kubernetes调度
一.kubernetes调度概述
调度器通过 kubernetes 的 watch 机制来发现集群中新创建且尚未被调度到 Node 上的 Pod。调度器会将发现的每一个未调度的 Pod 调度到一个合适的 Node 上来运行。
kube-scheduler 是 Kubernetes 集群的默认调度器,并且是集群控制面的一部分。如果你真的希望或者有这方面的需求,kube-scheduler 在设计上是允许你自己写一个调度组件并替换原有的 kube-scheduler。
在做调度决定时需要考虑的因素包括:单独和整体的资源请求、硬件/软件/策略限制、亲和以及反亲和要求、数据局域性、负载间的干扰等等。
默认策略可以参考:https://kubernetes.io/zh/docs/concepts/scheduling/kube-scheduler/
调度框架:https://kubernetes.io/zh/docs/concepts/configuration/scheduling-framework/
二.nodename节点选择约束
nodeName 是节点选择约束的最简单方法,但一般不推荐。如果 nodeName 在 PodSpec 中指定了,则它优先于其他的节点选择方法。
使用 nodeName 来选择节点的一些限制:
- 如果指定的节点不存在。
- 如果指定的节点没有资源来容纳 pod,则pod 调度失败。
- 云环境中的节点名称并非总是可预测或稳定的。
脚本内容,指定nodename server2
[root@server1 schedu]# cat pod1.yml
apiVersion: v1
kind: Pod
metadata:
name: nginx
spec:
containers:
- name: nginx
image: nginx
nodeName: server2
执行脚本,nginx部署在server2上
[root@server1 schedu]# kubectl apply -f pod1.yml
pod/nginx created
[root@server1 schedu]# kubectl get pod
NAME READY STATUS RESTARTS AGE
nginx 1/1 Running 0 5s
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx 1/1 Running 0 19s 10.244.179.115 server2 <none> <none>
三.nodeSelector节点选择约束
脚本内容,指定标签disktype=ssd
,具有标签内容的节点可以部署
[root@server1 schedu]# vim pod2.yml
[root@server1 schedu]# kubectl delete pod nginx
pod "nginx" deleted
[root@server1 schedu]# vim pod
pod1.yml pod2.yml
[root@server1 schedu]# vim pod2.yml
[root@server1 schedu]# cat pod2.yml
apiVersion: v1
kind: Pod
metadata:
name: nginx
labels:
env: test
spec:
containers:
- name: nginx
image: nginx
imagePullPolicy: IfNotPresent
nodeSelector:
disktype: ssd
执行脚本,查看pod,由于未识别到指定标签节点,处于pending状态,
[root@server1 schedu]# kubectl apply -f pod2.yml
pod/nginx created
[root@server1 schedu]# kubectl get pod
NAME READY STATUS RESTARTS AGE
nginx 0/1 Pending 0 5s
为server2添加标签,查看pod,部署成功
[root@server1 schedu]# kubectl label nodes server2 disktype=ssd
node/server2 labeled
[root@server1 schedu]# kubectl get pod
NAME READY STATUS RESTARTS AGE
nginx 1/1 Running 0 20s
四.亲和与反亲和
亲和与反亲和
- nodeSelector 提供了一种非常简单的方法来将 pod 约束到具有特定标签的节点上。亲和/反亲和功能极大地扩展了你可以表达约束的类型。
- 你可以发现规则是“软”/“偏好”,而不是硬性要求,因此,如果调度器无法满足该要求,仍然调度该 pod
- 你可以使用节点上的 pod 的标签来约束,而不是使用节点本身的标签,来允许哪些 pod 可以或者不可以被放置在一起。
1.节点亲和
- requiredDuringSchedulingIgnoredDuringExecution 必须满足
- preferredDuringSchedulingIgnoredDuringExecution 倾向满足
IgnoreDuringExecution 表示如果在Pod运行期间Node的标签发生变化,导致亲和性策略不能满足,则继续运行当前的Pod。
nodeaffinity还支持多种规则匹配条件的配置如
- In:label 的值在列表内
- NotIn:label 的值不在列表内
- Gt:label 的值大于设置的值,不支持Pod亲和性
- Lt:label 的值小于设置的值,不支持pod亲和性
- Exists:设置的label 存在
- DoesNotExist:设置的 label 不存在
参考:https://kubernetes.io/zh/docs/concepts/configuration/assign-pod-node/
节点反亲和
脚本内容,指定pod node-affinity
不在server4上部署
[root@server1 schedu]# cat pod3.yml
apiVersion: v1
kind: Pod
metadata:
name: node-affinity
spec:
containers:
- name: nginx
image: nginx
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: kubernetes.io/hostname
operator: NotIn
values:
- server4
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 1
preference:
matchExpressions:
- key: disktype
operator: In
values:
- ssd
执行脚本,查看node-affinity
部署于server2
[root@server1 schedu]# kubectl apply -f pod3.yml
pod/node-affinity created
[root@server1 schedu]# kubectl get pod
NAME READY STATUS RESTARTS AGE
node-affinity 1/1 Running 0 8s
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE
node-affinity 1/1 Running 0 15s 10.244.179.117 server2 <none> <none>
节点亲和
脚本内容,指定pod node-affinity
在server4上部署
[root@server1 schedu]# cat pod3.yml
apiVersion: v1
kind: Pod
metadata:
name: node-affinity
spec:
containers:
- name: nginx
image: nginx
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: kubernetes.io/hostname
operator: In
values:
- server4
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 1
preference:
matchExpressions:
- key: disktype
operator: In
values:
- ssd
执行脚本,查看node-affinity
部署于server4
[root@server1 schedu]# kubectl apply -f pod3.yml
pod/node-affinity created
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
node-affinity 1/1 Running 0 5s 10.244.22.61 server4 <none> <none>
2.pod亲和与反亲和
pod 亲和性和反亲和性
- podAffinity 主要解决POD可以和哪些POD部署在同一个拓扑域中的问题(拓扑域用主机标签实现,可以是单个主机,也可以是多个主机组成的cluster、zone等。)
- podAntiAffinity主要解决POD不能和哪些POD部署在同一个拓扑域中的问题。它们处理的是Kubernetes集群内部POD和POD之间的关系。
- Pod 间亲和与反亲和在与更高级别的集合(例如 ReplicaSets,StatefulSets,Deployments 等)一起使用时,它们可能更加有用。可以轻松配置一组应位于相同定义拓扑(例如,节点)中的工作负载。
pod亲和
脚本内容,设置nginx
与mysql
为pod亲和,即二者需要部署在同一节点上
[root@server1 schedu]# cat pod4.yml
apiVersion: v1
kind: Pod
metadata:
name: nginx
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx
---
apiVersion: v1
kind: Pod
metadata:
name: mysql
labels:
app: mysql
spec:
containers:
- name: mysql
image: mysql
env:
- name: "MYSQL_ROOT_PASSWORD"
value: "westos"
affinity:
podAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
- labelSelector:
matchExpressions:
- key: app
operator: In
values:
- nginx
topologyKey: kubernetes.io/hostname
执行脚本,mysql
与nginx
均位于server2
[root@server1 schedu]# kubectl apply -f pod4.yml
pod/nginx created
pod/mysql created
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
mysql 1/1 Running 0 11s 10.244.179.125 server2 <none> <none>
nginx 1/1 Running 0 11s 10.244.179.124 server2 <none> <none>
pod反亲和
脚本内容,设置nginx
与mysql
为pod反亲和,即二者需要部署在不同节点上
[root@server1 schedu]# cat pod4.yml
apiVersion: v1
kind: Pod
metadata:
name: nginx
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx
---
apiVersion: v1
kind: Pod
metadata:
name: mysql
labels:
app: mysql
spec:
containers:
- name: mysql
image: mysql:5.7
env:
- name: "MYSQL_ROOT_PASSWORD"
value: "westos"
affinity:
podAntiAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
- labelSelector:
matchExpressions:
- key: app
operator: In
values:
- nginx
topologyKey: kubernetes.io/hostname
执行脚本,mysql
与nginx
分别位于server4
和server2
[root@server1 schedu]# kubectl apply -f pod4.yml
pod/nginx unchanged
pod/mysql created
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
mysql 1/1 Running 0 36s 10.244.22.63 server4 <none> <none>
nginx 1/1 Running 0 7m8s 10.244.179.64 server2 <none> <none>
五.Taints(污点)
1.Taints(污点)简介
NodeAffinity节点亲和性,是Pod上定义的一种属性,使Pod能够按我们的要求调度到某个Node上,而Taints则恰恰相反,它可以让Node拒绝运行Pod,甚至驱逐Pod。
Taints(污点)是Node的一个属性,设置了Taints后,所以Kubernetes是不会将Pod调度到这个Node上的,于是Kubernetes就给Pod设置了个属性Tolerations(容忍),只要Pod能够容忍Node上的污点,那么Kubernetes就会忽略Node上的污点,就能够(不是必须)把Pod调度过去。
如Kubernetes集群主机,就是被设置了污点,因此一般部署pod时候不选择该节点为部署节点
查看matser的污点 master:NoSchedule
[root@server1 schedu]# kubectl describe nodes server1 | grep Taints
Taints: node-role.kubernetes.io/master:NoSchedule
可以使用命令 kubectl taint 给节点增加一个 taint:
- $ kubectl taint nodes node1 key=value:NoSchedule //创建
- $ kubectl describe nodes server1 |grep Taints //查询
- $ kubectl taint nodes node1 key:NoSchedule- //删除
其中[effect] 可取值: [ NoSchedule | PreferNoSchedule | NoExecute ]
- NoSchedule:POD 不会被调度到标记为 taints 节点。
- PreferNoSchedule:NoSchedule 的软策略版本。
- NoExecute:该选项意味着一旦 Taint 生效,如该节点内正在运行的 POD 没有对应 Tolerate 设置,会直接被逐出。
2.Nodename可以无视任何污点
设置指定nodename: server1
上部署,server1为集群master
[root@server1 schedu]# cat pod1.yml
apiVersion: v1
kind: Pod
metadata:
name: nginx
labels:
env: test
spec:
containers:
- name: nginx
image: nginx
nodeName: server1
server1污点
[root@server1 schedu]# kubectl describe nodes server1 | grep Taints
Taints: node-role.kubernetes.io/master:NoSchedule
查看pod,已经部署在server1,说明nodename
可以掩盖污点
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx 1/1 Running 0 94s 10.244.116.142 server1 <none>
3.标签方式选择node
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx 0/1 Pending 0 13s
脚本内容,指定标签 roles=master
节点安装
[root@server1 schedu]# cat pod1.yml
apiVersion: v1
kind: Pod
metadata:
name: nginx
labels:
env: test
spec:
containers:
- name: nginx
image: nginx
nodeSelector:
roles: master
授予server1 标签
[root@server1 schedu]# kubectl label nodes server1 roles=master
node/server1 labeled
执行脚本,查看pod部署,处于pending
状态,说明污点
优先级高于标签选择
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx 0/1 Pending 0 13s <none> <none> <none> <none>
删除标签
[root@server1 schedu]# kubectl label nodes server2 roles-
node/server2 labeled
4.容忍标签
tolerations中定义的key、value、effect,要与node上设置的taint保持一直:
- 如果 operator 是 Exists ,value可以省略。
- 如果 operator 是 Equal ,则key与value之间的关系必须相等。
- 如果不指定operator属性,则默认值为Equal。
还有两个特殊值:
- 当不指定key,再配合Exists 就能匹配所有的key与value ,可以容忍所有污点。
- 当不指定effect ,则匹配所有的effect
NoSchedule
设置sevrer2
taint key=value:NoSchedule
[root@server1 schedu]# kubectl taint nodes server2 key=value:NoSchedule
node/server2 tainted
测试部署,由于server2
有污点,部署在server4
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx 1/1 Running 0 8s 10.244.22.4 server4 <none> <none>
设置容忍标签,容忍effect: "NoSchedule"
[root@server1 schedu]# cat pod1.yml
apiVersion: v1
kind: Pod
metadata:
name: nginx
labels:
env: test
spec:
containers:
- name: nginx
image: nginx
nodeSelector:
roles: master
tolerations:
- operator: "Exists"
effect: "NoSchedule"
执行脚本,为server2添加rules=master
,查看到nginx部署到server2
,容忍标签生效
[root@server1 schedu]# kubectl apply -f pod1.yml
pod/nginx created
[root@server1 schedu]# kubectl label nodes server2 roles=master
node/server2 labeled
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx 1/1 Running 0 43s 10.244.179.70 server2 <none> <none>
[root@server1 schedu]#
污点注销
[root@server1 schedu]# kubectl taint nodes server2 key:NoSchedule-
node/server2 untainted
NoExecute
查看pod,分别部署于server2和server4
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
myapp-deployment-67f8c948cf-5l5vd 1/1 Running 0 8s 10.244.179.71 server2 <none> <none>
myapp-deployment-67f8c948cf-zffpv 1/1 Running 0 8s 10.244.179.74 server2 <none> <none>
myapp-deployment-67f8c948cf-zhzmm 1/1 Running 0 8s 10.244.22.5 server4 <none> <none>
nginx 1/1 Running 0 13m 10.244.179.70 server2 <none> <none>
nginx-deployment-6456d7c676-8lc4p 1/1 Running 0 8s 10.244.179.75 server2 <none> <none>
nginx-deployment-6456d7c676-9jhdv 1/1 Running 0 8s 10.244.22.3 server4 <none> <none>
nginx-deployment-6456d7c676-znjfr 1/1 Running 0 8s 10.244.22.6 server4 <none> <none>
给Server2节点打上taint: key1=v1:NoExecute
kubectl taint node server2 key1=v1:NoExecute
node/server1 tainted
可以看到server1上的Pod被驱离:
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
myapp-deployment-67f8c948cf-h6zmg 1/1 Running 0 15s 10.244.22.8 server4 <none> <none>
myapp-deployment-67f8c948cf-m886n 1/1 Running 0 15s 10.244.22.13 server4 <none> <none>
myapp-deployment-67f8c948cf-zhzmm 1/1 Running 0 47s 10.244.22.5 server4 <none> <none>
nginx-deployment-6456d7c676-9jhdv 1/1 Running 0 47s 10.244.22.3 server4 <none> <none>
nginx-deployment-6456d7c676-rrbf2 1/1 Running 0 15s 10.244.22.7 server4 <none> <none>
nginx-deployment-6456d7c676-znjfr 1/1 Running 0 47s 10.244.22.6 server4 <none> <none>
在PodSpec中为容器设定容忍标签:
tolerations:
- key: "key1"
operator: "Equal"
value: "v1"
effect: "NoExecute"
为Pod设置容忍后会,server1又可以运行Pod了。
[root@server1 schedu]# kubectl apply -f pod1.yml
pod/nginx created
[root@server1 schedu]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx 1/1 Running 0 6s 10.244.179.72 server2 <none> <none>
5.cordon、drain、delete
影响Pod调度的指令还有:cordon、drain、delete,后期创建的pod都不会被调度到该节点上,但操作的暴力程度不一样。
cordon 停止调度
影响最小,只会将node调为SchedulingDisabled,新创建pod,不会被调度到该节点,节点原有pod不受影响,仍正常对外提供服务。
$ kubectl cordon server3
$ kubectl get node
NAME STATUS ROLES AGE VERSION
server1 Ready <none> 29m v1.17.2
server2 Ready <none> 12d v1.17.2
server3 Ready,SchedulingDisabled <none> 9d v1.17.2
$ kubectl uncordon server3 //恢复
drain 驱逐节点
首先驱逐node上的pod,在其他节点重新创建,然后将节点调为SchedulingDisabled。
$ kubectl drain server3
node/server3 cordoned
evicting pod "web-1"
evicting pod "coredns-9d85f5447-mgg2k"
pod/coredns-9d85f5447-mgg2k evicted
pod/web-1 evicted
node/server3 evicted
$ kubectl uncordon server3
delete 删除节点
最暴力的一个,首先驱逐node上的pod,在其他节点重新创建,然后,从master节点删除该node,master失去对其控制,如要恢复调度,需进入node节点,重启kubelet服务
$ kubectl delete node server3
```# systemctl restart kubelet //基于node的自注册功能,恢复使用
```powershell
# systemctl restart kubelet //基于node的自注册功能,恢复使用