关于 Kubernetes中Pod健康检测和服务可用性检查的一些笔记(LivenessProbe+ReadinessProbe)

写在前面


  • 学习k8s这里整理记忆
  • 博文内容涉及:
  • LivenessProbe,ReadinessProbe两种探针的一些基本理论
  • ExecAction,TCPSocketAction,HTTPGetAction三种检测方式的Demo

中秋明月,豪门有,贫家也有。极慰人心。 ——烽火戏诸侯《剑来》
*

Pod健康检查和服务可用性检查

健康检查的目的

探测的目的: 用来维持 pod的健壮性,当pod挂掉之后,deployment会生成新的pod,但如果pod是正常运行的,但pod里面出了问题,此时deployment是监测不到的。故此需要探测(probe)-pod是不是正常提供服务的

探针类似

KubernetesPod 的健康状态可以通过两类探针来检查:LivenessProbe ReadinessProbe, kubelet定期执行这两类探针来诊断容器的健康状况。都是通过deployment实现的

探针类型 描述
LivenessProbe探针 用于判断容器是否存活(Running状态) ,如果LivenessProbe探针探测到容器不健康,则kubelet将杀掉该容器,并根据容器的重启策略做相应的处理。如果一个容器不包含LivenesspProbe探针,那么kubelet认为该容器的LivenessProbe探针返回的值永远是Success。
ReadinessProbe探针 用于判断容器服务是否可用(Ready状态) ,达到Ready状态的Pod才可以接收请求。对于被Service管理的Pod, Service与Pod Endpoint的关联关系也将基于Pod是否Ready进行设置。如果在运行过程中Ready状态变为False,则系统自动将其从Service的后端Endpoint列表中隔离出去,后续再把恢复到Ready状态的Pod加回后端Endpoint列表。这样就能保证客户端在访问Service时不会被转发到服务不可用的Pod实例上。

检测方式及参数配置

LivenessProbeReadinessProbe均可配置以下三种实现方式。

方式 描述
ExecAction 在容器内部执行一个命令,如果该命令的返回码为0,则表明容器健康。
TCPSocketAction 通过容器的IP地址和端口号执行TC检查,如果能够建立TCP连接,则表明容器健康。
HTTPGetAction 通过容器的IP地址、端口号及路径调用HTTP Get方法,如果响应的状态码大于等于200且小于400,则认为容器健康。

对于每种探测方式,需要设置initialDelaySecondstimeoutSeconds等参数,它们的含义分别如下。

参数 描述
initialDelaySeconds: 启动容器后进行首次健康检查的等待时间,单位为s。
timeoutSeconds: 健康检查发送请求后等待响应的超时时间,单位为s。当超时发生时, kubelet会认为容器已经无法提供服务,将会重启该容器。
periodSeconds 执行探测的频率,默认是10秒,最小1秒。
successThreshold 探测失败后,最少连续探测成功多少次才被认定为成功,默认是1,对于liveness必须是1,最小值是1。
failureThreshold 当 Pod 启动了并且探测到失败,Kubernetes 的重试次数。存活探测情况下的放弃就意味着重新启动容器。就绪探测情况下的放弃 Pod 会被打上未就绪的标签。默认值是 3。最小值是 1

Kubernetes的ReadinessProbe机制可能无法满足某些复杂应用对容器内服务可用状态的判断

所以Kubernetes从1.11版本开始,引入PodReady++特性对Readiness探测机制进行扩展,在1.14版本时达到GA稳定版,称其为Pod Readiness Gates。

通过Pod Readiness Gates机制,用户可以将自定义的ReadinessProbe探测方式设置在Pod上,辅助Kubernetes设置Pod何时达到服务可用状态(Ready) 。为了使自定义的ReadinessProbe生效,用户需要提供一个外部的控制器(Controller)来设置相应的Condition状态。

Pod的Readiness Gates在Pod定义中的ReadinessGate字段进行设置。下面的例子设置了一个类型为www.example.com/feature-1的新ReadinessGate:

关于 Kubernetes中Pod健康检测和服务可用性检查的一些笔记(LivenessProbe+ReadinessProbe)
新增的自定义Condition的状态(status)将由用户自定义的外部控·制器设置,默认值为False. Kubernetes将在判断全部readinessGates条件都为True时,才设置Pod为服务可用状态(Ready为True) 。
这个不是太懂,需要以后再研究下

学习环境准备

┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$mkdir liveness-probe
┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$cd liveness-probe/
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  create ns liveness-probe
namespace/liveness-probe created
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl config current-context
kubernetes-admin@kubernetes
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl config set-context $(kubectl config current-context) --namespace=liveness-probe
Context "kubernetes-admin@kubernetes" modified.

LivenessProbe探针

用于判断容器是否存活(Running状态) ,如果LivenessProbe探针探测到容器不健康,则kubelet将杀掉该容器,并根据容器的重启策略做相应的处理

ExecAction方式:command

在容器内部执行一个命令,如果该命令的返回码为0,则表明容器健康。

资源文件定义

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$cat liveness-probe.yaml
apiVersion: v1
kind: Pod
metadata:
  creationTimestamp: null
  labels:
    run: pod-liveness
  name: pod-liveness
spec:
  containers:
  - args:
    - /bin/sh
    - -c
    - touch /tmp/healthy; sleep 30; rm -rf /tmp/healthy; slee 10
    livenessProbe:
      exec:
        command:
        - cat
        - /tmp/healthy
      initialDelaySeconds: 5 #容器启动的5s内不监测
      periodSeconds: 5 #每5s钟检测一次
    image: busybox
    imagePullPolicy: IfNotPresent
    name: pod-liveness
    resources: {}
  dnsPolicy: ClusterFirst
  restartPolicy: Always 
status: {}

运行这个deploy。当pod创建成功后,新建文件,并睡眠30s,删掉文件在睡眠。使用liveness检测文件的存在

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  apply  -f liveness-probe.yaml
pod/pod-liveness created
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  get pods
NAME           READY   STATUS    RESTARTS     AGE
pod-liveness   1/1     Running   1 (8s ago)   41s   # 30文件没有重启

运行超过30s后。文件被删除,所以被健康检测命中,pod根据重启策略重启

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  get pods
NAME           READY   STATUS    RESTARTS      AGE
pod-liveness   1/1     Running   2 (34s ago)   99s

99s后已经从起了第二次

┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$ansible 192.168.26.83 -m shell -a "docker ps | grep pod-liveness"
192.168.26.83 | CHANGED | rc=0 >>
00f4182c014e   7138284460ff                                        "/bin/sh -c 'touch /…"   6 seconds ago   Up 5 seconds             k8s_pod-liveness_pod-liveness_liveness-probe_81b4b086-fb28-4657-93d0-bd23e67f980a_0
01c5cfa02d8c   registry.aliyuncs.com/google_containers/pause:3.5   "/pause"                 7 seconds ago   Up 6 seconds             k8s_POD_pod-liveness_liveness-probe_81b4b086-fb28-4657-93d0-bd23e67f980a_0
┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$kubectl get pods
NAME           READY   STATUS    RESTARTS   AGE
pod-liveness   1/1     Running   0          25s
┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$kubectl get pods
NAME           READY   STATUS    RESTARTS      AGE
pod-liveness   1/1     Running   1 (12s ago)   44s
┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$ansible 192.168.26.83 -m shell -a "docker ps | grep pod-liveness"
192.168.26.83 | CHANGED | rc=0 >>
1eafd7e8a12a   7138284460ff                                        "/bin/sh -c 'touch /…"   15 seconds ago   Up 14 seconds             k8s_pod-liveness_pod-liveness_liveness-probe_81b4b086-fb28-4657-93d0-bd23e67f980a_1
01c5cfa02d8c   registry.aliyuncs.com/google_containers/pause:3.5   "/pause"                 47 seconds ago   Up 47 seconds             k8s_POD_pod-liveness_liveness-probe_81b4b086-fb28-4657-93d0-bd23e67f980a_0
┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$

查看节点机docker中的容器ID,前后不一样,确定是POD被杀掉后重启。

HTTPGetAction的方式

通过容器的IP地址、端口号及路径调用HTTP Get方法,如果响应的状态码大于等于200且小于400,则认为容器健康。
创建资源文件,即相关参数使用

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$cat liveness-probe-http.yaml
apiVersion: v1
kind: Pod
metadata:
  creationTimestamp: null
  labels:
    run: pod-livenss-probe
  name: pod-livenss-probe
spec:
  containers:
  - image: nginx
    imagePullPolicy: IfNotPresent
    name: pod-livenss-probe
    livenessProbe:
      failureThreshold: 3 #当 Pod 启动了并且探测到失败,Kubernetes 的重试次数
      httpGet:
        path: /index.html
        port: 80
        scheme: HTTP
      initialDelaySeconds: 10  #容器启动后第一次执行探测是需要等待多少秒
      periodSeconds: 10   #执行探测的频率,默认是10秒,最小1秒
      successThreshold: 1 #探测失败后,最少连续探测成功多少次才被认定为成功
      timeoutSeconds: 10 #探测超时时间,默认1秒,最小1秒
    resources: {}
  dnsPolicy: ClusterFirst
  restartPolicy: Always
status: {}

运行deploy,这个的探测机制访问Ngixn的默认欢迎页

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$vim liveness-probe-http.yaml
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl apply  -f liveness-probe-http.yaml
pod/pod-livenss-probe created
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  get pods
NAME                READY   STATUS    RESTARTS   AGE
pod-livenss-probe   1/1     Running   0          15s
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl exec -it pod-livenss-probe -- rm /usr/share/nginx/html/index.html
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  get pods
NAME                READY   STATUS    RESTARTS     AGE
pod-livenss-probe   1/1     Running   1 (1s ago)   2m31s

当欢迎页被删除时,访问报错,被检测命中,pod重启

TCPSocketAction方式

通过容器的IP地址和端口号执行TCP检查,如果能够建立TCP连接,则表明容器健康。
资源文件定义

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$cat liveness-probe-tcp.yaml
apiVersion: v1
kind: Pod
metadata:
  creationTimestamp: null
  labels:
    run: pod-livenss-probe
  name: pod-livenss-probe
spec:
  containers:
  - image: nginx
    imagePullPolicy: IfNotPresent
    name: pod-livenss-probe
    livenessProbe:
      failureThreshold: 3
      tcpSocket:
        port: 8080
      initialDelaySeconds: 10
      periodSeconds: 10
      successThreshold: 1
      timeoutSeconds: 10
    resources: {}
  dnsPolicy: ClusterFirst
  restartPolicy: Always
status: {}

访问8080端口,但是8080端口未开放,所以访问会超时,不能建立连接,命中检测,重启Pod

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  apply  -f liveness-probe-tcp.yaml
pod/pod-livenss-probe created
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  get pods
NAME                READY   STATUS    RESTARTS   AGE
pod-livenss-probe   1/1     Running   0          8s
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  get pods
NAME                READY   STATUS    RESTARTS     AGE
pod-livenss-probe   1/1     Running   1 (4s ago)   44s
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$

ReadinessProbe探针

用于判断容器服务是否可用(Ready状态) ,达到Ready状态的Pod才可以接收请求。负责不能进行访问

ExecAction方式:command

资源文件定义,使用钩子建好需要检查的文件

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$cat readiness-probe.yaml
apiVersion: v1
kind: Pod
metadata:
  creationTimestamp: null
  labels:
    run: pod-liveness
  name: pod-liveness
spec:
  containers:
  - readinessProbe:
      exec:
        command:
        - cat
        - /tmp/healthy
      initialDelaySeconds: 5 #容器启动的5s内不监测
      periodSeconds: 5 #每5s钟检测一次
    image: nginx
    imagePullPolicy: IfNotPresent
    name: pod-liveness
    resources: {}
    lifecycle:
      postStart:
        exec:
          command: ["/bin/sh", "-c","touch /tmp/healthy"]
  dnsPolicy: ClusterFirst
  restartPolicy: Always
status: {}

创建3个有Ngixn的pod,通过POD创建一个SVC做测试用

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$sed 's/pod-liveness/pod-liveness-1/' readiness-probe.yaml | kubectl apply  -f -
pod/pod-liveness-1 created
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$sed 's/pod-liveness/pod-liveness-2/' readiness-probe.yaml | kubectl apply  -f -
pod/pod-liveness-2 created
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  get pods -o wide
NAME             READY   STATUS    RESTARTS   AGE    IP             NODE                         NOMINATED NODE   READINESS GATES
pod-liveness     1/1     Running   0          3m1s   10.244.70.50   vms83.liruilongs.github.io   <none>           <none>
pod-liveness-1   1/1     Running   0          2m     10.244.70.51   vms83.liruilongs.github.io   <none>           <none>
pod-liveness-2   1/1     Running   0          111s   10.244.70.52   vms83.liruilongs.github.io   <none>           <none>

修改主页文字

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$serve=pod-liveness
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl exec  -it $serve -- sh -c "echo $serve > /usr/share/nginx/html/index.html"
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl exec  -it $serve -- sh -c "cat /usr/share/nginx/html/index.html"
pod-liveness
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$serve=pod-liveness-1
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl exec  -it $serve -- sh -c "echo $serve > /usr/share/nginx/html/index.html"
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$serve=pod-liveness-2
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl exec  -it $serve -- sh -c "echo $serve > /usr/share/nginx/html/index.html"
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$

修改标签

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl get pods --show-labels
NAME             READY   STATUS    RESTARTS   AGE   LABELS
pod-liveness     1/1     Running   0          15m   run=pod-liveness
pod-liveness-1   1/1     Running   0          14m   run=pod-liveness-1
pod-liveness-2   1/1     Running   0          14m   run=pod-liveness-2
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl edit pods pod-liveness-1
pod/pod-liveness-1 edited
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl edit pods pod-liveness-2
pod/pod-liveness-2 edited
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl get pods --show-labels
NAME             READY   STATUS    RESTARTS   AGE   LABELS
pod-liveness     1/1     Running   0          17m   run=pod-liveness
pod-liveness-1   1/1     Running   0          16m   run=pod-liveness
pod-liveness-2   1/1     Running   0          16m   run=pod-liveness
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$

要删除文件检测

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  exec -it pod-liveness -- ls /tmp/
healthy
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  exec -it pod-liveness-1 -- ls /tmp/
healthy

使用POD创建SVC

┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl expose  --name=svc pod pod-liveness --port=80
service/svc exposed
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl  get ep
NAME   ENDPOINTS                                         AGE
svc    10.244.70.50:80,10.244.70.51:80,10.244.70.52:80   16s
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl get svc
NAME   TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)   AGE
svc    ClusterIP   10.104.246.121   <none>        80/TCP    36s
┌──[root@vms81.liruilongs.github.io]-[~/ansible/liveness-probe]
└─$kubectl get pods -o wide
NAME             READY   STATUS    RESTARTS   AGE   IP             NODE                         NOMINATED NODE   READINESS GATES
pod-liveness     1/1     Running   0          24m   10.244.70.50   vms83.liruilongs.github.io   <none>           <none>
pod-liveness-1   1/1     Running   0          23m   10.244.70.51   vms83.liruilongs.github.io   <none>           <none>
pod-liveness-2   1/1     Running   0          23m   10.244.70.52   vms83.liruilongs.github.io   <none>           <none>

测试SVC正常,三个POD会正常 负载

┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$while true; do curl 10.104.246.121 ; sleep 1
> done
pod-liveness
pod-liveness-2
pod-liveness
pod-liveness-1
pod-liveness-2
^C

删除文件测试

┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$kubectl exec -it pod-liveness -- rm  -rf /tmp/
┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$kubectl exec -it pod-liveness -- ls /tmp/
ls: cannot access '/tmp/': No such file or directory
command terminated with exit code 2
┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$while true; do curl 10.104.246.121 ; sleep 1; done
pod-liveness-2
pod-liveness-2
pod-liveness-2
pod-liveness-1
pod-liveness-2
pod-liveness-2
pod-liveness-1
^C

会发现pod-liveness的pod已经不提供服务了

kubeadm 中的一些健康检测

kube-apiserver.yaml中的使用,两种探针同时使用

┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$cat /etc/kubernetes/manifests/kube-apiserver.yaml | grep -A 8 readi
    readinessProbe:
      failureThreshold: 3
      httpGet:
        host: 192.168.26.81
        path: /readyz
        port: 6443
        scheme: HTTPS
      periodSeconds: 1
      timeoutSeconds: 15
┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$cat /etc/kubernetes/manifests/kube-apiserver.yaml | grep -A 9 liveness
    livenessProbe:
      failureThreshold: 8
      httpGet:
        host: 192.168.26.81
        path: /livez
        port: 6443
        scheme: HTTPS
      initialDelaySeconds: 10
      periodSeconds: 10
      timeoutSeconds: 15
┌──[root@vms81.liruilongs.github.io]-[~/ansible]
└─$
上一篇:MaxCompute中使用OSS外部表读取JSON数据


下一篇:关于 Kubernetes中一些基本概念和术语笔记