client-go gin的简单整合四-list-watch初探

背景：

完成了client-go gin的简单整合三（list列表相关再进阶关于Pods），恩如果有代理是可以看到每次的请求都要访问后端服务的，如何避免频繁调用后端apiserver呢？list-watch监听机制可以使用一下？

关于list-watch:

参照：https://blog.51cto.com/u_15127559/3377812(错别字好多？最后还是引用了沈老师的ppt上面的概念！),

• list http短链接调用资源的api，获取列表。
• 使用http长连接持续监听资源，有变化则返回一个WatchEvent

client-go informer

client-go k8s.io/client-go/tools/cache包informer对象对list-watch机制进行了封装

• 初始化调用List api获得全量list 缓存（本地缓存）
• 调用watch api watch资源，当资源发生变更通过一定机制维护缓存，减少访问apiserver的压力

个人觉得不错的文章Client-go源码分析之Reflector,华为云不错的视频list-watch机制原理详解,
client-go(kubernetes)的ListerWatcher解析.

client-go gin的简单整合四-list-watch

以deployment简单例子的开始

文件名** /src/service/test.go，监控deployment的变化， **开始其实是不是可以跟java是的弄一个单独的测试包？这里就简单操作了偷懒…

cache.NewInformer()开始：

cache.NewInformer()

goland查看源码功能：

实现Handler方法：

只实现了OnUpdate方法（仅仅打印deployment名字），OnAdd，OnDelete是空的：

type DepHandler struct {
}

func (d *DepHandler) OnAdd(obj interface{}) {}
func (d *DepHandler) OnUpdate(oldObj, newObj interface{}) {
	if dep, ok := newObj.(*v1.Deployment); ok {
		fmt.Println(dep.Name)
	}
}
func (d *DepHandler) OnDelete(obj interface{}) {
}

最终test.go如下：

package main

import (
	"fmt"
	"k8s-demo1/src/lib"
	v1 "k8s.io/api/apps/v1"
	"k8s.io/apimachinery/pkg/fields"
	"k8s.io/apimachinery/pkg/util/wait"
	"k8s.io/client-go/tools/cache"
)

type DepHandler struct {
}

func (d *DepHandler) OnAdd(obj interface{}) {}
func (d *DepHandler) OnUpdate(oldObj, newObj interface{}) {
	if dep, ok := newObj.(*v1.Deployment); ok {
		fmt.Println(dep.Name)
	}
}
func (d *DepHandler) OnDelete(obj interface{}) {
}
func main() {
	s, c := cache.NewInformer(cache.NewListWatchFromClient(lib.K8sClient.AppsV1().RESTClient(),
		"deployments", "default", fields.Everything()),
		&v1.Deployment{},
		0,
		&DepHandler{},
	)
	c.Run(wait.NeverStop)
	s.List()
}

关于s c 源码中Store, Controller

运行test.go

go run test.go
手动修改nginx deployment副本数量，查看 goland输出：

[zhangpeng@zhangpeng ~]$ kubectl get deployments
NAME    READY   UP-TO-DATE   AVAILABLE   AGE
nginx   2/2     2            2           10d
[zhangpeng@zhangpeng ~]$ kubectl get deployment
NAME    READY   UP-TO-DATE   AVAILABLE   AGE
nginx   2/2     2            2           10d
[zhangpeng@zhangpeng ~]$ kubectl scale deployment/nginx --replicas=3
deployment.apps/nginx scaled
[zhangpeng@zhangpeng ~]$ 

注：resource 是deployments，不能是deployment, *DepHandler 为什么加指针运算符？

实现一个pod的list-watch?

首先停止test.go，并注释掉test.go中代码。

创建了一个/src/service/test1.go,照着test.go来一遍：

package main

import (
	"fmt"
	"k8s-demo1/src/lib"
	corev1 "k8s.io/api/core/v1"
	"k8s.io/apimachinery/pkg/fields"
	"k8s.io/apimachinery/pkg/util/wait"
	"k8s.io/client-go/tools/cache"
)

type PodHandler struct {
}

func (p *PodHandler) OnAdd(obj interface{}) {}
func (p *PodHandler) OnUpdate(oldObj, newObj interface{}) {
	if pods, ok := newObj.(*corev1.Pod); ok {
		fmt.Println(pods.Name)
	}
}
func (p *PodHandler) OnDelete(obj interface{}) {
}
func main() {
	s, c := cache.NewInformer(cache.NewListWatchFromClient(lib.K8sClient.CoreV1().RESTClient(),
		"pods", "default", fields.Everything()),
		&corev1.Pod{},
		0,
		&PodHandler{},
	)
	c.Run(wait.NeverStop)
	s.List()
}

注意：pod 的api是 corev1。参照：https://github.com/kubernetes/client-go/blob/release-1.23/informers/core/v1/pod.go#L58
运行test1.go,修改default 下nginx deployment副本数：

[zhangpeng@zhangpeng ~]$ kubectl scale deployment/nginx --replicas=3
deployment.apps/nginx scaled

SharedInformerFactory工厂模式

思考一下为什么要使用工厂模式呢？
关于SharedInformerFactory参考：https://stackoverflow.com/questions/40975307/how-to-watch-events-on-a-kubernetes-service-using-its-go-client,https://qiankunli.github.io/2020/07/20/client_go.html

先test1.go的pod开始：

src/service/test1.go

package main

import (
	"fmt"
	"k8s-demo1/src/lib"
	corev1 "k8s.io/api/core/v1"
	"k8s.io/apimachinery/pkg/util/wait"
	"k8s.io/client-go/informers"
)

type PodHandler struct {
}

func (p *PodHandler) OnAdd(obj interface{}) {}
func (p *PodHandler) OnUpdate(oldObj, newObj interface{}) {
	if pods, ok := newObj.(*corev1.Pod); ok {
		fmt.Println(pods.Name)
	}
}
func (p *PodHandler) OnDelete(obj interface{}) {
}
func main() {
	factory := informers.NewSharedInformerFactory(lib.K8sClient, 0)
	podinformer := factory.Core().V1().Pods()
	podinformer.Informer().AddEventHandler(&PodHandler{})
	factory.Start(wait.NeverStop)
	select {}
}

写的时候以为直接corev1…发现是core().v1(),为什么要用select{}呢？参照：https://blog.csdn.net/cbmljs/article/details/93497415。阻塞，防止程序退出！运行test1.go程序，修改nginx deployment副本：

[zhangpeng@zhangpeng ~]$ kubectl scale deployment/nginx --replicas=4
deployment.apps/nginx scaled

可不可以deployment pod list-watch搞在一起呢？现在写了一个test.go,test1.go?貌似也是可以的…先搞在一起，如下：

package main

import (
	"fmt"
	"k8s-demo1/src/lib"
	v1 "k8s.io/api/apps/v1"
	corev1 "k8s.io/api/core/v1"
	"k8s.io/apimachinery/pkg/util/wait"
	"k8s.io/client-go/informers"
)

type PodHandler struct {
}

func (p *PodHandler) OnAdd(obj interface{}) {}
func (p *PodHandler) OnUpdate(oldObj, newObj interface{}) {
	if pods, ok := newObj.(*corev1.Pod); ok {
		fmt.Println(pods.Name)
	}
}
func (p *PodHandler) OnDelete(obj interface{}) {
}

type DepHandler struct {
}

func (d *DepHandler) OnAdd(obj interface{}) {}
func (d *DepHandler) OnUpdate(oldObj, newObj interface{}) {
	if dep, ok := newObj.(*v1.Deployment); ok {
		fmt.Println(dep.Name)
	}
}
func (d *DepHandler) OnDelete(obj interface{}) {
}
func main() {
	factory := informers.NewSharedInformerFactory(lib.K8sClient, 0)
	podinformer := factory.Core().V1().Pods()
	podinformer.Informer().AddEventHandler(&PodHandler{})
	depinformer := factory.Apps().V1().Deployments()
	depinformer.Informer().AddEventHandler(&DepHandler{})
	factory.Start(wait.NeverStop)
	select {}
}

运行test1.go ，更改nginx deployment副本数量，打印了deployment name 和pod name！

Handler OnAdd

补全一下OnAdd方法，打印一下pod deployment列表：

func (p *PodHandler) OnAdd(obj interface{}) {
	fmt.Println(obj.(*corev1.Pod).Name)
}
func (d *DepHandler) OnAdd(obj interface{}) {
	fmt.Println(obj.(*v1.Deployment).Name)
}

package main

import (
	"fmt"
	"k8s-demo1/src/lib"
	v1 "k8s.io/api/apps/v1"
	corev1 "k8s.io/api/core/v1"
	"k8s.io/apimachinery/pkg/util/wait"
	"k8s.io/client-go/informers"
)

type PodHandler struct {
}

func (p *PodHandler) OnAdd(obj interface{}) {
	fmt.Println(obj.(*corev1.Pod).Name)
}
func (p *PodHandler) OnUpdate(oldObj, newObj interface{}) {
	if pods, ok := newObj.(*corev1.Pod); ok {
		fmt.Println(pods.Name)
	}
}
func (p *PodHandler) OnDelete(obj interface{}) {
}

type DepHandler struct {
}

func (d *DepHandler) OnAdd(obj interface{}) {
	fmt.Println(obj.(*v1.Deployment).Name)
}
func (d *DepHandler) OnUpdate(oldObj, newObj interface{}) {
	if dep, ok := newObj.(*v1.Deployment); ok {
		fmt.Println(dep.Name)
	}
}
func (d *DepHandler) OnDelete(obj interface{}) {
}
func main() {
	factory := informers.NewSharedInformerFactory(lib.K8sClient, 0)
	podinformer := factory.Core().V1().Pods()
	podinformer.Informer().AddEventHandler(&PodHandler{})
	depinformer := factory.Apps().V1().Deployments()
	depinformer.Informer().AddEventHandler(&DepHandler{})
	factory.Start(wait.NeverStop)
	select {}
}

sync.Map

为什么引用sync.Map呢？Go语言sync.Map（在并发环境中使用的map），还是考虑并发原因！
拿deployment为例，打印一下develop namespace命名空间下的deployment列表：
test1.go

package main

import (
	"context"
	"fmt"
	"github.com/gin-gonic/gin"
	"k8s-demo1/src/lib"
	v1 "k8s.io/api/apps/v1"
	"k8s.io/apimachinery/pkg/util/wait"
	"k8s.io/client-go/informers"
	"log"
	"sync"
	"time"
)

type DeploymentMap struct {
	data sync.Map
}

func (depmap *DeploymentMap) Add(dep *v1.Deployment) {
	if list, ok := depmap.data.Load(dep.Namespace); ok {
		list = append(list.([]*v1.Deployment), dep)
		depmap.data.Store(dep.Namespace, list)
	} else {
		depmap.data.Store(dep.Namespace, []*v1.Deployment{dep})
	}
}

type DepHandler struct {
}

func (d *DepHandler) OnAdd(obj interface{}) {
	//fmt.Println(obj.(*v1.Deployment).Name)
	DepMap.Add(obj.(*v1.Deployment))
}
func (d *DepHandler) OnUpdate(oldObj, newObj interface{}) {
	if dep, ok := newObj.(*v1.Deployment); ok {
		fmt.Println(dep.Name)
	}
}
func (d *DepHandler) OnDelete(obj interface{}) {
}

var DepMap *DeploymentMap

func init() {
	DepMap = &DeploymentMap{}
}
func main() {
	factory := informers.NewSharedInformerFactory(lib.K8sClient, 0)
	depinformer := factory.Apps().V1().Deployments()
	depinformer.Informer().AddEventHandler(&DepHandler{})
	factory.Start(wait.NeverStop)
	c, _ := context.WithTimeout(context.Background(), time.Second*3)
	select {
	case <-c.Done():
		log.Fatal("time out")
	default:
		r := gin.New()
		r.GET("/", func(c *gin.Context) {
			var res []string
			DepMap.data.Range(func(key, value interface{}) bool {
				if key == "develop" {
					for _, item := range value.([]*v1.Deployment) {
						res = append(res, item.Name)
					}
				}
				return true

			})
			c.JSON(200, res)

		})
		r.Run(":8080")
	}
}

没有作具体的路由，直接测试一下访问：http://127.0.0.1:8080/

总结

1. list-watch机制
2. cache informer,informer工厂模式。
3. handler实现
4. sync.map
5. 断言…