k8s之资源限制以及探针检查

k8s之资源限制以及探针检查

 

 

一、资源限制

1. 资源限制的使用

当定义Pod时可以选择性地为每个容器设定所需要的资源数量。最常见的可设定资源是CPU和内存大小,以及其他类型的资源。

2. reuqest资源(请求)和limit资源(约束)

当为Pod中的容器指定了request资源时,调度器就使用该信息来决定将Pod调度到哪个节点上。当还为容器指定了limit资源时,kubelet就会确保运行的容器不会使用超出所设的limit资源量。kubelet还会为容器预留所设的request资源量,供该容器使用。
如果Pod所在的节点具有足够的可用资源,容器可以使用超过所设置的request资源量。不过,容器不可以使用超出所设置的limit资源量。
如果给容器设置了内存的limit值,但未设置内存的request值,Kubernetes会自动为其设置与内存limit相匹配的request值。类似的,如果给容器设置了CPU的limit值但未设置CPU的request值,则Kubernetes自动为其设置CPU的request值,并使之与CPU的limit值匹配。

3. 官网示例

https://kubernetes.io/zh/docs/concepts/configuration/manage-resources-containers/

4. Pod和容器的资源请求和限制

  定义创建容器时预分配的CPU资源
  spec.containers[].resources.requests.cpu
  定义创建容器时预分配的内存资源
  spec.containers[].resources.requests.memory
  定义创建容器时预分配的巨页资源
  spec.containers[].resources.requests.hugepages-<size>
  定义cpu的资源上限
  spec.containers[].resources.limits.cpu
  定义内存的资源上限
  spec.containers[].resources.limits.memory
  定义巨页的资源上限
  spec.containers[].resources.limits.hugepages-<size>

5. 资源类型

CPU 和内存都是资源类型。每种资源类型具有其基本单位。 CPU 表达的是计算处理能力,其单位是 Kubernetes CPUs。 内存的单位是字节。 如果你使用的是 Kubernetes v1.14 或更高版本,则可以指定巨页(Huge Page)资源。 巨页是 Linux 特有的功能,节点内核在其中分配的内存块比默认页大小大得多。
例如,在默认页面大小为 4KiB 的系统上,你可以指定约束 hugepages-2Mi: 80Mi。 如果容器尝试分配 40 个 2MiB 大小的巨页(总共 80 MiB ),则分配请求会失败。
说明:
你不能过量使用 hugepages- * 资源。 这与 memory 和 cpu 资源不同。

6. CPU资源单位

CPU资源的request和limit以cpu为单位。kubernetes中的一个cpu相当于1个vCPU(1个超线程)。
Kubernetes也支持带小数CPU的请求。spec.containers[].resources.requests.cpu为0.5的容器能够获得一个cpu的一半CPU资源(类似于cgroup对CPU资源的时间分片)。表达式0.1等价于表达式100m(毫核),表示每1000毫秒内容器可以使用的CPU时间总量为0.1*1000毫秒。

7. 内存资源单位

内存的request和limit以字节为单位。可以用证书表示,
也可以用以10为底数的指数的单位(E、P、T、G、M、K)来表示,
或者以2为底数的指数的单位(Ei、Pi、Ti、Gi、Mi、Ki)来表示。
如1KB=103=1000,1MB=106=1000000=1000KB,1GB=10^9=1000000000=1000MB
1KiB=210=1024,1MiB=220=1048576=1024KiB

PS:在买硬盘的时候,操作系统报的数量要比产品标出或商家号称的小一些,主要原因是标出的是以MB、GB为单位的,1GB就是1,000,000,000Byte,而操作系统是以2进制为处理单位的,因此检查硬盘容量时是以MiB、GiB为单位,1GiB=2^30=1,073,741,824Byte。相比较而言,1GiB要比1GB多出73,741,824Byte,所以检测实际结果要比标出的少一些,单位越大,两者的差值也就越大。

8. 官方文档示例

  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 Pod 有两个 Container。每个 Container 的请求为 0.25 cpu 和 64MiB(226 字节)内存, 每个容器的资源约束为 0.5 cpu 和 128MiB 内存。 你可以认为该 Pod 的资源请求为 0.5 cpu 和 128 MiB 内存,资源限制为 1 cpu 和 256MiB 内存。

9. 资源限制实操

9.1 编写yaml资源配置清单

  [root@master ~]# mkdir /opt/test
  [root@master ~]# cd !$
  cd /opt/test
  [root@master test]# vim test1.yaml
   
  apiVersion: v1
  kind: Pod
  metadata:
  name: test1
  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: "password"
  resources:
  requests:
  memory: "64Mi"
  cpu: "250m"
  limits:
  memory: "128Mi"
  cpu: "500m"

9.2 释放内存(node节点,以node01为例)

由于mysql对于内存的使用要求比较高,因此需要先检查内存的可用空间是否能够满足mysql的正常运行,若剩余内存不够,可对其进行释放操作。

9.2.1 查看内存

free -mH

  [root@node01 ~]# free -mh
  total used free shared buff/cache available
  Mem: 1.9G 1.0G 86M 26M 870M 663M
  Swap: 0B 0B 0B

内存总量为1.9G,实际使用1G,因此可有内存应该为0.9G左右。
但是由于有870M的内存被用于缓存,导致了free仅为86M。
86M剩余可用内存显然是不够用的,因此需要释放缓存。

9.2.2 手动释放缓存

echo [1\2\3] > /proc/sys/vm/drop_caches

  [root@node01 ~]# cat /proc/sys/vm/drop_caches
  0
  [root@node01 ~]# echo 3 > /proc/sys/vm/drop_caches
  [root@node01 ~]# free -mh
  total used free shared buff/cache available
  Mem: 1.9G 968M 770M 26M 245M 754M
  Swap: 0B 0B 0B

0:0是系统默认值,默认情况下表示不释放内存,由操作系统自动管理
1:释放页缓存
2:释放dentries和inodes
3:释放所有缓存
注意:
如果因为是应用有像内存泄露、溢出的问题,从swap的使用情况是可以比较快速可以判断的,但free上面反而比较难查看。相反,如果在这个时候,我们告诉用户,修改系统的一个值,“可以”释放内存,free就大了。用户会怎么想?不会觉得操作系统“有问题”吗?所以说,既然核心是可以快速清空buffer或cache,也不难做到(这从上面的操作中可以明显看到),但核心并没有这样做(默认值是0),我们就不应该随便去改变它。
一般情况下,应用在系统上稳定运行了,free值也会保持在一个稳定值的,虽然看上去可能比较小。当发生内存不足、应用获取不到可用内存、OOM错误等问题时,还是更应该去分析应用方面的原因,如用户量太大导致内存不足、发生应用内存溢出等情况,否则,清空buffer,强制腾出free的大小,可能只是把问题给暂时屏蔽了。

9.3 创建资源

kubectl apply -f tets1.yaml

  [root@master test]# kubectl apply -f test1.yaml
  pod/test1 created

9.4 跟踪查看pod状态

kubectl get pod -o wide -w

  [root@master test]# kubectl get pod -o wide -w
  NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
  test1 0/2 ContainerCreating 0 4s <none> node01 <none> <none>
  test1 2/2 Running 0 18s 10.244.1.55 node01 <none> <none>
  test1 1/2 OOMKilled 0 21s 10.244.1.55 node01 <none> <none>
  test1 2/2 Running 1 37s 10.244.1.55 node01 <none> <none>
  test1 1/2 OOMKilled 1 40s 10.244.1.55 node01 <none> <none>
  ......

OOM(OverOfMemory)表示服务的运行超过了我们所设定的约束值。
Ready:2/2,status:Running说明该pod已成功创建并运行,但运行过程中发生OOM问题被kubelet杀死并重新拉起新的pod。

9.5 查看容器日志

kubectl logs test1 -c web

  [root@master test]# kubectl logs test1 -c web
  /docker-entrypoint.sh: /docker-entrypoint.d/ is not empty, will attempt to perform configuration
  /docker-entrypoint.sh: Looking for shell scripts in /docker-entrypoint.d/
  /docker-entrypoint.sh: Launching /docker-entrypoint.d/10-listen-on-ipv6-by-default.sh
  10-listen-on-ipv6-by-default.sh: info: Getting the checksum of /etc/nginx/conf.d/default.conf
  10-listen-on-ipv6-by-default.sh: info: Enabled listen on IPv6 in /etc/nginx/conf.d/default.conf
  /docker-entrypoint.sh: Launching /docker-entrypoint.d/20-envsubst-on-templates.sh
  /docker-entrypoint.sh: Launching /docker-entrypoint.d/30-tune-worker-processes.sh
  /docker-entrypoint.sh: Configuration complete; ready for start up
  2021/11/06 08:31:23 [notice] 1#1: using the "epoll" event method
  2021/11/06 08:31:23 [notice] 1#1: nginx/1.21.3
  2021/11/06 08:31:23 [notice] 1#1: built by gcc 8.3.0 (Debian 8.3.0-6)
  2021/11/06 08:31:23 [notice] 1#1: OS: Linux 3.10.0-693.el7.x86_64
  2021/11/06 08:31:23 [notice] 1#1: getrlimit(RLIMIT_NOFILE): 1048576:1048576
  2021/11/06 08:31:23 [notice] 1#1: start worker processes
  2021/11/06 08:31:23 [notice] 1#1: start worker process 31
  2021/11/06 08:31:23 [notice] 1#1: start worker process 32

nginx启动正常,接下来查看mysql日志
kubectl logs test1 -c mysql

  [root@master test]# kubectl logs test1 -c db
  2021-11-06 08:38:44+00:00 [Note] [Entrypoint]: Entrypoint script for MySQL Server 8.0.27-1debian10 started.
  2021-11-06 08:38:44+00:00 [Note] [Entrypoint]: Switching to dedicated user 'mysql'
  2021-11-06 08:38:44+00:00 [Note] [Entrypoint]: Entrypoint script for MySQL Server 8.0.27-1debian10 started.
  2021-11-06 08:38:44+00:00 [Note] [Entrypoint]: Initializing database files
  2021-11-06T08:38:44.274783Z 0 [System] [MY-013169] [Server] /usr/sbin/mysqld (mysqld 8.0.27) initializing of server in progress as process 41
  2021-11-06T08:38:44.279965Z 1 [System] [MY-013576] [InnoDB] InnoDB initialization has started.
  2021-11-06T08:38:44.711420Z 1 [System] [MY-013577] [InnoDB] InnoDB initialization has ended.
  2021-11-06T08:38:45.777355Z 0 [Warning] [MY-013746] [Server] A deprecated TLS version TLSv1 is enabled for channel mysql_main
  2021-11-06T08:38:45.777389Z 0 [Warning] [MY-013746] [Server] A deprecated TLS version TLSv1.1 is enabled for channel mysql_main
  2021-11-06T08:38:45.898121Z 6 [Warning] [MY-010453] [Server] root@localhost is created with an empty password ! Please consider switching off the --initialize-insecure option.
  /usr/local/bin/docker-entrypoint.sh: line 191: 41 Killed "$@" --initialize-insecure --default-time-zone=SYSTEM

锁定问题容器为mysql

9.6 删除pod

kubectl delete -f test1

  [root@master test]# kubectl delete -f test1.yaml
  pod "test1" deleted

9.7 修改yaml配置资源清单,提高mysql资源限制

  [root@master test]# vim test1.yaml
   
  apiVersion: v1
  kind: Pod
  metadata:
  name: test1
  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: "password"
  resources:
  requests:
  memory: "512Mi"
  cpu: "0.5"
  limits:
  memory: "1024Mi"
  cpu: "1"

9.8 再次创建资源

kubectl apply -f test1.yaml

  [root@master test]# kubectl apply -f test1.yaml
  pod/test1 created

9.9 跟踪查看pod状态

kubectl get pod -o wide -w

  [root@master test]# kubectl get pod -o wide -w
  NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
  test1 0/2 ContainerCreating 0 12s <none> node01 <none> <none>
  test1 2/2 Running 0 18s 10.244.1.56 node01 <none> <none>

9.10 查看pod详细信息

kubectl describe pod test1

  [root@master test]# kubectl describe pod test1
  ......
  Containers:
  web:
  Container ID: docker://caf5bef54f878ebba32728b5e43743e36bbdf1457973f3ca130c98de5e1803d3
  Image: nginx
  ......
  #nginx资源限制
  Limits:
  cpu: 500m
  memory: 128Mi
  Requests:
  cpu: 250m
  memory: 64Mi
  #nginx环境变量
  Environment:
  WEB_ROOT_PASSWORD: password
  Mounts:
  /var/run/secrets/kubernetes.io/serviceaccount from default-token-7lsdx (ro)
  db:
  Container ID: docker://2574f2bd02d9d7fc5bb0d2b74582b0bece3d8bd37d1d7ff3148ae8109df49367
  Image: mysql
  ......
  #mysql资源限制
  Limits:
  cpu: 1
  memory: 1Gi
  Requests:
  cpu: 500m
  memory: 512Mi
  #mysql环境变量
  Environment:
  MYSQL_ROOT_PASSWORD: password
  Mounts:
  /var/run/secrets/kubernetes.io/serviceaccount from default-token-7lsdx (ro)
  ......
  #pod创建过程/事件记录
  Events:
  Type Reason Age From Message
  ---- ------ ---- ---- -------
  Normal Scheduled 105s default-scheduler Successfully assigned default/test1 to node01
  Normal Pulling 104s kubelet, node01 Pulling image "nginx"
  Normal Pulled 103s kubelet, node01 Successfully pulled image "nginx"
  Normal Created 103s kubelet, node01 Created container web
  Normal Started 103s kubelet, node01 Started container web
  Normal Pulling 103s kubelet, node01 Pulling image "mysql"
  Normal Pulled 88s kubelet, node01 Successfully pulled image "mysql"
  Normal Created 88s kubelet, node01 Created container db
  Normal Started 88s kubelet, node01 Started container db

9.11 查看node资源使用

  [root@master test]# kubectl describe node node01
  ......
  Namespace Name CPU Requests CPU Limits Memory Requests Memory Limits AGE
  --------- ---- ------------ ---------- --------------- ------------- ---
  default test1 750m (37%) 1500m (75%) 576Mi (30%) 1152Mi (61%) 10m
  kube-system coredns-bccdc95cf-qrlbp 100m (5%) 0 (0%) 70Mi (3%) 170Mi (9%) 4d21h
  kube-system kube-flannel-ds-amd64-6927f 100m (5%) 100m (5%) 50Mi (2%) 50Mi (2%) 4d21h
  kube-system kube-proxy-hjqfc 0 (0%) 0 (0%) 0 (0%) 0 (0%) 4d21h
  Allocated resources:
  (Total limits may be over 100 percent, i.e., overcommitted.)
  Resource Requests Limits
  -------- -------- ------
  cpu 950m (47%) 1600m (80%)
  memory 696Mi (36%) 1372Mi (72%)
  ephemeral-storage 0 (0%) 0 (0%)
  Events: <none>

node01的配置为2C2G。
CPU Requests分析:
nginx的requests为250m,mysql的requests为500m,因此node01的CPU Requests为750m,在node01的两个核中使用占比为37%。
CPU Limits分析:
nginx到的limit为500m,mysql的limit为1,因此node01到的CPU Limits为1500m,在node01的两个核中使用占比为75%。
Memory Requests分析:
nginx的requests为64Mi,mysql的requests为512Mi,因此node01的内存Requests为576Mi,在node01的2G内存中使用占比为30%。
Memory Limits分析:
nginx的limits为128Mi,mysql的limit为1Gi,因此node01的1152Mi,在node01的2G内存中使用占比为61%。

二、健康检查

1. 健康检查的定义


健康检查又称为探针(Probe),是由kubelet对容器执行的定期诊断。

2. 探针的三种规则

2.1 livenessProbe存活探针

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。

2.2 readinessProbe就绪探针

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。

2.3 startupProbe启动探针(1.17版本新增)

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。如果匹配了startupProbe探测,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。

2.4 同时定义

以上三种规则可同时定义。在readinessProbe检测成功之前,Pod的running状态是不会变成reasy状态的。

3. Probe支持的三种检测方法

3.1 exec

在容器内执行执行命令,如果容器退出时返回码为0则认为诊断成功。

3.2 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。

3.3 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(2xx和3xx),则诊断被认为是成功的。

4. 探测结果

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

5. 官方文档

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

6. exec方式

6.1 官方示例1

  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 600
  livenessProbe:
  exec:
  command:
  - cat
  - /tmp/healthy
  initialDelaySeconds: 5
  periodSeconds: 5

initalDeploySeconds: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
补充:
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。)

在这个配置文件中,可以看到 Pod 中只有一个容器。 periodSeconds 字段指定了 kubelet 应该每 5 秒执行一次存活探测。 initialDelaySeconds 字段告诉 kubelet 在执行第一次探测前应该等待 5 秒。 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。 如果这个命令返回非 0 值,kubelet 会杀死这个容器并重新启动它。

6.2 编写yaml资源配置清单

  [root@master test]# 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

在这个配置文件中,可以看到Pod只有一个容器。
容器中的command字段表示创建一个/tmp/live文件后休眠30秒,休眠结束后删除该文件,并休眠10分钟。
仅使用livenessProbe存活探针,并使用exec检查方式,对/tmp/live文件进行存活检测。
initialDelaySeconds字段表示kubelet在执行第一次探测前应该等待1秒。
periodSeconds字段表示kubelet每隔3秒执行一次存活探测。

6.3 创建资源

kubectl create -f exec.yaml

  [root@master test]# kubectl create -f exec.yaml
  pod/liveness-exec created

6.4 跟踪查看pod状态

kubectl get pod -o wide -w

  [root@master ~]# kubectl get pod -o wide -w
  liveness-exec 0/1 Pending 0 0s <none> <none> <none> <none>
  liveness-exec 0/1 Pending 0 0s <none> node01 <none> <none>
  liveness-exec 0/1 ContainerCreating 0 0s <none> node01 <none> <none>
  liveness-exec 1/1 Running 0 2s 10.244.1.62 node01 <none> <none>
  liveness-exec 1/1 Running 1 68s 10.244.1.62 node01 <none> <none>

发现68秒时容器重启

6.5 查看pod事件描述

kubectl describe pod liveness-exec

  [root@master test]# kubectl describe pod liveness-exec
   
  ......
  Events:
  Type Reason Age From Message
  ---- ------ ---- ---- -------
  Normal Scheduled 67s default-scheduler Successfully assigned default/liveness-exec to node01
  Normal Started 66s kubelet, node01 Started container liveness-exec-container
  Warning Unhealthy 30s (x3 over 36s) kubelet, node01 Liveness probe failed:
  Normal Killing 30s kubelet, node01 Container liveness-exec-container failed liveness probe, will be restarted
  Normal Pulled 0s (x2 over 67s) kubelet, node01 Container image "busybox" already present on machine
  Normal Created 0s (x2 over 67s) kubelet, node01 Created container liveness-exec-container

在容器启动37秒时,健康检查三次失败(倒推第一次检查在31秒),kubelet启动了killing程序并在67秒时拉取镜像创建新的容器,在68秒时完成第一次容器重启

7. httpGet方式

7.1 官网示例2

  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只有一个容器。initialDealySeconds字段告诉kubelet再执行第一次探测前应该等待3秒。preiodSeconds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,其他返回代码都标示失败。

7.2 编写yaml资源配置清单

  [root@master test]# vim httpget.yaml
   
  apiVersion: v1
  kind: Pod
  metadata:
  name: liveness-httpget
  namespace: default
  spec:
  containers:
  - name: liveness-httpget-container
  image: nginx
  imagePullPolicy: IfNotPresent
  ports:
  - name: nginx
  containerPort: 80
  livenessProbe:
  httpGet:
  port: nginx
  path: /index.html
  initialDelaySeconds: 1
  periodSeconds: 3
  timeoutSeconds: 10

7.3 创建资源

kubectl create -f httpget.yaml

  [root@master test]# kubectl create -f httpget.yaml
  pod/liveness-httpget created

kubectl get pod

  [root@master test]# kubectl get pod
  NAME READY STATUS RESTARTS AGE
  liveness-httpget 1/1 Running 0 6s

7.4 删除Pod的index.html文件

kubectl exec -it liveness-httpget -- rm -rf /usr/share/nginx/html/index.html

[root@master test]# kubectl exec -it liveness-httpget -- rm -rf /usr/share/nginx/html/index.html

7.5 查看pod状态

kubectl get pod -w

  [root@master test]# kubectl get pod -w
  NAME READY STATUS RESTARTS AGE
  liveness-httpget 1/1 Running 0 5m35s
  liveness-httpget 1/1 Running 1 5m37s

容器发生重启

7.6 查看容器事件

kubectl describe pod liveness-httpget

  [root@master ~]# kubectl describe pod liveness-httpget
  ......
  Events:
  Type Reason Age From Message
  ---- ------ ---- ---- -------
  Normal Scheduled 5m47s default-scheduler Successfully assigned default/liveness-httpget to node01
  Normal Pulled 11s (x2 over 5m46s) kubelet, node01 Container image "nginx" already present on machine
  Normal Created 11s (x2 over 5m46s) kubelet, node01 Created container liveness-httpget-container
  Normal Started 11s (x2 over 5m46s) kubelet, node01 Started container liveness-httpget-container
  Warning Unhealthy 11s (x3 over 17s) kubelet, node01 Liveness probe failed: HTTP probe failed with statuscode: 404
  Normal Killing 11s kubelet, node01 Container liveness-httpget-container failed liveness probe, will be restarted

重启原因是HTTP探测得到的状态返回码是404,HTTP probe failed with statuscode: 404。
重启完成后,不会再次重启,因为重新拉取的镜像中包含了index.html文件。

8. tcpSocket方式

8.1 官方示例

  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readiness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。

8.2 编写yaml资源配置清单

  [root@master test]# vim tcpsocket.yaml
   
  apiVersion: v1
  kind: Pod
  metadata:
  name: liveness-tcpsocket
  spec:
  containers:
  - name: liveness-tcpsocket-container
  image: nginx
  livenessProbe:
  initialDelaySeconds: 5
  timeoutSeconds: 1
  tcpSocket:
  port: 8080
  periodSeconds: 3

8.3 创建资源

kubectl apply -f tcpsocket.yaml

  [root@master test]# kubectl apply -f tcpsocket.yaml
  pod/liveness-tcpsocket created

8.4 跟踪查看pod状态

kubectl get pod -w

  [root@master test]# kubectl get pod -w
  NAME READY STATUS RESTARTS AGE
  liveness-tcpsocket 0/1 ContainerCreating 0 6s
  liveness-tcpsocket 1/1 Running 0 17s
  liveness-tcpsocket 1/1 Running 1 44s
  liveness-tcpsocket 1/1 Running 2 71s

pod异常重启

8.5 查看pod事件

kubectl describe pod liveness-tcpsocket

  [root@master test]# kubectl describe pod liveness-tcpsocket
   
  ......
  Events:
  Type Reason Age From Message
  ---- ------ ---- ---- -------
  Normal Scheduled 93s default-scheduler Successfully assigned default/liveness-tcpsocket to node01
  Normal Pulled 23s (x3 over 77s) kubelet, node01 Successfully pulled image "nginx"
  Normal Created 23s (x3 over 77s) kubelet, node01 Created container liveness-tcpsocket-container
  Normal Started 23s (x3 over 77s) kubelet, node01 Started container liveness-tcpsocket-container
  Normal Pulling 11s (x4 over 92s) kubelet, node01 Pulling image "nginx"
  Warning Unhealthy 11s (x9 over 71s) kubelet, node01 Liveness probe failed: dial tcp 10.244.1.65:8080: connect: connection refused
  Normal Killing 11s (x3 over 65s) kubelet, node01 Container liveness-tcpsocket-container failed liveness probe, will be restarted

重启原因是nginx使用的默认端口为80,8080端口的健康检查被拒绝访问

8.5 删除pod

kubectl delete -f tcpsocket.yaml

8.6 修改tcpSocket端口

  [root@master test]# vim tcpsocket.yaml
   
  apiVersion: v1
  kind: Pod
  metadata:
  name: liveness-tcpsocket
  spec:
  containers:
  - name: liveness-tcpsocket-container
  image: nginx
  livenessProbe:
  initialDelaySeconds: 5
  timeoutSeconds: 1
  tcpSocket:
  #修改端口为80
  port: 80
  periodSeconds: 3

8.7 再次创建资源

kubectl apply -f tcpsocket.yaml

  [root@master test]# kubectl apply -f tcpsocket.yaml
  pod/liveness-tcpsocket created

8.8 跟踪查看pod状态

kubectl get pod -o wide -w

  [root@master ~]# kubectl get pod -o wide -w
  NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
  liveness-tcpsocket 1/1 Running 0 21s 10.244.1.66 node01 <none> <none>

启动正常,并未出现重启

8.9 查看pod事件

kubectl describe pod liveness-tcpsocket

  [root@master test]# kubectl describe pod liveness-tcpsocket
   
  ......
  Events:
  Type Reason Age From Message
  ---- ------ ---- ---- -------
  Normal Scheduled 33s default-scheduler Successfully assigned default/liveness-tcpsocket to node01
  Normal Pulling 32s kubelet, node01 Pulling image "nginx"
  Normal Pulled 17s kubelet, node01 Successfully pulled image "nginx"
  Normal Created 17s kubelet, node01 Created container liveness-tcpsocket-container
  Normal Started 17s kubelet, node01 Started container liveness-tcpsocket-container

启动正常

9. readinessProbe就绪探针1

9.1 编写yaml资源配置清单

  [root@master test]# vim readiness-httpget.yaml
   
  apiVersion: v1
  kind: Pod
  metadata:
  name: readiness-httpget
  namespace: default
  spec:
  containers:
  - name: readiness-httpget-container
  image: nginx
  imagePullPolicy: IfNotPresent
  ports:
  - name: http
  containerPort: 80
  readinessProbe:
  httpGet:
  port: 80
  #注意,这里设置个错误地址
  path: /index1.html
  initialDelaySeconds: 1
  periodSeconds: 3
  livenessProbe:
  httpGet:
  port: http
  path: /index.html
  initialDelaySeconds: 1
  periodSeconds: 3
  timeoutSeconds: 10

9.2 创建资源

kubectl apply -f readiness-httpget.yaml

  [root@master test]# kubectl apply -f readiness-httpget.yaml
  pod/readiness-httpget created

9.3 查看pod状态

kubectl get pod

  [root@master test]# kubectl get pod
  NAME READY STATUS RESTARTS AGE
  readiness-httpget 0/1 Running 0 25s

STATUS为Running,但无法进入READY状态

9.4 查看pod事件

kubectl describe pod readiness-httpget

  [root@master test]# kubectl describe pod readiness-httpget
  ......
  Events:
  Type Reason Age From Message
  ---- ------ ---- ---- -------
  Normal Scheduled 119s default-scheduler Successfully assigned default/readiness-httpget to node01
  Normal Pulled 119s kubelet, node01 Container image "nginx" already present on machine
  Normal Created 119s kubelet, node01 Created container readiness-httpget-container
  Normal Started 119s kubelet, node01 Started container readiness-httpget-container
  Warning Unhealthy 54s (x22 over 117s) kubelet, node01 Readiness probe failed: HTTP probe failed with statuscode: 404

异常原因为readinessProbe检测的状态返回值为404,kubelet阻止pod进入READY状态

9.5 查看日志

kubectl logs readiness-httpget

  [root@master test]# kubectl logs readiness-httpget
  ......
  2021/11/07 16:40:41 [error] 32#32: *164 open() "/usr/share/nginx/html/index1.html" failed (2: No such file or directory), client: 10.244.1.1, server: localhost, request: "GET /index1.html HTTP/1.1", host: "10.244.1.68:80"
  10.244.1.1 - - [07/Nov/2021:16:40:41 +0000] "GET /index1.html HTTP/1.1" 404 153 "-" "kube-probe/1.15" "-"
  10.244.1.1 - - [07/Nov/2021:16:40:43 +0000] "GET /index.html HTTP/1.1" 200 615 "-" "kube-probe/1.15" "-"

9.6 为容器创建index1.html

kubectl exec -it readiness-httpget -- touch /usr/share/nginx/html/index1.html

[root@master test]# kubectl exec -it readiness-httpget -- touch /usr/share/nginx/html/index1.html

9.7 查看容器状态

kubectl get pod

  [root@master test]# kubectl get pod
  NAME READY STATUS RESTARTS AGE
  readiness-httpget 1/1 Running 0 15m

10. readinessProbe就绪探针2

10.1 编写yaml资源配置清单

  [root@master test]# cat readiness-multi-nginx.yaml
  apiVersion: v1
  kind: Pod
  metadata:
  name: nginx1
  labels:
  app: nginx
  spec:
  containers:
  - name: nginx
  image: nginx
  imagePullPolicy: IfNotPresent
  ports:
  - name: http
  containerPort: 80
  readinessProbe:
  httpGet:
  port: http
  path: /index.html
  initialDelaySeconds: 5
  periodSeconds: 5
  timeoutSeconds: 10
  ---
  apiVersion: v1
  kind: Pod
  metadata:
  name: nginx2
  labels:
  app: nginx
  spec:
  containers:
  - name: nginx
  image: nginx
  imagePullPolicy: IfNotPresent
  ports:
  - name: http
  containerPort: 80
  readinessProbe:
  httpGet:
  port: http
  path: /index.html
  initialDelaySeconds: 5
  periodSeconds: 5
  timeoutSeconds: 10
  ---
  apiVersion: v1
  kind: Pod
  metadata:
  name: nginx3
  labels:
  app: nginx
  spec:
  containers:
  - name: nginx
  image: nginx
  imagePullPolicy: IfNotPresent
  ports:
  - name: http
  containerPort: 80
  readinessProbe:
  httpGet:
  port: http
  path: /index.html
  initialDelaySeconds: 5
  periodSeconds: 5
  timeoutSeconds: 10
  ---
  apiVersion: v1
  kind: Service
  metadata:
  name: nginx-svc
  spec:
  #service通过selector绑定到nginx集群中
  selector:
  app: nginx
  type: ClusterIP
  ports:
  - name: http
  port: 80
  targetPort: 80

10.2 创建资源

kubectl apply -f readiness-multi-nginx.yaml

  [root@master test]# kubectl apply -f readiness-multi-nginx.yaml
  pod/nginx1 created
  pod/nginx2 created
  pod/nginx3 created
  service/nginx-svc created

10.3 查看pod,service状态

kubectl get pod,svc -o wide

  [root@master test]# kubectl get pod,svc -o wide
  NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
  pod/nginx1 1/1 Running 0 22s 10.244.1.69 node01 <none> <none>
  pod/nginx2 1/1 Running 0 22s 10.244.2.31 node02 <none> <none>
  pod/nginx3 1/1 Running 0 22s 10.244.1.70 node01 <none> <none>
   
  NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE SELECTOR
  service/kubernetes ClusterIP 10.1.0.1 <none> 443/TCP 3d3h <none>
  service/nginx-svc ClusterIP 10.1.177.18 <none> 80/TCP 22s app=nginx

运行成功

10.4 删除nginx1中的index.html

kubectl exec -it nginx1 -- rm -rf /usr/share/nginx/html/index.html

[root@master test]# kubectl exec -it nginx1 -- rm -rf /usr/share/nginx/html/index.html

10.5 查看pod状态

kubectl get pod -o wide -w

  [root@master test]# kubectl get pod -o wide -w
  NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
  nginx1 1/1 Running 0 3m41s 10.244.1.69 node01 <none> <none>
  nginx2 1/1 Running 0 3m41s 10.244.2.31 node02 <none> <none>
  nginx3 1/1 Running 0 3m41s 10.244.1.70 node01 <none> <none>
  nginx1 0/1 Running 0 3m43s 10.244.1.69 node01 <none> <none>

nginx1的READY状态变为0/1

10.6 查看pod事件

kubectl describe pod nginx1

  [root@master test]# kubectl describe pod nginx1
  ......
  Events:
  Type Reason Age From Message
  ---- ------ ---- ---- -------
  Normal Scheduled 4m13s default-scheduler Successfully assigned default/nginx1 to node01
  Normal Pulled 4m12s kubelet, node01 Container image "nginx" already present on machine
  Normal Created 4m12s kubelet, node01 Created container nginx
  Normal Started 4m12s kubelet, node01 Started container nginx
  Warning Unhealthy 0s (x9 over 40s) kubelet, node01 Readiness probe failed: HTTP probe failed with statuscode: 404

由于httpGet检测到的状态返回码为404,所以readinessProbe失败,kubelet将其设定为noready状态。

10.7 查看service详情

kubectl describe svc nginx-svc

  [root@master test]# kubectl describe svc nginx-svc
  Name: nginx-svc
  Namespace: default
  Labels: <none>
  Annotations: kubectl.kubernetes.io/last-applied-configuration:
  {"apiVersion":"v1","kind":"Service","metadata":{"annotations":{},"name":"nginx-svc","namespace":"default"},"spec":{"ports":[{"name":"http"...
  Selector: app=nginx
  Type: ClusterIP
  IP: 10.1.177.18
  Port: http 80/TCP
  TargetPort: 80/TCP
  Endpoints: 10.244.1.70:80,10.244.2.31:80
  Session Affinity: None
  Events: <none>

nginx1被剔除出了service的终端列表

10.8 查看终端

kubectl get endpoints

  [root@master test]# kubectl get endpoints
  NAME ENDPOINTS AGE
  kubernetes 192.168.122.10:6443 3d3h
  nginx-svc 10.244.1.70:80,10.244.2.31:80 9m34s

终端中无nginx1

三、总结

1. 探针

探针分为3种

  1. livenessProbe(存活探针)∶判断容器是否正常运行,如果失败则杀掉容器(不是pod),再根据重启策略是否重启容器
  2. readinessProbe(就绪探针)∶判断容器是否能够进入ready状态,探针失败则进入noready状态,并从service的endpoints中剔除此容器
  3. startupProbe∶判断容器内的应用是否启动成功,在success状态前,其它探针都处于无效状态

2. 检查方式

检查方式分为3种

  1. exec∶使用 command 字段设置命令,在容器中执行此命令,如果命令返回状态码为0,则认为探测成功
  2. httpget∶通过访问指定端口和url路径执行http get访问。如果返回的http状态码为大于等于200且小于400则认为成功
  3. tcpsocket∶通过tcp连接pod(IP)和指定端口,如果端口无误且tcp连接成功,则认为探测成功

3. 常用的探针可选参数

常用的探针可选参数有4个

  1. initialDelaySeconds∶ 容器启动多少秒后开始执行探测
  2. periodSeconds∶探测的周期频率,每多少秒执行一次探测
  3. failureThreshold∶探测失败后,允许再试几次
  4. timeoutSeconds ∶ 探测等待超时的时间
posted @ 2021-12-07 08:06  隐姓埋名4869  阅读(546)  评论(0编辑  收藏  举报