cloud.redhat.com/blog/kubernetes-deep-dive-code-generation-customresources
【3】 https://github.com/kubernetes/sample-controller.git
【4】 https://cloud.tencent.com/developer/article/1659440
【5】 https://www.bookstack.cn/read/source-code-reading-notes/kubernetes-k8s_events.md