一、开启cpu Manager
# more /lib/systemd/system/kubelet.service --cpu-manager-policy=static --kube-reserved=cpu=200m,memory=1Gi,ephemeral-storage=1Gi --system-reserved=cpu=200m,memory=1G
二、如何使用
Pod 必须是 Guaranteed 的 QoS 类型
Pod中该Container的Cpu request必须为整数CPUs
三、实例
# more deploy1.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx-deployment
spec:
selector:
matchLabels:
app: nginx
replicas: 4
template:
metadata:
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx:1.7.9
ports:
- containerPort: 80
resources:
limits:
memory: "200Mi"
cpu: "2"
requests:
memory: "200Mi"
cpu: "2"
效果查看
# docker ps|grep nginx 7bead158cbf5 84581e99d807 "nginx -g 'daemon of…" 27 seconds ago Up 27 seconds k8s_nginx_nginx-deployment-c585c6cf8-22x5h_default_3babe9d2-e085-11eb-9b5a-fa163e1b0c2f_0 ab63ab8d72b3 mirrorgooglecontainers/pause-amd64:3.0 "/pause" 28 seconds ago Up 27 seconds k8s_POD_nginx-deployment-c585c6cf8-22x5h_default_3babe9d2-e085-11eb-9b5a-fa163e1b0c2f_0 # cd /sys/fs/cgroup/cpuset/kubepods/pod3babe9d2-e085-11eb-9b5a-fa163e1b0c2f/7bead158cbf532e325fc3c57afe77fda026b32b5eb0dc9ced509e2e3834df35f/ # more cpuset.cpus 1-2 //绑定在cpu1-2 # more cpuset.mems 0
四、报错处理
报错：configured policy "static" differs from state checkpoint policy "none"
# kubectl get node
NAME STATUS ROLES AGE VERSION
10.30.20.106 Ready <none> 8d v1.14.4
10.30.20.119 Ready <none> 47h v1.14.4
10.30.20.129 Ready <none> 8d v1.14.4
10.30.20.94 Ready <none> 8d v1.14.4
# kubectl drain 10.30.20.94 --ignore-daemonsets
# mv /var/lib/kubelet/cpu_manager_state /var/lib/kubelet/cpu_manager_state-old
# kubectl uncordon minikube
# kubectl uncordon 10.30.20.94
node/10.30.20.94 uncordoned
# systemctl daemon-reload
# systemctl restart kubelet
# systemctl status kubelet
# more /var/lib/kubelet/cpu_manager_state-old
{"policyName":"none","defaultCpuSet":"","checksum":3242152201}
参考文档：