VitePress

Appearance

Return to top
On this page

十九 容器资源限制

起始值 requests 最低保障

终结值 limits 硬限制

  • CPU
1 颗 CPU = 1000 millicores
0.5 颗 CPU = 500 m

1 颗 CPU = 1000 millicores
0.5 颗 CPU = 500 m
  • 内存
Ei、Pi、Ti、Gi、Mi、Ki

Ei、Pi、Ti、Gi、Mi、Ki

19.1 资源限制

  • 清单格式，详见：kubectl explain pods.spec.containers.resources
resources      <Object>               # 资源限制
  limits       <map[string]string>    # 资源最高限制
    cpu        <string>               # 单位 m
    memory     <string>               # 单位 Gi、Mi
  requests     <map[string]string>    # 资源最低要求
    cpu        <string>               # 单位 m
    memory     <string>               # 单位 Gi、Mi

resources      <Object>               # 资源限制
  limits       <map[string]string>    # 资源最高限制
    cpu        <string>               # 单位 m
    memory     <string>               # 单位 Gi、Mi
  requests     <map[string]string>    # 资源最低要求
    cpu        <string>               # 单位 m
    memory     <string>               # 单位 Gi、Mi
  • 清单示例，node 节点的 CPU 为 12 核心，cpu limits 设置为 1000m 也就是允许
apiVersion: v1
kind: Pod
metadata:
  name: pod-resources-demo
  namespace: default
  labels:
    app: myapp
    tier: frontend
spec:
  containers:
  - name: nginx
    image: ikubernetes/stress-ng
    command:
    - "/usr/bin/stress-ng"
    #- "-m 1"                       # 以单线程压测内存
    - "-c 1"                        # 以单线程压测CPU
    - "--metrics-brief"
    ports:
    - name: http
      containerPort: 80
    - name: https
      containerPort: 443
    resources:
      requests:
        cpu: 1000m                 # 它决定在预选阶段淘汰哪些主机
        memory: 512Mi
      limits:
        cpu: 1000m                 # 表示限制容器使用 node 节点的一颗 CPU，无论多少进程，它们最多只能占用 node 节点的可 CPU
        memory: 512Mi

apiVersion: v1
kind: Pod
metadata:
  name: pod-resources-demo
  namespace: default
  labels:
    app: myapp
    tier: frontend
spec:
  containers:
  - name: nginx
    image: ikubernetes/stress-ng
    command:
    - "/usr/bin/stress-ng"
    #- "-m 1"                       # 以单线程压测内存
    - "-c 1"                        # 以单线程压测CPU
    - "--metrics-brief"
    ports:
    - name: http
      containerPort: 80
    - name: https
      containerPort: 443
    resources:
      requests:
        cpu: 1000m                 # 它决定在预选阶段淘汰哪些主机
        memory: 512Mi
      limits:
        cpu: 1000m                 # 表示限制容器使用 node 节点的一颗 CPU，无论多少进程，它们最多只能占用 node 节点的可 CPU
        memory: 512Mi
  • 查看结果
Mem: 855392K used, 139916K free, 10188K shrd, 796K buff, 350368K cached
CPU0:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU1: 100% usr   0% sys   0% nic   0% idle   0% io   0% irq   0% sirq         # 占满了一颗 CPU
CPU2:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU3:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU4:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU5:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU6:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU7:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU8:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU9:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU10:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU11:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
Load average: 0.84 0.50 0.40 3/485 11
  PID  PPID USER     STAT   VSZ %VSZ CPU %CPU COMMAND
    6     1 root     R     6888   1%   1   8% {stress-ng-cpu} /usr/bin/stress-ng -c 1 --metrics-brief
    1     0 root     S     6244   1%  10   0% /usr/bin/stress-ng -c 1 --metrics-brief
    7     0 root     R     1504   0%  11   0% top

Mem: 855392K used, 139916K free, 10188K shrd, 796K buff, 350368K cached
CPU0:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU1: 100% usr   0% sys   0% nic   0% idle   0% io   0% irq   0% sirq         # 占满了一颗 CPU
CPU2:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU3:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU4:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU5:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU6:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU7:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU8:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU9:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU10:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
CPU11:   0% usr   0% sys   0% nic  99% idle   0% io   0% irq   0% sirq
Load average: 0.84 0.50 0.40 3/485 11
  PID  PPID USER     STAT   VSZ %VSZ CPU %CPU COMMAND
    6     1 root     R     6888   1%   1   8% {stress-ng-cpu} /usr/bin/stress-ng -c 1 --metrics-brief
    1     0 root     S     6244   1%  10   0% /usr/bin/stress-ng -c 1 --metrics-brief
    7     0 root     R     1504   0%  11   0% top

19.2 qos 质量管理

  • GuranteedW
每个容器同时设置了 CPU 和内存的 requests 和 limits，而且
    cpu.limits = cpu.requests
    memory.limits = memory.requests
那么它将优先被调度

每个容器同时设置了 CPU 和内存的 requests 和 limits，而且
    cpu.limits = cpu.requests
    memory.limits = memory.requests
那么它将优先被调度
  • Burstable
至少有一个容器设置 CPU 或内存资源的 requests 属性

那么它将具有中等优先级

至少有一个容器设置 CPU 或内存资源的 requests 属性

那么它将具有中等优先级
  • BestEffort
没有任何一个容器设置了 requests 或 limits 属性

那么它将只有最低优先级，当资源不够用的时候，这个容器可能最先被终止，以腾出资源来，为 Burstable 和 Guranteed

没有任何一个容器设置了 requests 或 limits 属性

那么它将只有最低优先级，当资源不够用的时候，这个容器可能最先被终止，以腾出资源来，为 Burstable 和 Guranteed
  • oom 策略
最先杀死占用量和需求量的比例大的

最先杀死占用量和需求量的比例大的