Static Pod, Pod Creation Process, Container Resource Limits

静态Pod

Static containers have the following characteristics：

• Pod由特定节点上的kubelet管理;
• 不能使用控制器;
• Pod名称标识当前节点名称.

在kubeletThe configuration file to enable staticPod的参数：

vi /var/lib/kubelet/config.yaml
...
staticPodPath: /etc/kubernetes/manifests

将部署的pod.yamlplaced in this directory will bekubelet自动创建.

K8S组件中除了kubeletOutside are running in the form of container.在K8SBefore the cluster is set up,Only by the above methodkubelet创建静态Pod.

MasterThe components deployed by the node areapi-server、controller-manager、scheduler和etcd;NodeThe components deployed by the node arekube-proxy、kubelet和docker.

[[email protected] ~]# kubectl get pods -n kube-system
NAME                                       READY   STATUS    RESTARTS          AGE
calico-kube-controllers-5c64b68895-kthkh   1/1     Running   4 (6h16m ago)     8d
calico-node-6mxfm                          1/1     Running   4 (6h16m ago)     8d
calico-node-nzn8p                          1/1     Running   150 (6h16m ago)   8d
calico-node-vg6s9                          1/1     Running   4 (6h16m ago)     8d
coredns-6d8c4cb4d-k9245                    1/1     Running   4 (6h16m ago)     8d
coredns-6d8c4cb4d-xzl49                    1/1     Running   4 (6h16m ago)     8d
etcd-k8s-master                            1/1     Running   4 (6h16m ago)     8d
kube-apiserver-k8s-master                  1/1     Running   4 (6h16m ago)     8d
kube-controller-manager-k8s-master         1/1     Running   4 (6h16m ago)     8d
kube-proxy-58kfs                           1/1     Running   5 (6h16m ago)     8d
kube-proxy-75z28                           1/1     Running   4 (6h16m ago)     8d
kube-proxy-f9522                           1/1     Running   4 (6h16m ago)     8d
kube-scheduler-k8s-master                  1/1     Running   4 (6h16m ago)     8d
metrics-server-798c598bb8-j69cc            1/1     Running   3 (6h16m ago)     6d19h

[[email protected] ~]# kubectl get deploy -n kube-system
NAME                      READY   UP-TO-DATE   AVAILABLE   AGE
calico-kube-controllers   1/1     1            1           8d
coredns                   2/2     2            2           8d
metrics-server            1/1     1            1           6d19h

可以发现,etcd-k8s-master、kube-apiserver-k8s-master、kube-controller-manager-k8s-master、kube-scheduler-k8s-master都是以masterThe name of the node ends with,And there is no corresponding controller.

实际上,masterThese four components of the node are all generated by this nodekubelet直接管理的静态Pod部署.

[[email protected] ~]# grep staticPod /var/lib/kubelet/config.yaml
staticPodPath: /etc/kubernetes/manifests
[[email protected] ~]#
[[email protected] ~]# ls /etc/kubernetes/manifests/
etcd.yaml  kube-apiserver.yaml  kube-controller-manager.yaml  kube-scheduler.yaml

Pod创建流程

k8s基于list-watch机制的控制器架构,实现组件间交互的解耦.其他组件监控自己负责的资源,当这些资源发生变化时,kube-apiserver会通知这些组件,这个过程类似于发布与订阅.

from the moment we issue the command,一个PodThe general process that is created is as follows：

1. 所有使用kubectlCommands issued are sent toapi-server,The latter after receiving the writtenetcd中.
2. etcdWrite successfullyapi-serverReturn to write the message of success,否则会返回错误信息.
3. api-serverNotify the schedulerscheduler把要创建的Pod分配到合适的节点上.
4. schedulerChoose good after the target node,打上标记,and report the results toapi-server.
5. api-server收到scheduler反馈的Pod-Node绑定信息后,将其写入etcd.
6. api-server通知目标node节点上的kubelet组件创建Pod.
7. 目标节点上的kubelet组件调用dockeror other container engines to create containers,and report the creation result information toapi-server.
8. api-server将收到的PodCreate result information writeetcd.

Since creating a separatePodwill not use the controller,So the above process does not involvecontroller-manager.此外,也没有创建Service,So it doesn't involvekube-proxy组件.

容器资源限制

Use the following two properties to limit what the container can useCPU和内存资源的上限：

• resources.limits.cpu
• resources.limits.memory

Set the minimum resource requirements used by the container through the following two properties（下限）,作为容器调度时资源分配的依据：

• resources.requests.cpu
• resources.requests.memory

k8s会根据requestsvalue to find nodes with enough available resources to schedulePod.

注意事项：

• When no resource limit is configured,The container can use all the resources of the node where it is located.
• limitsGenerally lower than the node configuration20%.
• requests的值必须小于limits,一般低于limits的20%到30%.
• 只设置limits限制时,requestsBy default alsolimits的值.
• requestsis a reserved resource,The actual resources used when the container is running may be less thanrequests.
• 如果requests配置过高,will result in node allocationPod过少,will lead to idle resources.
• 如果requests配置过低,will result in node allocationPod过多,Saturation may lead to resources.
• CPUThe unit of measure for resources is millicores,1核CPU=1000m,0.5=500m.

View the allocation of resources of the current node（Allocated resources）:

kubectl describe node k8s-node1