您好,登录后才能下订单哦!
关于作者 刘海平(HappyLau )云计算高级顾问 目前在腾讯云从事公有云相关工作,曾就职于酷狗,EasyStack,拥有多年公有云+私有云计算架构设计,运维,交付相关经验,参与了酷狗,南方电网,国泰君安等大型私有云平台建设,精通Linux,Kubernetes,OpenStack,Ceph等开源技术,在云计算领域具有丰富实战经验,拥有RHCA/OpenStack/Linux授课经验。
上一篇文章中kubernetes系列教程(六)kubernetes资源管理和服务质量初步介绍了kubernetes中的resource资源调度和服务质量Qos,介绍了kubernetes中如何定义pod的资源和资源调度,以及设置resource之后的优先级别Qos,接下来介绍kubernetes系列教程pod的调度机制。
kubernets是容器编排引擎,其中最主要的一个功能是容器的调度,通过kube-scheduler实现容器的完全自动化调度,调度周期分为:调度周期Scheduling Cycle和绑定周期Binding Cycle,其中调度周期细分为过滤filter和weight称重,按照指定的调度策略将满足运行pod节点的node赛选出来,然后进行排序;绑定周期是经过kube-scheduler调度优选的pod后,由特定的node节点watch然后通过kubelet运行。
过滤阶段包含预选Predicate和scoring排序,预选是筛选满足条件的node,排序是最满足条件的node打分并排序,预选的算法包含有:
过滤条件需要检查node上满足的条件,可以通过kubectl describe node node-id方式查看,如下图:
优选调度算法有:
nodeName是PodSpec中的一个字段,可以通过pod.spec.nodeName指定将pod调度到某个具体的node节点上,该字段比较特殊一般都为空,如果有设置nodeName字段,kube-scheduler会直接跳过调度,在特定节点上通过kubelet启动pod。通过nodeName调度并非是集群的智能调度,通过指定调度的方式可能会存在资源不均匀的情况,建议设置Guaranteed的Qos,防止资源不均时候Pod被驱逐evince。如下以创建一个pod运行在node-3上为例:
[root@node-1 demo]# cat nginx-nodeName.yaml
apiVersion: v1
kind: Pod
metadata:
name: nginx-run-on-nodename
annotations:
kubernetes.io/description: "Running the Pod on specific nodeName"
spec:
containers:
- name: nginx-run-on-nodename
image: nginx:latest
ports:
- name: http-80-port
protocol: TCP
containerPort: 80
nodeName: node-3 #通过nodeName指定将nginx-run-on-nodename运行在特定节点node-3上
[root@node-1 demo]# kubectl apply -f nginx-nodeName.yaml
pod/nginx-run-on-nodename created
[root@node-1 demo]# kubectl get pods nginx-run-on-nodename -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-run-on-nodename 1/1 Running 0 6m52s 10.244.2.15 node-3 <none> <none>
nodeSelector是PodSpec中的一个字段,nodeSelector是最简单实现将pod运行在特定node节点的实现方式,其通过指定key和value键值对的方式实现,需要node设置上匹配的Labels,节点调度的时候指定上特定的labels即可。如下以node-2添加一个app:web的labels,调度pod的时候通过nodeSelector选择该labels:
[root@node-1 demo]# kubectl label node node-2 app=web
node/node-2 labeled
[root@node-1 demo]# kubectl get nodes --show-labels
NAME STATUS ROLES AGE VERSION LABELS
node-1 Ready master 15d v1.15.3 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-1,kubernetes.io/os=linux,node-role.kubernetes.io/master=
node-2 Ready <none> 15d v1.15.3 app=web,beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-2,kubernetes.io/os=linux
node-3 Ready <none> 15d v1.15.3 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-3,kubernetes.io/os=linux
[root@node-1 demo]# cat nginx-nodeselector.yaml
apiVersion: v1
kind: Pod
metadata:
name: nginx-run-on-nodeselector
annotations:
kubernetes.io/description: "Running the Pod on specific node by nodeSelector"
spec:
containers:
- name: nginx-run-on-nodeselector
image: nginx:latest
ports:
- name: http-80-port
protocol: TCP
containerPort: 80
nodeSelector: #通过nodeSelector将pod调度到特定的labels
app: web
[root@node-1 demo]# kubectl apply -f nginx-nodeselector.yaml
pod/nginx-run-on-nodeselector created
[root@node-1 demo]# kubectl get pods nginx-run-on-nodeselector -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-run-on-nodeselector 1/1 Running 0 51s 10.244.1.24 node-2 <none> <none>
系统默认预先定义有多种内置的labels,这些labels可以标识node的属性,如arch架构,操作系统类型,主机名等
affinity/anti-affinity和nodeSelector功能相类似,相比于nodeSelector,affinity的功能更加丰富,未来会取代nodeSelector,affinity增加了如下的一些功能增强:
下面通过一个例子来演示node affinity的使用,requiredDuringSchedulingIgnoredDuringExecution指定需要满足的条件,preferredDuringSchedulingIgnoredDuringExecution指定优选的条件,两者之间取与关系。
[root@node-1 ~]# kubectl get nodes --show-labels
NAME STATUS ROLES AGE VERSION LABELS
node-1 Ready master 15d v1.15.3 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-1,kubernetes.io/os=linux,node-role.kubernetes.io/master=
node-2 Ready <none> 15d v1.15.3 app=web,beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-2,kubernetes.io/os=linux
node-3 Ready <none> 15d v1.15.3 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-3,kubernetes.io/os=linux
[root@node-1 demo]# cat nginx-node-affinity.yaml
apiVersion: v1
kind: Pod
metadata:
name: nginx-run-node-affinity
annotations:
kubernetes.io/description: "Running the Pod on specific node by node affinity"
spec:
containers:
- name: nginx-run-node-affinity
image: nginx:latest
ports:
- name: http-80-port
protocol: TCP
containerPort: 80
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: kubernetes.io/hostname
operator: In
values:
- node-1
- node-2
- node-3
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 1
preference:
matchExpressions:
- key: app
operator: In
values: ["web"]
[root@node-1 demo]# kubectl apply -f nginx-node-affinity.yaml
pod/nginx-run-node-affinity created
[root@node-1 demo]# kubectl get pods --show-labels nginx-run-node-affinity -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES LABELS
nginx-run-node-affinity 1/1 Running 0 106s 10.244.1.25 node-2 <none> <none> <none>
本文介绍了kubernetes中的调度机制,默认创建pod是全自动调度机制,调度由kube-scheduler实现,调度过程分为两个阶段调度阶段(过滤和沉重排序)和绑定阶段(在node上运行pod)。通过干预有四种方式:
调度框架介绍:https://kubernetes.io/docs/concepts/configuration/scheduling-framework/
Pod调度方法:https://kubernetes.io/docs/concepts/configuration/assign-pod-node/
当你的才华撑不起你的野心时,你就应该静下心来学习
返回kubernetes系列教程目录
免责声明:本站发布的内容(图片、视频和文字)以原创、转载和分享为主,文章观点不代表本网站立场,如果涉及侵权请联系站长邮箱:is@yisu.com进行举报,并提供相关证据,一经查实,将立刻删除涉嫌侵权内容。