K8SPod advanced resource limits and probes

1. Pod Advanced

1. Resource limitations

When defining a Pod, you can optionally set the number of resources required for each container. The most common configurable resources are CPU and memory size, among other types of resources.

When a request resource is specified for a container in a Pod, the scheduler uses this information to decide which node to schedule the Pod to. When limit resources are also specified for the container, kubelet will ensure that the running container does not use more than the set limit resources. Kubelet will also reserve the set request resource amount for the container for use by the container.

If the node where the Pod is running has enough available resources, the container can use more resources than the set request. However, the container cannot use more resources than the set limit.

If the memory limit value is set for the container, but the memory request value is not set, Kubernetes will automatically set a request value that matches the memory limit. Similarly, if the CPU limit value is set for a container but the CPU request value is not set, Kubernetes automatically sets the CPU request value for it and matches it with the CPU limit value.

Official website example:

https://kubernetes.io/docs/concepts/configuration/manage-compute-resources-container/

2. Resource requests and limitations of Pods and containers:

spec.containers[].resources.requests.cpu        //定义创建容器时预分配的CPU资源
spec.containers[].resources.requests.memory      //定义创建容器时预分配的内存资源
spec.containers[].resources.limits.cpu         //定义 cpu 的资源上限 
spec.containers[].resources.limits.memory       //定义内存的资源上限

3. CPU resource unit

The request and limit of CPU resources are in cpu. One CPU in Kubernetes is equivalent to 1 vCPU (1 hyperthread).
Kubernetes also supports requests with fractional CPUs. A container whose spec.containers[].resources.requests.cpu is 0.5 can obtain half of the CPU resources of a cpu (similar to Cgroup's time slicing of CPU resources). The expression 0.1 is equivalent to the expression 100m (millicore), which means that the total amount of CPU time that the container can use every 1000 milliseconds is 0.1*1000 milliseconds.
Kubernetes does not allow setting CPU resources with a precision less than 1m. 

4. Memory resource unit 

The memory request and limit are in bytes. It can be expressed as an integer, or in units of base 10 exponent (E, P, T, G, M, K), or in units of base 2 exponent (Ei, Pi, Ti, Gi, Mi, Ki) to represent.
For example: 1KB=10^3=1000, 1MB=10^6=1000000=1000KB, 1GB=10^9=1000000000=1000MB 1KiB=2^10=1024, 1MiB=2^20=1048576=
1024KiB

Note: When buying a hard drive, the quantity quoted by the operating system is smaller than what is marked on the product or claimed by the merchant. The main reason is that the quantity marked is in MB and GB. 1GB is 1,000,000,000 Byte, while the operating system is in MB and GB. Binary is the processing unit, so when checking the hard disk capacity, MiB and GiB are used as the unit, 1GiB=2^30=1,073,741,824. In comparison, 1GiB is 1,073,741,824-1,000,000,000=73,741,824Byte more than 1GB, so the actual detection The result is less than indicated.

5. Example operation

5.1 Example 1

apiVersion: v1
kind: Pod
metadata:
  name: frontend
spec:
  containers:
  - name: app
    image: images.my-company.example/app:v4
    env:
    - name: MYSQL_ROOT_PASSWORD
      value: "password"
    resources:
      requests:
        memory: "64Mi"
        cpu: "250m"
      limits:
        memory: "128Mi"
        cpu: "500m"
  - name: log-aggregator
    image: images.my-company.example/log-aggregator:v6
    resources:
      requests:
        memory: "64Mi"
        cpu: "250m"
      limits:
        memory: "128Mi"
        cpu: "500m"

此例子中的 Pod 有两个容器。每个容器的 request 值为 0.25 cpu 和 64MiB 内存，每个容器的 limit 值为 0.5 cpu 和 128MiB 内存。那么可以认为该 Pod 的总的资源 request 为 0.5 cpu 和 128 MiB 内存，总的资源 limit 为 1 cpu 和 256MiB 内存。

5.2 示例2

5.2.1 编写yaml文件

vim pod1.yaml

apiVersion: v1
kind: Pod
metadata:
  name: frontend
spec:
  containers:
  - name: web
    image: nginx
    env:
    - name: WEB_ROOT_PASSWORD
      value: "password"
    resources:
      requests:
        memory: "64Mi"
        cpu: "250m"
      limits:
        memory: "128Mi"
        cpu: "500m"
  - name: db
    image: mysql
    env:
    - name: MYSQL_ROOT_PASSWORD
      value: "abc123"
    resources:
      requests:
        memory: "512Mi"
        cpu: "0.5"
      limits:
        memory: "1Gi"
        cpu: "1"

5.2.2 创建资源

kubectl apply -f pod1.yaml

5.2.3 查看pod1的创建过程

kubectl describe pod frontend

5.2.4 查看资源的具体状态

kubectl get pods -o wide

5.2.5 查看node节点的具体信息

kubectl describe nodes node01	#由于当前虚拟机有2个CPU，所以Pod的CPU Limits一共占用了50%

6、重启策略(restartPolicy)

当 Pod 中的容器退出时通过节点上的 kubelet 重启容器。适用于 Pod 中的所有容器。

• Always：当容器终止退出后，总是重启容器，默认策略
• OnFailure：当容器异常退出(退出状态码非0)时，重启容器；正常退出则不重启容器
• Never：当容器终止退出，从不重启容器。

#注意：K8S 中不支持重启 Pod 资源，只有删除重建

kubectl edit deployment nginx-deployment
......
 restartPolicy: Always

6.1 示例1

6.1.1 创建yaml文件

vim pod2.yaml

apiVersion: v1
kind: Pod
metadata:
  name: foo
spec:
  containers:
  - name: busybox
    image: busybox
    args:
    - /bin/sh
    - -c
    - sleep 30; exit 3

6.1.2 创建资源

kubectl apply -f pod2.yaml

6.1.3 查看pod状态

#查看Pod状态，等容器启动后30秒后执行exit退出进程进入error状态，就会重启次数加1
kubectl get pod

6.1.4 删除原有资源，创建新的资源

kubectl delete -f pod2.yaml

vim pod2.yaml

apiVersion: v1
kind: Pod
metadata:
  name: foo
spec:
  containers:
  - name: busybox
    image: busybox
    args:
    - /bin/sh
    - -c
    - sleep 30; exit 3
  restartPolicy: Never
#注意：跟container同一个级别

kubectl apply -f pod2.yaml

6.1.5 容器进入error状态不会进行重启

#容器进入error状态不会重启
kubectl get pods -w

二、健康检查：又称为探针(Probe) 

探针是由kubelet对容器执行的定期诊断。

1、探针的三种规则

• livenessProbe ：判断容器是否正在运行。如果探测失败，则kubelet会杀死容器，并且容器将根据 restartPolicy 来设置 Pod 状态。 如果容器不提供存活探针，则默认状态为Success。
• readinessProbe ：判断容器是否准备好接受请求。如果探测失败，端点控制器将从与 Pod 匹配的所有 service endpoints 中剔除删除该Pod的IP地址。 初始延迟之前的就绪状态默认为Failure。如果容器不提供就绪探针，则默认状态为Success。
• startupProbe(这个1.17版本增加的)：判断容器内的应用程序是否已启动，主要针对于不能确定具体启动时间的应用。如果配置了 startupProbe 探测，在则在 startupProbe 状态为 Success 之前，其他所有探针都处于无效状态，直到它成功后其他探针才起作用。 如果 startupProbe 失败，kubelet 将杀死容器，容器将根据 restartPolicy 来重启。如果容器没有配置 startupProbe， 则默认状态为 Success。

#注：以上规则可以同时定义。在readinessProbe检测成功之前，Pod的running状态是不会变成ready状态的。

2、Probe支持三种检查方法

• exec ：在容器内执行指定命令。如果命令退出时返回码为0则认为诊断成功。
• tcpSocket ：对指定端口上的容器的IP地址进行TCP检查(三次握手)。如果端口打开，则诊断被认为是成功的。
• httpGet ：对指定的端口和路径上的容器的IP地址执行HTTPGet请求。如果响应的状态码大于等于200且小于400，则诊断被认为是成功的

每次探测都将获得以下三种结果之一：

• 成功：容器通过了诊断。
• 失败：容器未通过诊断。
• 未知：诊断失败，因此不会采取任何行动

3、官网示例

https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-startup-probes/

4、exec方式

exec方式
apiVersion: v1
kind: Pod
metadata:
  labels:
    test: liveness
  name: liveness-exec
spec:
  containers:
  - name: liveness
    image: k8s.gcr.io/busybox
    args:
    - /bin/sh
    - -c
    - touch /tmp/healthy; sleep 30; rm -rf /tmp/healthy; sleep 60
    livenessProbe:
      exec:
        command:
        - cat
        - /tmp/healthy
      failureThreshold: 1
      initialDelaySeconds: 5
      periodSeconds: 5

• initialDelaySeconds：指定 kubelet 在执行第一次探测前应该等待5秒，即第一次探测是在容器启动后的第6秒才开始执行。默认是 0 秒，最小值是 0。
• periodSeconds：指定了 kubelet 应该每 5 秒执行一次存活探测。默认是 10 秒。最小值是 1。
• failureThreshold: 当探测失败时，Kubernetes 将在放弃之前重试的次数。 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。
• timeoutSeconds：探测的超时后等待多少秒。默认值是 1 秒。最小值是 1。(在 Kubernetes 1.20 版本之前，exec 探针会忽略 timeoutSeconds 探针会无限期地 持续运行，甚至可能超过所配置的限期，直到返回结果为止)。

可以看到 Pod 中只有一个容器。kubelet 在执行第一次探测前需要等待 5 秒，kubelet 会每 5 秒执行一次存活探测。kubelet 在容器内执行命令 cat /tmp/healthy 来进行探测。如果命令执行成功并且返回值为 0，kubelet 就会认为这个容器是健康存活的。 当到达第 31 秒时，这个命令返回非 0 值，kubelet 会杀死这个容器并重新启动它。

4.1 示例1：exec方式

vim exec.yaml

apiVersion: v1
kind: Pod
metadata:
  name: liveness-exec
  namespace: default
spec:
  containers:
  - name: liveness-exec-container
    image: busybox
    imagePullPolicy: IfNotPresent
    command: ["/bin/sh","-c","touch /tmp/live ; sleep 30; rm -rf /tmp/live; sleep 3600"]
    livenessProbe:
      exec:
        command: ["test","-e","/tmp/live"]
      initialDelaySeconds: 1
      periodSeconds: 3

4.1.1 创建exec资源

kubectl create -f exec.yaml

4.1.2 查看pod创建的过程

kubectl describe pods liveness-exec

4.1.3 查看pod的状态

kubectl get pods -w

5、httpGet方式

apiVersion: v1
kind: Pod
metadata:
  labels:
    test: liveness
  name: liveness-http
spec:
  containers:
  - name: liveness
    image: k8s.gcr.io/liveness
    args:
    - /server
    livenessProbe:
      httpGet:
        path: /healthz
        port: 8080
        httpHeaders:
        - name: Custom-Header
          value: Awesome
      initialDelaySeconds: 3
      periodSeconds: 3

在这个配置文件中，可以看到 Pod 也只有一个容器。initialDelaySeconds 字段告诉 kubelet 在执行第一次探测前应该等待 3 秒。periodSeconds 字段指定了 kubelet 每隔 3 秒执行一次存活探测。kubelet 会向容器内运行的服务(服务会监听 8080 端口)发送一个 HTTP GET 请求来执行探测。如果服务器上 /healthz 路径下的处理程序返回成功代码，则 kubelet 认为容器是健康存活的。如果处理程序返回失败代码，则 kubelet 会杀死这个容器并且重新启动它。

任何大于或等于 200 并且小于 400 的返回代码标示成功，其它返回代码都标示失败。

5.1 示例2：httpGet方式

vim httpget.yaml

apiVersion: v1
kind: Pod
metadata:
  name: liveness-httpget
  namespace: default
spec:
  containers:
  - name: liveness-httpget-container
    image: soscscs/myapp:v1
    imagePullPolicy: IfNotPresent
    ports:
    - name: http
      containerPort: 80
    livenessProbe:
      httpGet:
        port: http
        path: /index.html
      initialDelaySeconds: 1
      periodSeconds: 3
      timeoutSeconds: 10

5.1.1 创建httpget方式

kubectl create -f httpget.yaml

5.1.2 查看pod状态

6、tcpSocket方式

apiVersion: v1
kind: Pod
metadata:
  name: goproxy
  labels:
    app: goproxy
spec:
  containers:
  - name: goproxy
    image: k8s.gcr.io/goproxy:0.1
    ports:
    - containerPort: 8080
    readinessProbe:
      tcpSocket:
        port: 8080
      initialDelaySeconds: 5
      periodSeconds: 10
    livenessProbe:
      tcpSocket:
        port: 8080
      initialDelaySeconds: 15
      periodSeconds: 20

这个例子同时使用 readinessProbe 和 livenessProbe 探测。kubelet 会在容器启动 5 秒后发送第一个 readinessProbe 探测。这会尝试连接 goproxy 容器的 8080 端口。如果探测成功，kubelet 将继续每隔 10 秒运行一次检测。除了 readinessProbe 探测，这个配置包括了一个 livenessProbe 探测。kubelet 会在容器启动 15 秒后进行第一次 livenessProbe 探测。就像 readinessProbe 探测一样，会尝试连接 goproxy 容器的 8080 端口。如果 livenessProbe 探测失败，这个容器会被重新启动。

6.1 示例3：tcpSocket方式

vim tcpsocket.yaml

apiVersion: v1
kind: Pod
metadata:
  name: probe-tcp
spec:
  containers:
  - name: nginx
    image: soscscs/myapp:v1
    livenessProbe:
      initialDelaySeconds: 5
      timeoutSeconds: 1
      tcpSocket:
        port: 8080
      periodSeconds: 10
      failureThreshold: 2

6.1.1 创建tcpsocket资源

kubectl create -f tcpsocket.yaml

6.1.2 查看状态

kubectl exec -it probe-tcp  -- netstat -natp

6.1.3 查看pod状态

kubectl get pods -w

NAME        READY   STATUS    RESTARTS   AGE
probe-tcp   1/1     Running    0          1s
probe-tcp   1/1     Running    1          25s       #第一次是 init(5秒) + period(10秒) * 2
probe-tcp   1/1     Running    2          45s       #第二次是 period(10秒) + period(10秒)  重试了两次
probe-tcp   1/1     Running    3          65s

7、就绪检测

vim readiness-httpget.yaml

apiVersion: v1
kind: Pod
metadata:
  name: readiness-httpget
  namespace: default
spec:
  containers:
  - name: readiness-httpget-container
    image: soscscs/myapp:v1
    imagePullPolicy: IfNotPresent
    ports:
    - name: http
      containerPort: 80
    readinessProbe:
      httpGet:
        port: 80
        path: /index1.html
      initialDelaySeconds: 1
      periodSeconds: 3
    livenessProbe:
      httpGet:
        port: http
        path: /index.html
      initialDelaySeconds: 1
      periodSeconds: 3
      timeoutSeconds: 10

7.1 创建readiness-httpget资源

kubectl create -f readiness-httpget.yaml

7.2 查看pod状态

7.3 进入资源里

kubectl exec -it readiness-httpget sh
 # cd /usr/share/nginx/html/
 # ls
50x.html    index.html
 # echo 123 > index.html 
 # exit

7.4 查看pod状态

kubectl get pods 
NAME                READY   STATUS    RESTARTS   AGE
readiness-httpget   1/1     Running   0          2m31s

8、就绪检测2

vim readiness-myapp.yaml

apiVersion: v1
kind: Pod
metadata:
  name: myapp1
  labels:
     app: myapp
spec:
  containers:
  - name: myapp
    image: soscscs/myapp:v1
    ports:
    - name: http
      containerPort: 80
    readinessProbe:
      httpGet:
        port: 80
        path: /index.html
      initialDelaySeconds: 5
      periodSeconds: 5
      timeoutSeconds: 10 
---
apiVersion: v1
kind: Pod
metadata:
  name: myapp2
  labels:
     app: myapp
spec:
  containers:
  - name: myapp
    image: soscscs/myapp:v1
    ports:
    - name: http
      containerPort: 80
    readinessProbe:
      httpGet:
        port: 80
        path: /index.html
      initialDelaySeconds: 5
      periodSeconds: 5
      timeoutSeconds: 10 
---
apiVersion: v1
kind: Pod
metadata:
  name: myapp3
  labels:
     app: myapp
spec:
  containers:
  - name: myapp
    image: soscscs/myapp:v1
    ports:
    - name: http
      containerPort: 80
    readinessProbe:
      httpGet:
        port: 80
        path: /index.html
      initialDelaySeconds: 5
      periodSeconds: 5
      timeoutSeconds: 10 
---
apiVersion: v1
kind: Service
metadata:
  name: myapp
spec:
  selector:
    app: myapp
  type: ClusterIP
  ports:
  - name: http
    port: 80
    targetPort: 80

8.1 创建readiness-myapp资源

kubectl create -f readiness-myapp.yaml

8.2 查看资源的具体状态

kubectl get pods,svc,endpoints -o wide

8.4 readiness探测失败，Pod 无法进入READY状态，且端点控制器将从 endpoints 中剔除删除该 Pod 的 IP 地址

kubectl exec -it pod/myapp1 -- rm -rf /usr/share/nginx/html/index.html

9、启动、退出动作

vim post.yaml

apiVersion: v1
kind: Pod
metadata:
  name: lifecycle-demo
spec:
  containers:
  - name: lifecycle-demo-container
    image: soscscs/myapp:v1
    lifecycle:   #此为关键字段
      postStart:
        exec:
          command: ["/bin/sh", "-c", "echo Hello from the postStart handler >> /var/log/nginx/message"]      
      preStop:
        exec:
          command: ["/bin/sh", "-c", "echo Hello from the poststop handler >> /var/log/nginx/message"]
    volumeMounts:
    - name: message-log
      mountPath: /var/log/nginx/
      readOnly: false
  initContainers:
  - name: init-myservice
    image: soscscs/myapp:v1
    command: ["/bin/sh", "-c", "echo 'Hello initContainers'   >> /var/log/nginx/message"]
    volumeMounts:
    - name: message-log
      mountPath: /var/log/nginx/
      readOnly: false
  volumes:
  - name: message-log
    hostPath:
      path: /data/volumes/nginx/log/
      type: DirectoryOrCreate

9.1 创建资源

kubectl create -f post.yaml

9.2 查看pod具体状态

kubectl get pods -o wide

9.3 查看/var/log/nginx/nessage的信息

kubectl exec -it lifecycle-demo -- cat /var/log/nginx/message

9.4 在node02节点上查看

cd /data/volumes/nginx/log/
ls
cat message

#由上可知，init Container先执行，然后当一个主容器启动后，Kubernetes 将立即发送 postStart 事件。

9.5 在master节点上删除pod，再到node02节点上查看

kubectl delete pod lifecycle-demo

cat message

#由上可知，当在容器被终结之前， Kubernetes 将发送一个 preStop 事件。