Kubernetes

Kubernetes in Action读书笔记第五章:Service资源对象

本章主要内容

  • 我们为什么需要Service资源对象
  • Service资源对象是什么
  • 如何创建和使用Service
  • Service和pod是如何关联的
  • 访问Service资源对象的两种方式:内部|外部
  • Service资源对象的分类
  • Service资源对象的底层实现原理是什么
  • Pod的readiness Probe
  • Service故障诊断tips

Contents

1 我们为什么需要Service资源对象?

想要解释清楚这个问题。首先,我们先回顾一下,当前在我们的Kubernetes cluster中,我们部署了诸多应用,它们最终是运行在Pod中的container中的。如果我们想要访问这些container提供的服务的话,我们该怎么办呢?可以通过container或者Pod的IP来访问吗?不能。因为:

Pod的生命周期是短暂的,有可能被重新调度,被重启;

Pod一旦被重启,它的IP地址将发生变化,显然通过它的IP来访问它提供的服务是不太现实的;

Pod有可能被水平扩展(向上变多,向下变少),那么究竟该选择哪个具体的Pod来访问服务呢?

如果选定某个Pod来访问服务的话,一旦这个Pod出现故障怎么办?单点故障避免不了;

水平扩展多个Pod来提供服务的负载均衡,访问某个特定的pod话,就没法实现负载均衡了。

因此,基于上述种种原因,我们不应该通过直接访问pod来访问它提供的服务。于是,人们在Pod的基础上封装出了一个资源对象,称之为Service。它可以解决上述种种问题。具体是怎么解决的呢?我们继续。

2 Service资源对象是什么

Service是我们在Kubernetes里创建的1个资源对象,用于封装1个或1组pod(把它们绑定在一起),Service自身拥有1个静态的IP地址和端口,我们通过访问Service来间接的访问它底层的Pod提供的服务。

A Kubernetes Service is a resource you create to make a single, constant point of entry to a group of pods providing the same service. Each service has an IP address and port that never change while the service exists. Clients can open connections to that IP and port, and those connections are then routed to one of the pods backing that service. This way, clients of a service don’t need to know the location of individual pods providing the service, allowing those pods to be moved around the cluster at any time.

P153

2.1 Service的一个典型应用场景举例

假定1个前端服务,由多个pod组成,1个后端的数据库给前端工程提供服务。此时,我们可以把前端的3个Pod封装在一起,抽象出1个Service对象出来,把它的IP地址和端口,开放给客户端访问,而不是让客户端直接找到我这3个pod。同时,后端的数据库也可以封装成1个服务,把它的IP和端口,开放给前端的那3个Pod去访问,而不是让pod直接来访问数据库的这个后端pod。

3 Service和pod是如何关联的?

既然Service是基于一个或一组pod封装出来的资源对象,且可以通过访问Service对象来间接的访问它底层的pods提供的服务的。那么它是怎么实现的呢?

答案是label selector。即Pod上有label,Service上有pod的label selector。这样就把它们关联在一起了,如同前面我们知道的ReplicationController、ReplicaSet、DaemonSet和Job以及CronJob都是通过pod selector和pod产生关联,一样的道理。

4 Service资源对象的两种访问方式

显然,pod提供的服务就可以给cluster中的其它pod访问,也可以提供给cluster外的应用访问。因此,可以把访问服务分为两类:内部访问,外部访问。

很显然,服务是有cluster中的pod提供的,那么想要访问它就可以分为两种方式:在cluster内部访问,cluster外部访问。

5 创建Service资源对象的两种方式

我们在前面的第二章 3.3章节学习了通过命令kubectl expose pod把pod以Service的形式暴露出去,也可以通过yaml文件来创建1个Service资源对象。我们通常采用yaml文件的方式来创建Service资源对象。

6 Service资源对象的分类

  • ClusterIP:默认类型的Service资源对象;只能在Kubernetes cluster内部通过cluster IP+端口来访问服务;
  • NodePort: ClusterIP类型的Service对象的超集,除了可以在Kubernetes cluster 内部通过Cluster IP+端口访问访问之外,还可以通过整个集群中的每个node的IP+端口来访问服务;
  • Loadbalancer:它是NodePort 类型Service资源对象的超集,除了可以通过整个集群中的每个node的IP+端口来访问服务之外,还可以通过一个公网IP地址来访问服务;
  • Ingress:它是Loadbalancer的超集,1个Loadbalancer只能对应1个Service资源对象;而1个Ingress可以对应多个不同的服务;

7如何创建并验证ClusterIP类型的Service资源对象

7.1先创建ReplicaSet,通过ReplicaSet间接创建pod出来

创建前,default namespace下,没有pod,也没有ReplicaSet,只有1个Kubernetes Cluster系统默认的Service,名为kubernetes,我们先不管它:

[root@master-node Chapter04]# kubectl get pods
No resources found in default namespace.
[root@master-node Chapter04]# kubectl get rs
No resources found in default namespace.
[root@master-node Chapter04]# kubectl get svc
NAME         TYPE        CLUSTER-IP   EXTERNAL-IP   PORT(S)   AGE
kubernetes   ClusterIP   10.96.0.1    <none>        443/TCP   14d
[root@master-node Chapter04]# 

使用第四章的kubia-replicaset.yaml 来创建ReplicaSet和pods:

[root@master-node Chapter04]# pwd
/root/kubernetes-in-action/Chapter04
[root@master-node Chapter04]# cat kubia-replicaset.yaml 
apiVersion: apps/v1
kind: ReplicaSet
metadata:
  name: kubia
spec:
  replicas: 3
  selector:
    matchLabels:
      app: kubia
  template:
    metadata:
      labels:
        app: kubia
    spec:
      containers:
      - name: kubia
        image: luksa/kubia
[root@master-node Chapter04]# kubectl apply -f kubia-replicaset.yaml 
replicaset.apps/kubia created
[root@master-node Chapter04]# kubectl get pods -owide --show-labels 
NAME          READY   STATUS    RESTARTS   AGE   IP             NODE     NOMINATED NODE   READINESS GATES   LABELS
kubia-4mwx6   1/1     Running   0          33s   10.244.2.198   node-2   <none>           <none>            app=kubia
kubia-htskq   1/1     Running   0          33s   10.244.2.199   node-2   <none>           <none>            app=kubia
kubia-nt72z   1/1     Running   0          33s   10.244.1.20    node-1   <none>           <none>            app=kubia
[root@master-node Chapter04]# kubectl get replicasets.apps 
NAME    DESIRED   CURRENT   READY   AGE
kubia   3         3         3       42s
[root@master-node Chapter04]# 

7.2 创建Service资源对象

[root@master-node Chapter05]# pwd
/root/kubernetes-in-action/Chapter05
[root@master-node Chapter05]# cat kubia-svc.yaml 
apiVersion: v1
kind: Service                               #类型为Service资源对象
metadata:
  name: kubia                               #Service名为kubia
spec:
  ports: 
  - port: 80                                #Service对外提供的端口是80
    targetPort: 8080                        #Pod端口是8080
  selector:
    app: kubia
[root@master-node Chapter05]# kubectl apply -f kubia-svc.yaml 
service/kubia created
[root@master-node Chapter05]# kubectl get svc
NAME         TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)   AGE
kubernetes   ClusterIP   10.96.0.1        <none>        443/TCP   14d
kubia        ClusterIP   10.101.149.183   <none>        80/TCP    21s
[root@master-node Chapter05]# 

7.3验证ClusterIP类型的Service资源对象

在当前Kubernetes Cluster内部,所有的节点上,都可以通过Service资源对象的Cluster IP+端口访问:

[root@master-node Chapter05]# kubectl get svc
NAME         TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)   AGE
kubernetes   ClusterIP   10.96.0.1        <none>        443/TCP   14d
kubia        ClusterIP   10.101.149.183   <none>        80/TCP    21s
[root@master-node Chapter05]# curl 10.101.149.183
You've hit kubia-nt72z
[root@master-node Chapter05]# curl 10.101.149.183
You've hit kubia-nt72z
[root@master-node Chapter05]# curl 10.101.149.183
You've hit kubia-4mwx6
[root@master-node Chapter05]# curl 10.101.149.183
You've hit kubia-4mwx6
[root@master-node Chapter05]# curl 10.101.149.183
You've hit kubia-htskq
[root@master-node Chapter05]# 

我们的Service对象名为kubia,它的Cluster-IP为10.101.149.183,访问端口是80。我们可以在当前Cluster的任意节点上通过该IP+端口访问。我们有可以看到curl 10.101.149.183每次访问返回的主机名不固定,因为该Service下其实是有ReplicaSet下的3个不同的pod提供的服务。

那么,离开当前的Kubernetes Cluster能不能访问该服务呢?

不能。这就是默认的Cluster IP类型的Service资源对象的限制特征,只能在当前Kubernetes Cluster内部访问。

8 如何创建并验证NodePort类型的Service资源对象

8.1 创建NodePort类型的Service资源对象

[root@master-node Chapter05]# cat kubia-svc-nodeport.yaml 
apiVersion: v1
kind: Service
metadata:
  name: kubia-nodeport
spec:
  type: NodePort
  ports:
  - port: 80                              #Service访问端口,80
    targetPort: 8080                      #底层Pod端口:8080
    nodePort: 30123                       #NodePort端口,即任意节点都通过该端口来访问服务,该参数可以不在这里指定,由Kubernetes帮我们选定一个端口          
  selector:
    app: kubia
[root@master-node Chapter05]# kubectl get svc
NAME         TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)   AGE
kubernetes   ClusterIP   10.96.0.1        <none>        443/TCP   14d
kubia        ClusterIP   10.101.149.183   <none>        80/TCP    167m
[root@master-node Chapter05]# kubectl apply -f kubia-svc-nodeport.yaml 
service/kubia-nodeport created
[root@master-node Chapter05]# 

我们仅仅创建了1个NodePort的Service资源对象,底层的Pod还是采用前面的示例的名为kubia的ReplicaSet管控的3个pod。

8.2验证NodePort Service资源对象

[root@master-node Chapter05]# kubectl get svc
NAME             TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)        AGE
kubernetes       ClusterIP   10.96.0.1        <none>        443/TCP        14d
kubia            ClusterIP   10.101.149.183   <none>        80/TCP         167m
kubia-nodeport   NodePort    10.110.90.145    <none>        80:30123/TCP   7s
[root@master-node Chapter05]# curl 10.110.90.145
You've hit kubia-htskq
[root@master-node Chapter05]# curl 10.110.90.145:80
You've hit kubia-4mwx6
[root@master-node Chapter05]# curl 172.16.11.168:30123
You've hit kubia-4mwx6
[root@master-node Chapter05]# curl 172.16.11.148:30123
You've hit kubia-4mwx6
[root@master-node Chapter05]# curl 172.16.11.161:30123
You've hit kubia-nt72z
[root@master-node Chapter05]# 

我们看到名为kubia-nordport的Service对象,它的Cluter IP是10.110.90.145,访问端口在80。同时,它的类型是NodePort,那么可以通过cluster中任意节点的IP加端口30123来访问服务。

接下来,我们可以通过cluster外的其它客户端来访问该NodePort类型的服务:

通过我的Mac客户端来访问:

$ curl 172.16.11.161:30123
You've hit kubia-nt72z
​
asher at MacBook-Air-3 in ~
$ 

这,验证了NodePort Service除了可以通过Node IP+端口访问服务,还可以通过Cluster IP+端口访问。因为它是Cluster IP的超集。

8.3 NodePort访问架构图

客户端通过NodePort类型服务的IP+端口30123访问服务,请求到每个node的30123端口,然后再把服务请求转发到运行在各个node上的pod的8080端口上去。

9 如何创建并验证Loadbalancer类型的Service资源对象

9.1创建Loadbalancer类型的Service资源对象

如果是通过本地部署的Kubernetes cluster,或者采用的minikube环境是无法验证该类型的Service资源对象。因为,它需要一个公网IP。而云服务商,(Google的GCE或者亚马逊、微软都有提供,Oracle 的Oracle Kubernetes Engine也有提供)。我这里,以我的Oracle Cloud环境来演示。

user@cloudshell:~ (ap-osaka-1)$ kubectl apply -f nginx_lb.yaml 
deployment.apps/my-nginx created
service/my-nginx-svc created
user@cloudshell:~ (ap-osaka-1)$ kubectl get svc
NAME           TYPE           CLUSTER-IP      EXTERNAL-IP   PORT(S)        AGE
kubernetes     ClusterIP      10.96.0.1       <none>        443/TCP        10h
my-nginx-svc   LoadBalancer   10.96.252.106   <pending>     80:32064/TCP   9s
user@cloudshell:~ (ap-osaka-1)$

过一会儿或者,Oracle Kubernetes Engine帮我创建了LoadBalancer:

user@cloudshell:~ (ap-osaka-1)$ kubectl get svc
NAME           TYPE           CLUSTER-IP      EXTERNAL-IP     PORT(S)        AGE
kubernetes     ClusterIP      10.96.0.1       <none>          443/TCP        10h
my-nginx-svc   LoadBalancer   10.96.252.106   168.138.55.28   80:32064/TCP   2m48s
user@cloudshell:~ (ap-osaka-1)$ kubectl get pods -owide
NAME                        READY   STATUS    RESTARTS   AGE     IP           NODE        NOMINATED NODE   READINESS GATES
my-nginx-5d59d67564-8lwh7   1/1     Running   0          3m21s   10.244.0.7   10.0.10.9   <none>           <none>
user@cloudshell:~ (ap-osaka-1)$ 

其中 nginx_lb.yaml文件如下:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-nginx
  labels:
    app: nginx
spec:
  replicas: 3
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - name: nginx
        image: renguzi/oke                  #换成了我自己的image,修改NGINX的index.html,已经推送至DockerHub
        ports:
        - containerPort: 80
---
apiVersion: v1
kind: Service
metadata:
  name: my-nginx-svc
  labels:
    app: nginx
spec:
  type: LoadBalancer                        #Service类型为Loadbalancer
  ports:
  - port: 80
  selector:
    app: nginx

9.2 验证Loadbalancer Service资源对象

通过Loadbalancer公网IP+端口(这里是80)。

通过互联网直接访问服务:http://168.138.55.28/

9.3访问Loadbalancer服务架构图

客户端通过Loadbalancer IP+端口访问服务,Loadbalancer先把收到的请求路由到后端的Node上,然后再到各个Node上的pod上。可以看到,Loadbalancer是NodePort的超集,服务请求到了Node之后,服务请求走的流程跟NodePort一样。

10 如何创建并验证Ingress类型的Service资源对象

10.1为什么需要Ingress服务

因为对于Loadbalancer而言,每一个后端的Service资源对象,都需要1个独立的Loadbalancer与之相对应。那么有没有后端有多个服务,前端可以通过一个公网IP+不同的路径来区分后端不同的服务呢?

有,那就是Ingress类型的Service资源对象。

One important reason is that each LoadBalancer service requires its own load balancer with its own public IP address, whereas an Ingress only requires one, even when providing access to dozens of services. When a client sends an HTTP request to the Ingress, the host and path in the request determine which service the request is forwarded to, as shown in figure 5.9.

10.2 创建部署ingress Service资源对象

参考:https://docs.oracle.com/en-us/iaas/Content/ContEng/Tasks/contengsettingupingresscontroller.htm

同样,在Oracle cloud kubernetes环境上,执行:

# 1 创建初始化配置信息:包含namespace,角色,权限,ingress controller等,该配置文件地址:https://raw.githubusercontent.com/kubernetes/ingress-nginx/controller-v0.44.0/deploy/static/provider/cloud/deploy.yaml
user@cloudshell:~ (ap-osaka-1)$ kubectl apply -f deploy.yaml
namespace/ingress-nginx created
serviceaccount/ingress-nginx created
configmap/ingress-nginx-controller created
clusterrole.rbac.authorization.k8s.io/ingress-nginx unchanged
clusterrolebinding.rbac.authorization.k8s.io/ingress-nginx unchanged
role.rbac.authorization.k8s.io/ingress-nginx created
rolebinding.rbac.authorization.k8s.io/ingress-nginx created
service/ingress-nginx-controller-admission created
service/ingress-nginx-controller created
deployment.apps/ingress-nginx-controller created
validatingwebhookconfiguration.admissionregistration.k8s.io/ingress-nginx-admission configured
serviceaccount/ingress-nginx-admission created
clusterrole.rbac.authorization.k8s.io/ingress-nginx-admission unchanged
clusterrolebinding.rbac.authorization.k8s.io/ingress-nginx-admission unchanged
role.rbac.authorization.k8s.io/ingress-nginx-admission created
rolebinding.rbac.authorization.k8s.io/ingress-nginx-admission created
job.batch/ingress-nginx-admission-create created
job.batch/ingress-nginx-admission-patch created
user@cloudshell:~ (ap-osaka-1)$ 
​
# 2 创建loadbalancer类型的Service:
user@cloudshell:~ (ap-osaka-1)$ cat cloud-generic.yaml 
kind: Service
apiVersion: v1
metadata:
  name: ingress-nginx
  namespace: ingress-nginx
  labels:
    app.kubernetes.io/name: ingress-nginx
    app.kubernetes.io/part-of: ingress-nginx
spec:
  type: LoadBalancer
  selector:
    app.kubernetes.io/name: ingress-nginx
    app.kubernetes.io/part-of: ingress-nginx
  ports:
    - name: http
      port: 80
      targetPort: http
    - name: https
      port: 443
      targetPort: https
user@cloudshell:~ (ap-osaka-1)$ kubectl apply -f cloud-generic.yaml
service/ingress-nginx created
user@cloudshell:~ (ap-osaka-1)$ 
#### 3  拿到ingress-nginx的公网IP: 152.69.205.99 
user@cloudshell:~ (ap-osaka-1)$ kubectl get svc -n ingress-nginx
NAME                                 TYPE           CLUSTER-IP      EXTERNAL-IP      PORT(S)                      AGE
ingress-nginx                        LoadBalancer   10.96.101.118   152.69.205.99    80:32489/TCP,443:32700/TCP   58s
ingress-nginx-controller             LoadBalancer   10.96.3.197     152.69.203.208   80:32481/TCP,443:32581/TCP   105s
ingress-nginx-controller-admission   ClusterIP      10.96.158.245   <none>           443/TCP                      105s
user@cloudshell:~ (ap-osaka-1)$ 
​
#### 4 #创建加密的TLS(Transport Layer Security) 的secret
openssl req -x509 -nodes -days 365 -newkey rsa:2048 -keyout tls.key -out tls.crt -subj "/CN=nginxsvc/O=nginxsvc"
user@cloudshell:~ (ap-osaka-1)$ openssl req -x509 -nodes -days 365 -newkey rsa:2048 -keyout tls.key -out tls.crt -subj "/CN=nginxsvc/O=nginxsvc"
Generating a 2048 bit RSA private key
************************************************************************************************************************************************************************************************************************************+++++
writing new private key to 'tls.key'
-----
user@cloudshell:~ (ap-osaka-1)$ 
​
###### 5 创建secret
kubectl create secret tls tls-secret --key tls.key --cert tls.crt
​
###### 6 创建后端应用的Service
user@cloudshell:~ (ap-osaka-1)$ cat hello-world-ingress.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: docker-hello-world
  labels:
    app: docker-hello-world
spec:
  selector:
    matchLabels:
      app: docker-hello-world
  replicas: 3
  template:
    metadata:
      labels:
        app: docker-hello-world
    spec:
      containers:
      - name: docker-hello-world
        image: scottsbaldwin/docker-hello-world:latest
        ports:
        - containerPort: 80
---
apiVersion: v1
kind: Service
metadata:
  name: docker-hello-world-svc
spec:
  selector:
    app: docker-hello-world
  ports:
    - port: 8088
      targetPort: 80
  type: ClusterIP 
user@cloudshell:~ (ap-osaka-1)$ kubectl create -f hello-world-ingress.yaml
deployment.apps/docker-hello-world created
service/docker-hello-world-svc created
user@cloudshell:~ (ap-osaka-1)$ 
​
#### 7 创建ingress资源
user@cloudshell:~ (ap-osaka-1)$ cat ingress.yaml 
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: hello-world-ing
  annotations:
    kubernetes.io/ingress.class: "nginx"
spec:
  tls:
  - secretName: tls-secret
  rules:
  - http:
      paths:
        - path: /
          pathType: Prefix
          backend:
            service:
              name: docker-hello-world-svc
              port:
                number: 8088
user@cloudshell:~ (ap-osaka-1)$ 
​
#### 8 访问服务:
user@cloudshell:~ (ap-osaka-1)$ kubectl get svc --all-namespaces
NAMESPACE       NAME                                 TYPE           CLUSTER-IP      EXTERNAL-IP      PORT(S)                      AGE
default         docker-hello-world-svc               ClusterIP      10.96.171.156   <none>           8088/TCP                     71m
default         kubernetes                           ClusterIP      10.96.0.1       <none>           443/TCP                      3d17h
default         my-nginx-svc                         LoadBalancer   10.96.16.101    168.138.55.28    80:32751/TCP                 3d5h
ingress-nginx   ingress-nginx                        LoadBalancer   10.96.101.118   152.69.205.99    80:32489/TCP,443:32700/TCP   73m
ingress-nginx   ingress-nginx-controller             LoadBalancer   10.96.3.197     152.69.203.208   80:32481/TCP,443:32581/TCP   74m
ingress-nginx   ingress-nginx-controller-admission   ClusterIP      10.96.158.245   <none>           443/TCP                      74m
kube-system     kube-dns                             ClusterIP      10.96.5.5       <none>           53/UDP,53/TCP,9153/TCP       3d17h
user@cloudshell:~ (ap-osaka-1)$ 
user@cloudshell:~ (ap-osaka-1)$ kubectl get ingress --all-namespaces
NAMESPACE   NAME              CLASS    HOSTS   ADDRESS          PORTS     AGE
default     hello-world-ing   <none>   *       152.69.203.208   80, 443   123m
user@cloudshell:~ (ap-osaka-1)$ 

在本地通过ingress-nginx-controller的公网IP可以访问服务,http://152.69.203.208/

$ curl http://152.69.203.208/
<h1>Hello webhook world from: docker-hello-world-5cf675d97d-2nr6q</h1>
​
asher at MacBook-Air-3 in ~/.ssh
$ curl -I http://152.69.203.208/
HTTP/1.1 200 OK
Date: Fri, 18 Mar 2022 08:49:42 GMT
Content-Type: text/html
Content-Length: 71
Connection: keep-alive
Last-Modified: Fri, 18 Mar 2022 07:35:50 GMT
ETag: "62343656-47"
Accept-Ranges: bytes
​
​
asher at MacBook-Air-3 in ~/.ssh
$ 

但是通过ingress-nginx的公网IP却不行:

$ curl http://152.69.205.99/
curl: (56) Recv failure: Connection reset by peer
​
asher at MacBook-Air-3 in ~/.ssh
$ curl -I http://152.69.205.99/
curl: (56) Recv failure: Connection reset by peer
​
asher at MacBook-Air-3 in ~/.ssh

通过HTTPS协议访问ingres-nginx-controller 的公网IP https://152.69.203.208/

$ curl -k https://152.69.203.208/
<h1>Hello webhook world from: docker-hello-world-5cf675d97d-l8rbz</h1>
​
asher at MacBook-Air-3 in ~/.ssh
$ curl -k https://152.69.203.208/
<h1>Hello webhook world from: docker-hello-world-5cf675d97d-g2btw</h1>
​
asher at MacBook-Air-3 in ~/.ssh
$ curl -k https://152.69.203.208/
<h1>Hello webhook world from: docker-hello-world-5cf675d97d-2nr6q</h1>
​
asher at MacBook-Air-3 in ~/.ssh
$ 

每次返回的地址都不同,说明负载均衡到后端的不同pods上了:

user@cloudshell:~ (ap-osaka-1)$ kubectl get pods
NAME                                  READY   STATUS    RESTARTS   AGE
docker-hello-world-5cf675d97d-2nr6q   1/1     Running   0          77m
docker-hello-world-5cf675d97d-g2btw   1/1     Running   0          77m
docker-hello-world-5cf675d97d-l8rbz   1/1     Running   0          77m
my-nginx-7746788f68-crcfq             1/1     Running   0          3d4h
my-nginx-7746788f68-gz87k             1/1     Running   0          3d4h
my-nginx-7746788f68-rx8hg             1/1     Running   0          3d4h
user@cloudshell:~ (ap-osaka-1)$ 

10.3 killercoda演示ingress类型的Service

10.3.1 实验场景要求和步骤:

Kubernetes cluster中已经安装配置NGINX ingress Controller,有了该Controller才可以创建类型为ingress的Service资源对象;

Kubernetes cluster中已经配置好不同的Service资源对象,指向底层不同的Deployment管控的不同的pod;

最后创建ingress资源对象,通过不同的上下文prefix,指向不同的Service资源对象。这样,通过访问ingress的不同上下文,把服务请求分发到底层不同的Service资源对象上。从而实现通过一个ingress类型的Service可以访问到多个不同的Service。

10.3.2 该环境下,已经安装配置NGINX ingress Controller:
controlplane $ kubectl get ns
NAME              STATUS   AGE
default           Active   28d
ingress-nginx     Active   47m
kube-node-lease   Active   28d
kube-public       Active   28d
kube-system       Active   28d
world             Active   47m
controlplane $ kubectl -n ingress-nginx get all
NAME                                            READY   STATUS      RESTARTS   AGE
pod/ingress-nginx-admission-create-9gtdp        0/1     Completed   0          47m
pod/ingress-nginx-admission-patch-jj22g         0/1     Completed   0          47m
pod/ingress-nginx-controller-7df574b57c-skncq   1/1     Running     0          47m
​
NAME                                         TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)                      AGE
service/ingress-nginx-controller             NodePort    10.111.219.156   <none>        80:30080/TCP,443:30443/TCP   47m
service/ingress-nginx-controller-admission   ClusterIP   10.103.86.21     <none>        443/TCP                      47m
​
NAME                                       READY   UP-TO-DATE   AVAILABLE   AGE
deployment.apps/ingress-nginx-controller   1/1     1            1           47m
​
NAME                                                  DESIRED   CURRENT   READY   AGE
replicaset.apps/ingress-nginx-controller-7df574b57c   1         1         1       47m
​
NAME                                       COMPLETIONS   DURATION   AGE
job.batch/ingress-nginx-admission-create   1/1           14s        47m
job.batch/ingress-nginx-admission-patch    1/1           14s        47m
controlplane $ 
10.3.3 当前world namespace的资源信息如下:
controlplane $ kubectl -n world get all
NAME                          READY   STATUS    RESTARTS   AGE
pod/asia-6b67487686-mdzvb     1/1     Running   0          20m
pod/asia-6b67487686-vkp7s     1/1     Running   0          20m
pod/europe-6676bc64c8-cmstv   1/1     Running   0          20m
pod/europe-6676bc64c8-zzjs4   1/1     Running   0          20m
​
NAME                     READY   UP-TO-DATE   AVAILABLE   AGE
deployment.apps/asia     2/2     2            2           20m
deployment.apps/europe   2/2     2            2           20m
​
NAME                                DESIRED   CURRENT   READY   AGE
replicaset.apps/asia-6b67487686     2         2         2       20m
replicaset.apps/europe-6676bc64c8   2         2         2       20m
controlplane $ kubectl -n world get all -owide
NAME                          READY   STATUS    RESTARTS   AGE   IP            NODE           NOMINATED NODE   READINESS GATES
pod/asia-6b67487686-mdzvb     1/1     Running   0          20m   192.168.0.8   controlplane   <none>           <none>
pod/asia-6b67487686-vkp7s     1/1     Running   0          20m   192.168.0.6   controlplane   <none>           <none>
pod/europe-6676bc64c8-cmstv   1/1     Running   0          20m   192.168.0.5   controlplane   <none>           <none>
pod/europe-6676bc64c8-zzjs4   1/1     Running   0          20m   192.168.0.7   controlplane   <none>           <none>
​
NAME                     READY   UP-TO-DATE   AVAILABLE   AGE   CONTAINERS   IMAGES                SELECTOR
deployment.apps/asia     2/2     2            2           20m   c            nginx:1.21.5-alpine   app=asia
deployment.apps/europe   2/2     2            2           20m   c            nginx:1.21.5-alpine   app=europe
​
NAME                                DESIRED   CURRENT   READY   AGE   CONTAINERS   IMAGES                SELECTOR
replicaset.apps/asia-6b67487686     2         2         2       20m   c            nginx:1.21.5-alpine   app=asia,pod-template-hash=6b67487686
replicaset.apps/europe-6676bc64c8   2         2         2       20m   c            nginx:1.21.5-alpine   app=europe,pod-template-hash=6676bc64c8
controlplane $ 
10.3.4 创建类型为cluster IP的Service资源对象:
controlplane $ cat asia-servcie.yaml 
apiVersion: v1
kind: Service                               #类型为Service资源对象
metadata:
  name: asia                               #Service名为asia
spec:
  ports: 
  - port: 80                                #Service对外提供的端口是80
    targetPort: 80                           #Pod端口是80
  selector:
    app: asia
controlplane $ 
​
controlplane $ cat europe-servcie.yaml 
apiVersion: v1
kind: Service                               #类型为Service资源对象
metadata:
  name: europe                               #Service名为europe
spec:
  ports: 
  - port: 80                                #Service对外提供的端口是80
    targetPort: 80                          #Pod端口是80
  selector:
    app: europe
controlplane $ 
​
controlplane $ kubectl -n world apply -f asia-servcie.yaml 
service/asia created
controlplane $ kubectl -n world apply -f europe-servcie.yaml 
service/europe created
controlplane $ 
10.3.5 验证Service资源对象:
controlplane $ kubectl -n world get all
NAME                          READY   STATUS    RESTARTS   AGE
pod/asia-6b67487686-mdzvb     1/1     Running   0          22m
pod/asia-6b67487686-vkp7s     1/1     Running   0          22m
pod/europe-6676bc64c8-cmstv   1/1     Running   0          22m
pod/europe-6676bc64c8-zzjs4   1/1     Running   0          22m
​
NAME             TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)   AGE
service/asia     ClusterIP   10.101.25.128    <none>        80/TCP    46s
service/europe   ClusterIP   10.107.123.252   <none>        80/TCP    42s
​
NAME                     READY   UP-TO-DATE   AVAILABLE   AGE
deployment.apps/asia     2/2     2            2           22m
deployment.apps/europe   2/2     2            2           22m
​
NAME                                DESIRED   CURRENT   READY   AGE
replicaset.apps/asia-6b67487686     2         2         2       22m
replicaset.apps/europe-6676bc64c8   2         2         2       22m
controlplane $ curl 10.101.25.128
hello, you reached ASIA
controlplane $ curl 10.107.123.252
hello, you reached EUROPE
controlplane $ 
10.3.6 创建ingress资源对象
controlplane $ cat world-service-ingress.yaml 
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: world
  namespace: world
spec:
  ingressClassName: nginx
  rules:
  - host: "world.universe.mine"
    http:
      paths:
      - path: /europe
        pathType: Prefix
        backend:
          service:
            name: europe
            port:
              number: 80
      - path: /asia
        pathType: Prefix
        backend:
          service:
            name: asia
            port:
              number: 80
controlplane $ 
controlplane $ kubectl apply -f world-service-ingress.yaml 
ingress.networking.k8s.io/world created
controlplane $ kubectl -n world get ingress
NAME    CLASS   HOSTS                 ADDRESS   PORTS   AGE
world   nginx   world.universe.mine             80      28s
controlplane $ 
10.3.7 通过ingress访问验证
controlplane $ kubectl -n world get ingress
NAME    CLASS   HOSTS                 ADDRESS   PORTS   AGE
world   nginx   world.universe.mine             80      28s
controlplane $ curl http://world.universe.mine:30080/europe/
hello, you reached EUROPE
controlplane $ curl http://world.universe.mine:30080/asia/  
hello, you reached ASIA
controlplane $ 

从上,我们看到通过ingress一个入口地址,可以访问到底层的不同Service资源对象上。

11 Service的底层实现原理是什么?

表面上看,我们说底层的pod上有label,然后Service通过label selector把它自己和底层的pod关联在了一起。然后通过访问Service的IP+端口就可以了。

11.1 那么Service和pod是直接关联在一起的吗?

其实不是。在它们二者中间还有一个抽象层endpoint。那么什么是endpoint呢?如何查看呢?

11.2 如何查看endpoints的信息?

我们可以通过kubectl describe svc service_name时,就可以看到endpoint信息,如下:

[root@master-node Chapter05]# kubectl get svc
NAME              TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)        AGE
kubernetes        ClusterIP   10.96.0.1        <none>        443/TCP        21d
kubia             ClusterIP   10.101.149.183   <none>        80/TCP         7d3h
kubia-nodeport    NodePort    10.110.90.145    <none>        80:30123/TCP   7d1h
[root@master-node Chapter05]# kubectl describe svc kubia-nodeport 
Name:                     kubia-nodeport
Namespace:                default
Labels:                   <none>
Annotations:              <none>
Selector:                 app=kubia
Type:                     NodePort
IP Family Policy:         SingleStack
IP Families:              IPv4
IP:                       10.110.90.145
IPs:                      10.110.90.145
Port:                     <unset>  80/TCP
TargetPort:               8080/TCP
NodePort:                 <unset>  30123/TCP
Endpoints:                10.244.1.20:8080,10.244.2.198:8080,10.244.2.199:8080     #这就是endpoints信息
Session Affinity:         None
External Traffic Policy:  Cluster
Events:                   <none>
[root@master-node Chapter05]# 

我们也可以直接查看endpoints的信息:

[root@master-node Chapter05]# kubectl get endpoints
NAME              ENDPOINTS                                              AGE
kubernetes        172.16.11.168:6443                                     21d
kubia             10.244.1.20:8080,10.244.2.198:8080,10.244.2.199:8080   7d4h
kubia-nodeport    10.244.1.20:8080,10.244.2.198:8080,10.244.2.199:8080   7d1h
[root@master-node Chapter05]# kubectl get endpoints kubia-nodeport 
NAME             ENDPOINTS                                              AGE
kubia-nodeport   10.244.1.20:8080,10.244.2.198:8080,10.244.2.199:8080   7d1h
[root@master-node Chapter05]# 

从上,我们看到两个不同的Service对象(kubia和kubia-nodeport)它们的endpoints其实是一样的,都指向了相同的pod ip+port。

11.3 endpoints的作用是什么?

当客户端请求找到Service的IP和端口时,并不是直接到底层的pod,而是先到endpoints,然后Service Proxy从endpoints中选择1个IP+port对儿,并把请求交给它。我们也可以看到endpoints里的信息,其实也就是对应的Service包含的后端pod的IP和端口。

[root@master-node Chapter05]# kubectl get pods -owide
NAME                 READY   STATUS    RESTARTS   AGE    IP             NODE     NOMINATED NODE   READINESS GATES
kubia-4mwx6          1/1     Running   0          7d4h   10.244.2.198   node-2   <none>           <none>
kubia-htskq          1/1     Running   0          7d4h   10.244.2.199   node-2   <none>           <none>
kubia-nt72z          1/1     Running   0          7d4h   10.244.1.20    node-1   <none>           <none>
[root@master-node Chapter05]# 

从另外一个方面,可以说即使Service是通过label selector和Pod发生关联的。但是,endpoints的引入,解耦了Service和Pod的关联。为啥?比如,cluster内部的应用如果想要访问外部的应用提供的服务,那么这时候在cluster内部创建的Service资源对象,是无法通过label selector和pod绑定的,再说了,外部的应用提供的服务,也不一定是通过pod提供的呀,有可能是一个普通的应用。

如果我们单独创建1个Service对象,没有label selector,则Kubernetes不会帮我们创建对应的endpoints对象。接下来,我们可以手工再去创建endpoints资源对象。此时,手工创建的endpoints资源名称必须要与前面创建的Service资源对象同名。否则,endpoints倒是可以创建成功,但是Service没有对应的endpoints,此时的Service是没有意义的

11.4 如何创建1个没有label selector的Service资源对象?

[root@master-node Chapter05]# cat my-external-service-without-label-selector.yaml 
apiVersion: v1
kind: Service
metadata:
  name: my-external-service
spec:
  ports:
  - port: 80
[root@master-node Chapter05]# kubectl apply -f my-external-service-without-label-selector.yaml 
service/my-external-service created
[root@master-node Chapter05]# kubectl get svc
NAME                  TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)        AGE
kubernetes            ClusterIP   10.96.0.1        <none>        443/TCP        21d
kubia                 ClusterIP   10.101.149.183   <none>        80/TCP         7d5h
kubia-nodeport        NodePort    10.110.90.145    <none>        80:30123/TCP   7d2h
my-external-service   ClusterIP   10.96.179.130    <none>        80/TCP         7s
[root@master-node Chapter05]# kubectl describe svc my-external-service 
Name:              my-external-service
Namespace:         default
Labels:            <none>
Annotations:       <none>
Selector:          <none>
Type:              ClusterIP
IP Family Policy:  SingleStack
IP Families:       IPv4
IP:                10.96.179.130
IPs:               10.96.179.130
Port:              <unset>  80/TCP
TargetPort:        80/TCP
Endpoints:         <none>
Session Affinity:  None
Events:            <none>
[root@master-node Chapter05]# 

如上,创建了1个名为my-external-service的Service资源对象,其并没有任何label selector,同时,也没有对应的endpoints信息,默认情况下,Service资源对象是ClusterIP类型的。

11.5 如何手工创建endpoints资源对象?

在创建之前,我们现在其它机器上启动一个NGINX container,模拟cluster外部的服务,待会儿我们通过endpoints把这个服务关联进来。最终,我们可以在cluster内部通过my-external-service访问外部的NGINX服务。

172.16.17.7启动1个NGINX container:

[root@guoxin7 ~]# hostname -I
172.16.17.7 172.17.0.1 192.168.49.1 
[root@guoxin7 ~]# docker run --name mynginx -p 8888:80 -itd nginx
Unable to find image 'nginx:latest' locally
...
Status: Downloaded newer image for docker.io/nginx:latest
2ac149aef6cf7a5a233878b64a25c707c358d6047d2daf5c9a52bf8bdc67a675
[root@guoxin7 ~]# docker ps
CONTAINER ID        IMAGE               COMMAND                  CREATED             STATUS              PORTS                  NAMES
2ac149aef6cf        nginx               "/docker-entrypoin..."   2 minutes ago       Up 2 minutes        0.0.0.0:8888->80/tcp   mynginx
[root@guoxin7 ~]# 

然后,在我们的cluster机器上确保可以访问到NGINX container服务:

[root@master-node Chapter05]# curl  http://172.16.17.7:8888
<!DOCTYPE html>
<html>
<head>
<title>Welcome to nginx!</title>
<style>
...
​

接下来创建我们的endpoints对象:

[root@master-node Chapter05]# pwd
/root/kubernetes-in-action/Chapter05
[root@master-node Chapter05]# cat my-external-service-endpoints.yaml 
apiVersion: v1
kind: Endpoints
metadata:
  name: my-external-service
subsets:
  - addresses:
    - ip: 172.16.17.7 
    ports:
    - port: 8888
[root@master-node Chapter05]# kubectl apply -f my-external-service-endpoints.yaml 
endpoints/my-external-service created
[root@master-node Chapter05]# kubectl describe svc my-external-service 
Name:              my-external-service
Namespace:         default
Labels:            <none>
Annotations:       <none>
Selector:          <none>
Type:              ClusterIP
IP Family Policy:  SingleStack
IP Families:       IPv4
IP:                10.96.179.130
IPs:               10.96.179.130
Port:              <unset>  80/TCP
TargetPort:        80/TCP
Endpoints:         172.16.17.7:8888                  #endpoints出现了在Service里
Session Affinity:  None
Events:            <none>
[root@master-node Chapter05]# kubectl get endpoints my-external-service 
NAME                  ENDPOINTS          AGE
my-external-service   172.16.17.7:8888   24s
[root@master-node Chapter05]# 

11.6 通过Service对象来访问外部服务

前面,我们看到Service的IP是10.96.179.130,它的端口是80。我们通过这个Service对象来访问cluster外部的NGINX服务。

[root@master-node Chapter05]# curl  http://10.96.179.130
<!DOCTYPE html>
<html>
<head>
<title>Welcome to nginx!</title>
...
</html>
[root@master-node Chapter05]#

发现可以正常访问外部的服务。此外,我们也可以从当前cluster中的pod内部来访问这个cluster外部的服务:

[root@master-node Chapter05]# kubectl get pods
NAME                 READY   STATUS    RESTARTS   AGE
kubia-4mwx6          1/1     Running   0          7d20h
kubia-htskq          1/1     Running   0          7d20h
kubia-nt72z          1/1     Running   0          7d20h
[root@master-node Chapter05]# kubectl exec kubia-4mwx6 -it -- /bin/bash
root@kubia-4mwx6:/# hostname
kubia-4mwx6
root@kubia-4mwx6:/# curl http://10.96.179.130
<!DOCTYPE html>
<html>
<head>
<title>Welcome to nginx!</title>
...

11.7 Service和endpoints的关系?

endpoints的出现,解耦了Service和Pod的关系。从describe Service命令中,也可以看到endpoints的信息。但是,endpoints并不隶属于Service资源对象,它是可以单独于Service而创建出来的。

endpoints的出现,用来解耦Service资源对象和pod是非常有必要的。为什么这么说呢?比方说,当我们想从Kubernetes cluster内部访问一个位于cluster外部的服务的时候,我们在cluster内部创建1个指向外部服务的Service资源对象时,我们是无法给这个Service资源对象指定pod selector,进而让其和提供此服务的pod对应的。我们怎么知道提供这个外部的服务程序是跑在pod上的,还是1个普通的应用程序呢?

12 Service discovery是怎么实现的?

至此,我们已经在Kubernetes cluster中创建了不同类型的Service,并且验证了可以正常访问这些服务。那么,Kubernetes cluster中的服务发现是如何实现的呢?

在Kubernetes中有两种方式来实现Service discovery:

  • 环境变量
  • 内置提供DNS功能的Service对象;

12.1 环境变量实现Service discovery

在Kubernetes里,每个pod里都有若干个环境变量,其中就有几个比较特殊的环境变量,用来标识系统中的服务对象的。{SVCNAME}SERVICE_HOST 表示服务IP;

{SVCNAME}_SERVICE_PORT表示服务端口;

其中{SVCNAME}表示系统中已有的服务名,比如系统中有个名为kubernetes的Service资源对象:

[root@master-node Chapter05]# kubectl get svc
NAME                  TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)        AGE
kubernetes            ClusterIP   10.96.0.1        <none>        443/TCP        23d      #已经运行23天
kubia                 ClusterIP   10.101.149.183   <none>        80/TCP         9d
kubia-nodeport        NodePort    10.110.90.145    <none>        80:30123/TCP   9d
my-external-service   ClusterIP   10.96.179.130    <none>        80/TCP         47h      #启动运行47小时
[root@master-node Chapter05]#

这里,看到kubernetes 的IP地址和端口分别是:10.96.0.1和443/TCP

那么pod的环境变量就是KUBERNETES_SERVICE_PORT和KUBERNETES_SERVICE_HOST。

我们进入1个pod中,来验证它的环境变量:

[root@master-node Chapter05]# kubectl get pods
NAME                 READY   STATUS    RESTARTS   AGE
kubia-4mwx6          1/1     Running   0          9d
kubia-htskq          1/1     Running   0          9d
kubia-nt72z          1/1     Running   0          9d  #才运行9天,后于kubernetes对象创建,所以它可以通过环境变量发现kubernets服务对象                       
[root@master-node Chapter05]# kubectl exec kubia-nt72z -it  -- /bin/bash
root@kubia-nt72z:/# env
NODE_VERSION=7.9.0
HOSTNAME=kubia-nt72z
KUBERNETES_PORT_443_TCP_PORT=443
KUBERNETES_PORT=tcp://10.96.0.1:443
TERM=xterm
KUBERNETES_SERVICE_PORT=443                      #服务的环境变量
KUBERNETES_SERVICE_HOST=10.96.0.1                #服务的环境变量
NPM_CONFIG_LOGLEVEL=info
PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
PWD=/
SHLVL=1
HOME=/root
YARN_VERSION=0.22.0
KUBERNETES_PORT_443_TCP_PROTO=tcp
KUBERNETES_SERVICE_PORT_HTTPS=443
KUBERNETES_PORT_443_TCP_ADDR=10.96.0.1
KUBERNETES_PORT_443_TCP=tcp://10.96.0.1:443
_=/usr/bin/env
root@kubia-nt72z:/# 

因此,我们说pod kubia-nt72z通过它的环境变量发现了服务kubernetes。

但是,这种方式有一个弊端,那就是,要求Service必须要先于pod创建出来,后创建的pod才知道已经存在的Service对象。这里的例子看到kubernetes先于kubia-nt72z创建出来,它可以被pod通过环境变量来发现服务没问题。

比如上例中,我们在它的环境变量中,并没有发现my-external-service这个服务对象的环境变量。但是,我们再手工创建一个pod(/root/kubernetes-in-action/Chapter03/kubia-manual.yaml)出来,它的创建要晚于my-external-service这个服务对象,那么它的环境变量中,就会包含进来了。

[root@master-node Chapter05]# kubectl get pods
NAME                 READY   STATUS    RESTARTS   AGE
kubia-4mwx6          1/1     Running   0          9d
kubia-htskq          1/1     Running   0          9d
kubia-manual         1/1     Running   0          5h39m #我晚于my-external-service创建出来,所有我的环境变量里有它
kubia-nt72z          1/1     Running   0          9d
[root@master-node Chapter05]# kubectl exec kubia-manual -it -- env
PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
HOSTNAME=kubia-manual
TERM=xterm
KUBIA_NODEPORT_SERVICE_PORT=80
KUBIA_NODEPORT_PORT=tcp://10.110.90.145:80
....
MY_EXTERNAL_SERVICE_SERVICE_HOST=10.96.179.130      #my-external-service的环境变量  
MY_EXTERNAL_SERVICE_PORT_80_TCP_PROTO=tcp
KUBERNETES_PORT_443_TCP_ADDR=10.96.0.1
MY_EXTERNAL_SERVICE_PORT_80_TCP_ADDR=10.96.179.130
...

如果pod比Service先创建出来,那么pod如何发现Service呢?答案是环境变量的方式行不通了。

12.2 内置提供DNS功能的Service对象

通过环境变量的方式,在Kubernetes内部提供服务发现的功能有所限制。于是,Kubernetes的人们研究出了另外一种方式,通过内置一个提供DNS功能的玩意儿,cluster中,一旦新创建1个Service资源对象,你就向我这边注册过来,将来新创建的所有pod,都到我这儿来找服务对象,就能找到你这个Service了。

我们来看一下它实现的原理:

首先,在kube-system中内置1个名为kube-dns的Service资源对象,它随着Kubernetes cluster的启动而启动:

[root@master-node Chapter05]# kubectl get svc -n kube-system 
NAME       TYPE        CLUSTER-IP   EXTERNAL-IP   PORT(S)                  AGE
kube-dns   ClusterIP   10.96.0.10   <none>        53/UDP,53/TCP,9153/TCP   75d
[root@master-node Chapter05]# 

然后,系统中所有新建的pod,都在它的/etc/resolv.conf文件中,指向该kube-dns的Service资源对象的cluster IP:

[root@master-node Chapter05]# kubectl get pods
NAME                 READY   STATUS    RESTARTS   AGE
kubia-4mwx6          1/1     Running   0          9d
kubia-htskq          1/1     Running   0          9d
kubia-manual         1/1     Running   0          6h4m
kubia-nt72z          1/1     Running   0          9d
[root@master-node Chapter05]# kubectl exec kubia-nt72z -- cat /etc/resolv.conf
nameserver 10.96.0.10
search default.svc.cluster.local svc.cluster.local cluster.local
options ndots:5
[root@master-node Chapter05]# kubectl exec kubia-manual -- cat /etc/resolv.conf
nameserver 10.96.0.10
search default.svc.cluster.local svc.cluster.local cluster.local
options ndots:5
[root@master-node Chapter05]# 

从上,我们看到,所有的pod的/etc/resolv.conf文件都指向了10.96.0.10。

我们再来分析一下kube-dns这个Service资源对象:kubectl describe -n kube-system svc kube-dns

[root@master-node Chapter05]# kubectl describe -n kube-system svc kube-dns 
Name:              kube-dns
Namespace:         kube-system
Labels:            k8s-app=kube-dns
                   kubernetes.io/cluster-service=true
                   kubernetes.io/name=CoreDNS
Annotations:       prometheus.io/port: 9153
                   prometheus.io/scrape: true
Selector:          k8s-app=kube-dns
Type:              ClusterIP
IP Family Policy:  SingleStack
IP Families:       IPv4
IP:                10.96.0.10
IPs:               10.96.0.10
Port:              dns  53/UDP
TargetPort:        53/UDP
Endpoints:         10.244.1.3:53,10.244.2.210:53
Port:              dns-tcp  53/TCP
TargetPort:        53/TCP
Endpoints:         10.244.1.3:53,10.244.2.210:53
Port:              metrics  9153/TCP
TargetPort:        9153/TCP
Endpoints:         10.244.1.3:9153,10.244.2.210:9153
Session Affinity:  None
Events:            <none>
[root@master-node Chapter05]# 

这里,我们看到原来kube-system下的kube-dns这个Service自己也是通过endpoints来和10.244.1.3:53,10.244.2.210:53关联起来的。那么这个endpoints(10.244.1.3:53,10.244.2.210:53)指向的是什么呢?

[root@master-node Chapter05]# kubectl get pods -n kube-system -owide
NAME                                  READY   STATUS    RESTARTS          AGE   IP              NODE          NOMINATED NODE   READINESS GATES
coredns-6d8c4cb4d-bzlc5               1/1     Running   0                 26h   10.244.2.210    node-2        <none>           <none>
coredns-6d8c4cb4d-nnbxg               1/1     Running   0                 75d   10.244.1.3      node-1        <none>           <none>
...

原来是运行在kube-system下的2个pods。再看一下这个pod的详细信息:

[root@master-node Chapter05]# kubectl describe pod -n kube-system coredns-6d8c4cb4d-bzlc5 
Name:                 coredns-6d8c4cb4d-bzlc5
Namespace:            kube-system
Priority:             2000000000
Priority Class Name:  system-cluster-critical
Node:                 node-2/172.16.11.161
Start Time:           Wed, 23 Mar 2022 11:40:55 +0800
Labels:               k8s-app=kube-dns
                      pod-template-hash=6d8c4cb4d
Annotations:          <none>
Status:               Running
IP:                   10.244.2.210
IPs:
  IP:           10.244.2.210
Controlled By:  ReplicaSet/coredns-6d8c4cb4d         #pod被上层的ReplicaSet管控
Containers:
....

从上看到,它是属于ReplicaSet/coredns-6d8c4cb4d 管控的:

[root@master-node Chapter05]# kubectl get rs -n kube-system 
NAME                DESIRED   CURRENT   READY   AGE
coredns-6d8c4cb4d   2         2         2       75d
[root@master-node Chapter05]# kubectl describe rs -n kube-system 
Name:           coredns-6d8c4cb4d
Namespace:      kube-system
Selector:       k8s-app=kube-dns,pod-template-hash=6d8c4cb4d
Labels:         k8s-app=kube-dns
                pod-template-hash=6d8c4cb4d
Annotations:    deployment.kubernetes.io/desired-replicas: 2
                deployment.kubernetes.io/max-replicas: 3
                deployment.kubernetes.io/revision: 1
Controlled By:  Deployment/coredns               #ReplicaSet被上层的Deployment管控
Replicas:       2 current / 2 desired
Pods Status:    2 Running / 0 Waiting / 0 Succeeded / 0 Failed
Pod Template:
  Labels:           k8s-app=kube-dns
                    pod-template-hash=6d8c4cb4d

而ReplicaSet/coredns-6d8c4cb4d 这个ReplicaSet又被Deployment/coredns管控:

[root@master-node Chapter05]# kubectl get deployments.apps  -n kube-system 
NAME      READY   UP-TO-DATE   AVAILABLE   AGE
coredns   2/2     2            2           75d
[root@master-node Chapter05]# kubectl describe deployments.apps -n kube-system coredns 
Name:                   coredns
Namespace:              kube-system
CreationTimestamp:      Fri, 07 Jan 2022 17:11:49 +0800
Labels:                 k8s-app=kube-dns
Annotations:            deployment.kubernetes.io/revision: 1
Selector:               k8s-app=kube-dns
Replicas:               2 desired | 2 updated | 2 total | 2 available | 0 unavailable
StrategyType:           RollingUpdate

到了Deployment层就结束了。我们暂时还没有学习和了解Kubernetes里的Deployment,我们在后面会单独研究它。

总结一下,Kubernetes通过内置的dns实现服务发现的方式是这样的,起了一个Deployment名为coredns,它管控了一个ReplicaSet:coredns-6d8c4cb4d,这个ReplicaSet管控了2个pod:coredns-6d8c4cb4d-bzlc5和coredns-6d8c4cb4d-nnbxg 。然后,创建了一个cluster IP类型的Service对象叫做:kube-dns,一个同名的endpoints,指向了那2个pods。

从上,看到coredns-6d8c4cb4d-bzlc5和coredns-6d8c4cb4d-nnbxg 的生命周期不同,1个是26h,一个是75d。其中一个是我特意手工杀掉之后,由Deployment自动帮我启动的一个新的pod出来之后的结果。这也验证了Kubernetes的强大,

13 Service的sessionAffinity

13.1 什么是Service的sessionAffinity?

从名字上,我们大概猜测到,就是给Service资源对象设置”会话亲和性”。具体啥意思呢?就是说,当客户端会话每次请求访问我这个Service对象的时候,我每次都是以某个特定的pod来响应你,还是从我这个Service对象中的endpoints里任意挑选1个pod来响应你。

“碗里有3个蛋,每次都随机挑选1个吃,还是每次都吃最大的那一个。”

13.2 sessionAffinity的作用和应用场景?

  • 作用:简而言之,就是可以限定客户端的服务请求,每次都打到后端的某个特定的pod上,而不是随机一个pod来响应来自同一个客户端的请求。
  • 应用场景:对于某些特定的load balancer的场景,我们希望特定的会话每次都到特定的后端服务,可以命中缓存,提高效率。

13.3 Service支持的sessionAffinity类型有哪些?

Kubernetes里的Service目前只支持两种类型的sessionAffinity:None(默认值),ClientIP。

13.4不同类型的sessionAffinity的区别是什么?

默认情况下None类型的sessionAffinity,表示Service资源对象,不对特定客户端发起的Service请求进行限定和区分,而是随机挑选1个pod响应;

ClientIP:服务端会根据客户端的IP,经过特定的算法,从endpoints中选定1个pod响应该会话,而且以后每次该客户端发来的请求,都是该pod来响应。当然,不同IP的client发起的请求,可能由不同的pod来响应。

13.5 如何查看|修改|设置、验证Service的sessionAffinity?

查看Service的sessionAffinity:

[root@master-node ~]# kubectl get svc
NAME                  TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)        AGE
kubernetes            ClusterIP   10.96.0.1        <none>        443/TCP        24d
kubia                 ClusterIP   10.101.149.183   <none>        80/TCP         10d
kubia-nodeport        NodePort    10.110.90.145    <none>        80:30123/TCP   10d
my-external-service   ClusterIP   10.96.179.130    <none>        80/TCP         2d22h
[root@master-node ~]# kubectl describe svc kubia
Name:              kubia
Namespace:         default
Labels:            <none>
Annotations:       <none>
Selector:          app=kubia
Type:              ClusterIP
IP Family Policy:  SingleStack
IP Families:       IPv4
IP:                10.101.149.183
IPs:               10.101.149.183
Port:              <unset>  80/TCP
TargetPort:        8080/TCP
Endpoints:         10.244.1.20:8080,10.244.2.198:8080,10.244.2.199:8080
Session Affinity:  None              #这里就是Service的sessionAffinity,默认值是None。
Events:            <none>
[root@master-node ~]# 

验证None类型的sessionAffinity,每次请求都到不同的pod上去了:

[root@master-node ~]# curl http://10.101.149.183
You've hit kubia-htskq
[root@master-node ~]# curl http://10.101.149.183
You've hit kubia-4mwx6
[root@master-node ~]# curl http://10.101.149.183
You've hit kubia-nt72z
[root@master-node ~]# kubectl get pods
NAME                 READY   STATUS    RESTARTS   AGE
kubia-4mwx6          1/1     Running   0          10d
kubia-htskq          1/1     Running   0          10d
kubia-manual         1/1     Running   0          28h
kubia-nt72z          1/1     Running   0          10d
[root@master-node ~]# 

修改|设置Service的sessionAffinity为ClientIP:

[root@master-node ~]# kubectl edit svc kubia
...
  selector:
    app: kubia
  sessionAffinity: ClientIP       #None改为ClientIP,保存退出即可
  type: ClusterIP
...  

验证ClientIP类型的sessionAffinity,每次请求都到同1个pod上去了:

[root@master-node ~]# kubectl describe svc kubia
Name:              kubia
...
IP:                10.101.149.183
IPs:               10.101.149.183
Port:              <unset>  80/TCP
TargetPort:        8080/TCP
Endpoints:         10.244.1.20:8080,10.244.2.198:8080,10.244.2.199:8080
Session Affinity:  ClientIP              #类型为SessionAffinity
Events:            <none>
[root@master-node ~]# curl http://10.101.149.183
You've hit kubia-htskq
[root@master-node ~]# curl http://10.101.149.183
You've hit kubia-htskq
[root@master-node ~]# curl http://10.101.149.183
You've hit kubia-htskq
[root@master-node ~]# 
​
#从其它client向kubia发起Service请求:
[root@node-1 ~]#  curl http://10.101.149.183
You've hit kubia-nt72z
[root@node-1 ~]#  curl http://10.101.149.183
You've hit kubia-nt72z
[root@node-1 ~]#  curl http://10.101.149.183
You've hit kubia-nt72z
[root@node-1 ~]# 

对于类型为ClientIP的sessionAffinity,不同IP的客户端,访问到的pod不同,但是同1个IP的client,每次请求都到同1个pod。

14 Service的externalTrafficPolicy属性

前面,我们介绍了Service的sessionAffinity属性,它对所有类型(cluster IP,node port,Loadbalancer,ingress)的Service都生效。接下来,我们来看Service的另外一种属性,externalTrafficPolicy,它只针对node port及以上类型的Service资源对象有效。

14.1 externalTrafficPolicy的作用?

对于pod中的container,如果它提供的服务被来自于cluster内部的其它pod访问的话,它是知道对方的IP地址的。如果,来自于cluster外部的客户端访问Service(访问node port或者Loadbalancer类型的Service),那么底层的pod中的container是看不到访问源头的IP地址的。此时,如果想要看到访问源头的IP地址的话,需要设置Service对象的externalTrafficPolicy属性。

14.2 externalTrafficPolicy的类型有哪些?

  • Cluster:默认值。看到的访问源头IP不准确,有可能会导致服务请求的二次跳转,但是可以提供较好的Load balance,即会把访问请求均衡的分散到后端的pod上,不会导致服务请求失败
  • Local:保留了访问源头的IP地址,不会导致服务请求的二次跳转,不能保证较好的load balance,可能导致服务请求失败。

14.3 externalTrafficPolicy的生效范围

它只对node port及以上类型的Service资源对象有效。cluster IP类型的Service没有该属性值。

14.4 externalTrafficPolicy举例说明

  • 场景1:当前有2个node:node1,node2 ;3个pods:pod-a,pod-b,pod-c,其中pod-a和pod-b运行在node1上,pod-c运行在node2上,我们的服务类型是node port的。

当externalTrafficPolicy=Local,我们通过node_ip+端口的方式访问服务的话,则可能出现node1和node2分别承担50%的请求,但是到pod上,则是pod-a和pod-b各承担25%,pod-c承担了50%。

当externalTrafficPolicy=Cluster时,则不存在上述问题。

  • 场景2:另外,当我们的3个如果都在node1上,node2上没有运行pod。

如果externalTrafficPolicy=Local,当通过node2的IP+端口来访问服务时,则出现尴尬,发现访问服务失败,请求hang住,因为本地node上没有任何pod;

如果externalTrafficPolicy=Local,当通过node2的IP+端口来访问服务时,则访问成功,但是此时,底层服务调用发生了二次跳转,因为本地node2没有pod,转而把请求转到node1上的pod。

14.5 如何查看|修改|设置|验证Service的externalTrafficPolicy

查看:

[root@master-node ~]# kubectl get svc 
NAME                  TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)        AGE
kubernetes            ClusterIP   10.96.0.1        <none>        443/TCP        24d
kubia                 ClusterIP   10.101.149.183   <none>        80/TCP         10d
kubia-nodeport        NodePort    10.110.90.145    <none>        80:30123/TCP   10d
my-external-service   ClusterIP   10.96.179.130    <none>        80/TCP         3d
[root@master-node ~]# kubectl describe svc kubia-nodeport 
Name:                     kubia-nodeport
Namespace:                default
Labels:                   <none>
Annotations:              <none>
Selector:                 app=kubia
Type:                     NodePort
IP Family Policy:         SingleStack
IP Families:              IPv4
IP:                       10.110.90.145
IPs:                      10.110.90.145
Port:                     <unset>  80/TCP
TargetPort:               8080/TCP
NodePort:                 <unset>  30123/TCP
Endpoints:                10.244.1.20:8080,10.244.2.198:8080,10.244.2.199:8080
Session Affinity:         None
External Traffic Policy:  Cluster              #nodeport类型Service的externalTrafficPolicy默认值是cluster
Events:                   <none>
[root@master-node ~]# kubectl get pods -owide   #3个pod分别在node1和node2上,master node上没有任何pod
NAME                 READY   STATUS    RESTARTS   AGE    IP             NODE     NOMINATED NODE   READINESS GATES
kubia-4mwx6          1/1     Running   0          10d    10.244.2.198   node-2   <none>           <none>
kubia-htskq          1/1     Running   0          10d    10.244.2.199   node-2   <none>           <none>
kubia-manual         1/1     Running   0          30h    10.244.2.209   node-2   <none>           <none>
kubia-nt72z          1/1     Running   0          10d    10.244.1.20    node-1   <none>           <none>
[root@master-node ~]# 

此时,我们的客户端可以通过当前cluster的任意节点IP+port来访问服务:

$ curl  http://172.16.11.161:30123
You've hit kubia-nt72z
​
asher at MacBook-Air-3 in ~
$ curl  http://172.16.11.168:30123
You've hit kubia-4mwx6
​
asher at MacBook-Air-3 in ~
$ curl  http://172.16.11.148:30123
You've hit kubia-nt72z
​
asher at MacBook-Air-3 in ~
$ 
​

修改:

[root@master-node ~]# kubectl edit svc kubia-nodeport 
...
externalTrafficPolicy: Local       #改为Local,保存退出。
...

验证externalTrafficPolicy: Local :

$ curl  http://172.16.11.148:30123
You've hit kubia-nt72z
​
asher at MacBook-Air-3 in ~
$ curl  http://172.16.11.161:30123
You've hit kubia-htskq
​
asher at MacBook-Air-3 in ~
$ curl  http://172.16.11.168:30123
curl: (28) Failed to connect to 172.16.11.168 port 30123: Operation timed out
# 此时,服务请求被挂住了,

此时,由于底层的3个pod分别运行在2个worker node上,没有任何pod运行在master node 172.16.11.168机器上,那么通过curl http://172.16.11.168:30123 将被阻塞,挂住了。

14.6 参考

https://kubernetes.io/docs/tasks/access-application-cluster/create-external-load-balancer/#preserving-the-client-source-ip

15 Pod的readiness probe

在前面的第四章,我们学习和使用了pod的liveness probe来保证pod是健康运行状态的。接下来,我们看pod的另外一种probe:readiness probe。

15.1 我们为什么需要readiness probe及它的使用场景是什么

我们知道了,Service通过label selector把满足条件的pod当作它的endpoints,这样就形成了访问Service时,其实是访问底层的pod的架构。那么试想,如果当前Service的endpoints里添加了1个新的pod,在这个pod还没有完全启动成功,如果有客户端的访问请求被路由到这个新的pod上了。结果会怎样呢?

结果就是,在pod完全启动结束之前,这个客户端的请求被hang住了,好像并没有得到任何响应一样。我们当然不希望这种情况出现。我们希望这时的客户端请求要么被路由到已有的pod上,有么在新的pod完全启动成功之前,不响应客户端的服务请求。

于是,Kubernetes的开发者引入了一个新的readiness probe,用于检测该pod是否正常启动并可以对外提供服务。用来确保pod在启动结束之前,不对外提供服务。

15.2 readiness probe的3种工作方式

跟前面提到的pod中container的liveness probe一样的3种工作方式:

  • HTTP Get probe:向pod中的container发送一个HTTP请求,如果返回正常,就认为container是正常的,该pod可以接受服务请求;否则,不能接受服务请求;
  • TCP socket probe:跟HTTP类似,向pod中的container建立1个TCP socket连接,成功则pod可以接受服务;否则,不行;
  • Exec probe:在pod中的container里执行某个特定的命令,根据命令执行结果是否成功来判断pod是否可以接收服务请求;

15.3 如何创建|使用|验证带有readiness probe的pod

在pod的container中设置readiness probe跟设置liveness probe的方式几乎没有差别,无非是把关键字从livenessProbe改为readinessProbe。

先看一个container中有readinessProbe的ReplicaSet的代码:

[root@master-node Chapter05]# pwd
/root/kubernetes-in-action/Chapter05
[root@master-node Chapter05]# cat kubia-rc-readinessprobe.yaml 
apiVersion: v1
kind: ReplicationController
metadata:
  name: kubia
spec:
  replicas: 3
  selector:
    app: kubia
  template:
    metadata:
      labels:
        app: kubia
    spec:
      containers:
      - name: kubia
        image: luksa/kubia
        ports:
        - name: http
          containerPort: 8080
        readinessProbe:        #container中probe的关键字
          exec:                #exec类型的探测方式,如果container中ls /var/ready命令执行成功,则container可以接收服务请求,否则,container不是ready状态
            command:
            - ls
            - /var/ready
[root@master-node Chapter05]# 

创建这个ReplicationController,并查看pod的状态:

[root@master-node Chapter05]# kubectl get rc
No resources found in default namespace.
[root@master-node Chapter05]# kubectl apply -f kubia-rc-readinessprobe.yaml 
replicationcontroller/kubia created
[root@master-node Chapter05]# kubectl get rc
NAME    DESIRED   CURRENT   READY   AGE
kubia   3         3         0       5s
[root@master-node Chapter05]# kubectl get pods
NAME                 READY   STATUS    RESTARTS   AGE
kubia-4mwx6          1/1     Running   0          11d
kubia-76v4r          0/1     Running   0          8s
kubia-8chrc          0/1     Running   0          8s
kubia-htskq          1/1     Running   0          11d
kubia-manual         1/1     Running   0          2d3h
kubia-nt72z          1/1     Running   0          11d
kubia-wfpc4          0/1     Running   0          8s
[root@master-node Chapter05]# 

这里,看到pods的READY列中,始终没有正常的。原因何在,新建的这3个pod,它的container中的readiness probe执行失败,所以,看到pod的状态不是ready的。

执行kubectl describe pod kubia-76v4r 来验证:

[root@master-node Chapter05]# kubectl describe pod kubia-76v4r 
Name:         kubia-76v4r
Namespace:    default
Priority:     0
Node:         node-2/172.16.11.161
Start Time:   Fri, 25 Mar 2022 15:04:34 +0800
...
Events:
  Type     Reason     Age                 From               Message
  ----     ------     ----                ----               -------
  Normal   Scheduled  2m5s                default-scheduler  Successfully assigned default/kubia-76v4r to node-2
  Normal   Pulling    2m4s                kubelet            Pulling image "luksa/kubia"
  Normal   Pulled     2m1s                kubelet            Successfully pulled image "luksa/kubia" in 2.777117726s
  Normal   Created    2m1s                kubelet            Created container kubia
  Normal   Started    2m1s                kubelet            Started container kubia
  Warning  Unhealthy  6s (x15 over 118s)  kubelet            Readiness probe failed: ls: cannot access /var/ready: No such file or directory
[root@master-node Chapter05]# 

Warning Unhealthy 6s (x15 over 118s) kubelet Readiness probe failed: ls: cannot access /var/ready: No such file or directory说明,该pod不是正常状态的原因。

我们进入该pod中,手工touch对应的文件:

[root@master-node Chapter05]# kubectl exec kubia-76v4r -it -- /bin/bash
root@kubia-76v4r:/# ls -l /var/ready
ls: cannot access /var/ready: No such file or directory
root@kubia-76v4r:/# touch /var/ready
root@kubia-76v4r:/# exit
exit
[root@master-node Chapter05]# kubectl get pods kubia-76v4r 
NAME          READY   STATUS    RESTARTS   AGE
kubia-76v4r   1/1     Running   0          4m48s
[root@master-node Chapter05]# kubectl get pods
NAME                 READY   STATUS    RESTARTS   AGE
kubia-4mwx6          1/1     Running   0          11d
kubia-76v4r          1/1     Running   0          4m52s              #此时,pod状态变为ready
kubia-8chrc          0/1     Running   0          4m52s
kubia-htskq          1/1     Running   0          11d
kubia-manual         1/1     Running   0          2d3h
kubia-nt72z          1/1     Running   0          11d
kubia-wfpc4          0/1     Running   0          4m52s
[root@master-node Chapter05]# 

15.4 readiness和liveness probe的区别和联系

liveness probe如果发现container异常,它会主动杀死并重启container;而readiness probe既不会杀死container,也不会重启container;

liveness probe不会等待readiness probe执行成功之后再开始探测container是否健康;如果想要控制liveness probe在container启动一会儿之后,再来执行探测container是否健康运行的话,那么需要设置一个liveness probe的参数,initialDelaySeconds。

15.5 官方文档参考Define readiness probes

官方文档参考:Define readiness probes

16 Service故障诊断tips

16.1 永远不要去ping service_ip

通常我们诊断网络是否可达的方式是ping target_IP;但是,在Kubernetes里,我们去验证某个服务是否可用正常访问的时候,不要第一时间想到ping service_ip;那样的话,你永远ping不通,而要用curl service_ip;比如:

[root@master-node ~]# kubectl get svc
NAME                  TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)        AGE
kubernetes            ClusterIP   10.96.0.1        <none>        443/TCP        25d
kubia                 ClusterIP   10.101.149.183   <none>        80/TCP         11d
kubia-nodeport        NodePort    10.110.90.145    <none>        80:30123/TCP   11d
my-external-service   ClusterIP   10.96.179.130    <none>        80/TCP         3d23h
[root@master-node ~]# ping 10.101.149.183
PING 10.101.149.183 (10.101.149.183) 56(84) bytes of data.
^C
--- 10.101.149.183 ping statistics ---
2 packets transmitted, 0 received, 100% packet loss, time 1000ms
​
[root@master-node ~]# curl 10.101.149.183
You've hit kubia-nt72z
[root@master-node ~]# 

再比如,我在前面创建的Oracle cloud Kubernetes环境的那个Loadbalancer IP依然这样,ping不通,但是可以正常访问服务。

$ ping 168.138.55.28
PING 168.138.55.28 (168.138.55.28): 56 data bytes
Request timeout for icmp_seq 0
Request timeout for icmp_seq 1
^C
--- 168.138.55.28 ping statistics ---
3 packets transmitted, 0 packets received, 100.0% packet loss
​
asher at MacBook-Air-3 in ~
$ curl 168.138.55.28
<!DOCTYPE html>
<html>
<head>
  <meta charset="utf-8">
<title>Welcome to Oracle Kubernetes Engine!</title>
<style>
html { color-scheme: light dark; }
body { width: 35em; margin: 0 auto;
font-family: Tahoma, Verdana, Arial, sans-serif; }
</style>
</head>
<body>
<h1>Oracle Kubernetes Engine!</h1>
<p>部署在Oracle云亚太区数据中心上的Kubernetes LoadBalancer服务</p>
​
<p>触手可及,一起来玩
<a href="http://www.knockatdatabase.com/">by 数据库敲门人</a>.<br/>
</p>
</body>
</html>
​
asher at MacBook-Air-3 in ~
$ 

16.2 Service不可访问的可能原因

是否是正确的访问Service的port,而不是去访问其底层的pod的port;

底层pod是否在Service的endpoints列表里;

底层pod的readiness probe是否执行成功,即pod是否是ready状态;

17 本地Kubernetes cluster安装配置使用ingress

17.1 准备images和资源的yaml文件

docker pull registry.cn-hangzhou.aliyuncs.com/yutao517/ingress_nginx_controller:v1.1.0
docker tag registry.cn-hangzhou.aliyuncs.com/yutao517/ingress_nginx_controller:v1.1.0  k8s.gcr.io/ingress-nginx/controller:v1.1.1
​
docker pull registry.cn-hangzhou.aliyuncs.com/yutao517/kube_webhook_certgen:v1.1.1
docker tag registry.cn-hangzhou.aliyuncs.com/yutao517/kube_webhook_certgen:v1.1.1  k8s.gcr.io/ingress-nginx/kube-webhook-certgen:v1.1.1
​
wget https://download.yutao.co/mirror/deploy.yaml
修改其中的328行,修改其image为image: k8s.gcr.io/ingress-nginx/controller:v1.1.1而不是作者源文件中的v1.1.0。
​
[root@master-node ~]# mkdir local_nginx_ingress
[root@master-node ~]# cd local_nginx_ingress/
[root@master-node local_nginx_ingress]# wget https://download.yutao.co/mirror/deploy.yaml
​

17.2 创建world namespace和configmap以及Deployment资源对象

[root@master-node local_nginx_ingress]# cat asia.deployment.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  annotations:
    deployment.kubernetes.io/revision: "1"
  creationTimestamp: "2022-12-12T08:22:51Z"
  generation: 1
  labels:
    app: asia
  name: asia
  namespace: world
  resourceVersion: "1298"
  uid: 34afb7cc-9b30-4243-95b4-b39d0b893090
spec:
  progressDeadlineSeconds: 600
  replicas: 2
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      app: asia
  strategy:
    rollingUpdate:
      maxSurge: 25%
      maxUnavailable: 25%
    type: RollingUpdate
  template:
    metadata:
      creationTimestamp: null
      labels:
        app: asia
    spec:
      containers:
      - image: nginx:1.21.5-alpine
        imagePullPolicy: IfNotPresent
        name: c
        resources: {}
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /usr/share/nginx/html
          name: html
        - mountPath: /etc/nginx
          name: nginx-conf
          readOnly: true
      dnsPolicy: ClusterFirst
      initContainers:
      - command:
        - sh
        - -c
        - echo 'hello, you reached ASIA' > /html/index.html
        image: busybox:1.28
        imagePullPolicy: IfNotPresent
        name: init-container
        resources: {}
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /html
          name: html
      restartPolicy: Always
      schedulerName: default-scheduler
      securityContext: {}
      terminationGracePeriodSeconds: 30
      volumes:
      - emptyDir: {}
        name: html
      - configMap:
          defaultMode: 420
          items:
          - key: nginx.conf
            path: nginx.conf
          name: nginx-conf
        name: nginx-conf
[root@master-node local_nginx_ingress]# 
​
[root@master-node local_nginx_ingress]# cat europe.deployment.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  annotations:
    deployment.kubernetes.io/revision: "1"
  creationTimestamp: "2022-12-12T08:22:51Z"
  generation: 1
  labels:
    app: europe
  name: europe
  namespace: world
  resourceVersion: "1300"
  uid: 131fc90c-7ad9-4d9e-be4c-248486ddfab5
spec:
  progressDeadlineSeconds: 600
  replicas: 2
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      app: europe
  strategy:
    rollingUpdate:
      maxSurge: 25%
      maxUnavailable: 25%
    type: RollingUpdate
  template:
    metadata:
      creationTimestamp: null
      labels:
        app: europe
    spec:
      containers:
      - image: nginx:1.21.5-alpine
        imagePullPolicy: IfNotPresent
        name: c
        resources: {}
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /usr/share/nginx/html
          name: html
        - mountPath: /etc/nginx
          name: nginx-conf
          readOnly: true
      dnsPolicy: ClusterFirst
      initContainers:
      - command:
        - sh
        - -c
        - echo 'hello, you reached EUROPE' > /html/index.html
        image: busybox:1.28
        imagePullPolicy: IfNotPresent
        name: init-container
        resources: {}
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /html
          name: html
      restartPolicy: Always
      schedulerName: default-scheduler
      securityContext: {}
      terminationGracePeriodSeconds: 30
      volumes:
      - emptyDir: {}
        name: html
      - configMap:
          defaultMode: 420
          items:
          - key: nginx.conf
            path: nginx.conf
          name: nginx-conf
        name: nginx-conf
[root@master-node local_nginx_ingress]# 
​
[root@master-node local_nginx_ingress]# kubectl create namespace world
namespace/world created
[root@master-node local_nginx_ingress]# kubectl apply -f nginx-conf-configmap.yaml 
configmap/nginx-conf created
[root@master-node local_nginx_ingress]# kubectl apply -f asia.deployment.yaml 
deployment.apps/asia created
[root@master-node local_nginx_ingress]# kubectl apply -f europe.deployment.yaml 
deployment.apps/europe created
[root@master-node local_nginx_ingress]# 
[root@master-node local_nginx_ingress]# kubectl -n world get all
NAME                          READY   STATUS    RESTARTS   AGE
pod/asia-6644cdddb9-g8k6s     1/1     Running   0          51s
pod/asia-6644cdddb9-r7jpf     1/1     Running   0          51s
pod/europe-866945bcbf-4r44q   1/1     Running   0          10s
pod/europe-866945bcbf-92f25   1/1     Running   0          10s
​
NAME                     READY   UP-TO-DATE   AVAILABLE   AGE
deployment.apps/asia     2/2     2            2           51s
deployment.apps/europe   2/2     2            2           10s
​
NAME                                DESIRED   CURRENT   READY   AGE
replicaset.apps/asia-6644cdddb9     2         2         2       51s
replicaset.apps/europe-866945bcbf   2         2         2       10s
[root@master-node local_nginx_ingress]# 

参考:Kubernetes(十六)—国内安装Ingress-nginx服务

https://blog.yutao.co/blog/2022/05/09/Kubernetes(%E5%8D%81%E5%85%AD)-%E5%9B%BD%E5%86%85%E5%AE%89%E8%A3%85Ingress-nginx%E6%9C%8D%E5%8A%A1.html

18小结

在本章,我们学习了:

Service资源对象是什么,它能帮我们解决什么问题;

Service资源对象和pod的关联关系,endpoints如何解耦Service和pod;

如何创建|使用|验证cluster IP|node port|Loadbalancer|ingress类型的Service资源对象;

学习和掌握Service资源对象的底层实现原理;

掌握Service discovery的实现方式

如何设置pod中container的readiness probe

留言