目录
node调度策略nodeName和nodeSelector
指定nodeName
指定nodeSelector
node亲和性
node节点亲和性
硬亲和性
软亲和性
污点与容忍度
本文主要介绍了在pod中，与node相关的调度策略，亲和性，污点与容忍度等的内容
node调度策略nodeName和nodeSelector
在创建pod等资源时，可以通过调整字段进行node调度，指定资源调度到满足何种条件的node
指定nodeName
vim testpod1.yaml
apiVersion: v1
kind: Pod
metadata:
name: testpod1
namespace: default
labels:
app: tomcat
spec:
nodeName: ws-k8s-node1 #增加字段，将这个pod调度到node1
containers:
- name: test
image: docker.io/library/tomcat
imagePullPolicy: IfNotPresent
kubectl apply -f testpod1.yaml
kubectl get pods #可以看到已经调度到node1上了
testpod1 1/1 Running 0 116s 10.10.179.9 ws-k8s-node1 <none> <none>
指定nodeSelector
vim testpod2.yaml
apiVersion: v1
kind: Pod
metadata:
name: testpod2
namespace: default
labels:
app: tomcat
spec:
nodeSelector: #添加nodeSelector选项，
admin: ws #调度到具有admin=ws标签的node上
containers:
- name: test
image: docker.io/library/tomcat
imagePullPolicy: IfNotPresent
kubectl apply -f testpod2.yaml
但因为我没有admin=ws标签的node，所以应用后pod处于pending状态
#现在我给node1的节点打个标签
#kubectl --help | grep -i label
#kubectl label --help
Examples:
# Update pod 'foo' with the label 'unhealthy' and the value 'true'
#kubectl label pods foo unhealthy=true
kubectl label nodes ws-k8s-node1 admin=ws
#node/ws-k8s-node1 labeled
#调度情况恢复正常
kubectl get pods | grep testpod2
testpod2 1/1 Running 0 11m
#删除node标签
kubectl label nodes ws-k8s-node1 admin-
#删除testpod2
kubectl delete pods testpod2
如果同时使用nodeName和nodeSelector，则会报错亲和性错误，无法正常部署；
如果nodeName和nodeSelector指定的node同时满足这两项的条件，就可以部署
node亲和性
亲和性在Kubernetes中起着重要作用，通过定义规则和条件，它允许我们实现精确的Pod调度、资源优化、高性能计算以及容错性和可用性的增强。通过利用亲和性，我们可以更好地管理和控制集群中的工作负载，并满足特定的业务需求。
#查看帮助
kubectl explain pods.spec.affinity
RESOURCE: affinity <Object>
DESCRIPTION:
If specified, the pod's scheduling constraints
Affinity is a group of affinity scheduling rules.
FIELDS:
nodeAffinity <Object> #node亲和性
Describes node affinity scheduling rules for the pod.
podAffinity <Object> #pod亲和性
Describes pod affinity scheduling rules (e.g. co-locate this pod in the
same node, zone, etc. as some other pod(s)).
podAntiAffinity <Object> #pod反亲和性
Describes pod anti-affinity scheduling rules (e.g. avoid putting this pod
in the same node, zone, etc. as some other pod(s)).
node节点亲和性
在创建pod时，会根据nodeaffinity来寻找最适合该pod的条件的node
#查找帮助
kubectl explain pods.spec.affinity.nodeAffinity
KIND: Pod
VERSION: v1
RESOURCE: nodeAffinity <Object>
DESCRIPTION:
Describes node affinity scheduling rules for the pod.
Node affinity is a group of node affinity scheduling rules.
FIELDS:
preferredDuringSchedulingIgnoredDuringExecution <[]Object>
requiredDuringSchedulingIgnoredDuringExecution <Object>
#软亲和性，如果所有都不满足条件，也会找一个节点将就
preferredDuringSchedulingIgnoredDuringExecution
#硬亲和性，必须满足，如果不满足则不找节点，宁缺毋滥
requiredDuringSchedulingIgnoredDuringExecution
硬亲和性
kubectl explain pods.spec.affinity.nodeAffinity.requiredDuringSchedulingIgnoredDuringExecution
#nodeSelectorTerms <[]Object> -required-
# Required. A list of node selector terms. The terms are ORed.
kubectl explain pods.spec.affinity.nodeAffinity.requiredDuringSchedulingIgnoredDuringExecution.nodeSelectorTerms
FIELDS:
matchExpressions <[]Object> #匹配表达式
A list of node selector requirements by node's labels.
matchFields <[]Object> #匹配字段
A list of node selector requirements by node's fields.
#匹配表达式
kubectl explain pods.spec.affinity.nodeAffinity.requiredDuringSchedulingIgnoredDuringExecution.nodeSelectorTerms.matchExpressions
key <string> -required-
operator <string> -required-
values <[]string>
#可用operator
- `"DoesNotExist"`
- `"Exists"`
- `"Gt"`
- `"In"`
- `"Lt"`
- `"NotIn"`
#
vim ying-pod.yaml
apiVersion: v1
kind: Pod
metadata:
name: ying-pod
labels:
app: tomcat
user: ws
spec:
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: name #去找key=name
opertor: In # name = ws或=wws
values:
- ws
- wss
containers:
- name: test1
namespace: default
image: docker.io/library/tomcat
imagePullPolicy: IfNotPresent
kubectl apply -f ying-pod.yaml
#需要name=ws或name=wws，但是没有节点有标签，而且是硬亲和
#所以pod会处于pending状态
kubectl get pods | grep ying
ying-pod 0/1 Pending 0 15m
#修改node标签
kubectl label nodes ws-k8s-node1 name=ws
#开始构建，并且已经到node1节点了
kubectl get pod -owide | grep ying
ying-pod 0/1 ContainerCreating 0 80s <none> ws-k8s-node1 <none> <none>
#删除标签
kubectl label nodes ws-k8s-node1 name-
软亲和性
vim ruan-pod.yaml
apiVersion: v1
kind: Pod
metadata:
name: ruan-pod
namespace: default
spec:
containers:
- name: test
image: docker.io/library/alpine
imagePullPolicy: IfNotPresent
affinity:
nodeAffinity:
preferredDuringSchedulingIgnoredDuringExecution: #必选preference和weight
- preference:
matchExpressions:
- key: name
operate: In #还有Exists，Gt，Lt，NotIn等
values:
- ws
weight: 50 #软亲和性有“权重”说法，权重更高的更优先，范围1-100
- preference:
matchExpressions:
- key: name
operate: In
values:
- wws
weight: 70 #设置的比上面的高，用以做测试
kubectl apply -f ruan-pod.yaml
#不满足条件，所以随机找一个进行调度，能看到调度到了node2上
kubectl get pod -owide | grep ruan
ruan-pod 0/1 ContainerCreating 0 3m24s <none> ws-k8s-node2 <none> <none>
#修改node1的标签name=ws
kubectl label nodes ws-k8s-node1 name=ws
kubectl delete -f ruan-pod.yaml #删除再重新创建
kubectl apply -f ruan-pod.yaml
kubectl get pods -owide | grep ruan #调整到了node1上
ruan-pod 0/1 ContainerCreating 0 2s <none> ws-k8s-node1 <none> <none>
#修改node2的标签name=wws，此时node2权重比node1高
kubectl label nodes ws-k8s-node2 name=wss
kubectl delete -f ruan-pod.yaml
kubectl apply -f ruan-pod.yaml
kubectl get pods -owide | grep ruan #没有变化，还在node1
ruan-pod 0/1 ContainerCreating 0 4m29s <none> ws-k8s-node1 <none> <none>
#因为yaml的匹配顺序，已经匹配到了name=ws，如果没有另外标签不同的则不会变化
#修改ruan-pod.yaml
...
- preference:
matchExpressions:
- key: name
operator: In
values:
- ws
weight: 50
- preference:
matchExpressions:
- key: names
operator: In
values:
- wws
weight: 70
...
#添加node2标签name1=wws，权重比node1高，且标签key不同
kubectl label nodes ws-k8s-node2 names=wws
kubectl delete -f ruan-pod.yaml
kubectl apply -f ruan-pod.yaml
kubectl get po -owide | grep ruan #可以看到ruan-pod已经回到了node2上
ruan-pod 0/1 ContainerCreating 0 3m47s <none> ws-k8s-node2 <none> <none>
#清理环境
kubectl label nodes ws-k8s-node1 name-
kubectl label nodes ws-k8s-node2 names-
kubectl delete -f ruan-pod.yaml
kubectl delete -f ying-pod.yaml --fore --grace-period=0 #强制删除
污点与容忍度
污点类似于标签，可以给node打taints，如果pod没有对node上的污点有容忍，那么就不会调度到该node上。
在创建pod时可以通过tolerations来定义pod对于污点的容忍度
#查看node上的污点
#master节点是默认有污点
kubectl describe node ws-k8s-master1 | grep -i taint
Taints: node-role.kubernetes.io/control-plane:NoSchedule
#node默认没有污点
kubectl describe node ws-k8s-node1 | grep -i taint
Taints: <none>
#kubectl explain nodes.spec.taints查看帮助
kubectl explain nodes.spec.taints.effect
1.NoExecute
对已调度的pod不影响，仅对新需要调度的pod进行影响
2.NoSchedule
对已调度和新调度的pod都会有影响
3.PreferNoSchedule
软性的NoSchedule，就算不满足条件也可以调度到不容忍的node上
#查看当前master节点pod容忍情况
kubectl get pods -n kube-system -owide
kubectl describe pods kube-proxy-bg7ck -n kube-system | grep -i tolerations -A 10
Tolerations: op=Exists
node.kubernetes.io/disk-pressure:NoSchedule op=Exists
node.kubernetes.io/memory-pressure:NoSchedule op=Exists
node.kubernetes.io/network-unavailable:NoSchedule op=Exists
node.kubernetes.io/not-ready:NoExecute op=Exists
node.kubernetes.io/pid-pressure:NoSchedule op=Exists
node.kubernetes.io/unreachable:NoExecute op=Exists
node.kubernetes.io/unschedulable:NoSchedule op=Exists
Events: <none>
#给node1打一个污点，使其不接受
kubectl taint node ws-k8s-node1 user=ws:NoSchedule
#创建wudian.yaml进行测试
cat > wudian.yaml << EOF
apiVersion: v1
kind: Pod
metadata:
name: wudain-pod
namespace: default
labels:
app: app1
spec:
containers:
- name: wudian-pod
image: docker.io/library/tomcat
imagePullPolicy: IfNotPresent
EOF
kubectl apply -f wudian.yaml
#wudian-pod调度到了node2
kubectl get pods -owide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
wudain-pod 1/1 Running 0 18s 10.10.234.72 ws-k8s-node2 <none> <none>
#给node2添加污点
kubectl taint node ws-k8s-node2 user=xhy:NoExecute
#再查看发现wudain-pood已经被删除
kubectl get pods -owide
No resources found in default namespace.
#再次创建变成离线状态
kubectl apply -f wudian.yaml
kubectl get pods -owide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
wudain-pod 0/1 Pending 0 3s <none> <none> <none> <none>
#查看当前node污点状态
kubectl describe node ws-k8s-node1 | grep -i taint
Taints: user=ws:NoSchedule
kubectl describe node ws-k8s-node2 | grep -i taint
Taints: user=xhy:NoExecute
#创建带有容忍度的pod wudian2.yaml
cat > wudian2.yaml << EOF
apiVersion: v1
kind: Pod
metadata:
name: wudain2-pod
namespace: default
labels:
app: app1
spec:
containers:
- name: wudian2-pod
image: docker.io/library/tomcat
imagePullPolicy: IfNotPresent
tolerations: #容忍度
- key: "user"
operator: "Equal" #equal表示等于，exists代表存在
value: "ws" #根据字段，表示能容忍user=ws的污点
#如果operator为exists且value为空则代表容忍所有key相同的
effect: "NoSchedule" #需要准确匹配容忍等级，如果不匹配则不会生效
# tolerationSeconds: 1800 effect为NoExecute时才能使用，表示容忍污染的时间，默认是0，即永远容忍
EOF
#现在wudian2是能容忍node1的污点的
kubectl apply -f wudian2.yaml
kubectl get pods -owide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
wudain-pod 0/1 Pending 0 21m <none> <none> <none> <none>
wudain2-pod 1/1 Running 0 15s 10.10.179.13 ws-k8s-node1 <none> <none>
#创建带有容忍度的pod wudian3.yaml
cat > wudian3.yaml << EOF
apiVersion: v1
kind: Pod
metadata:
name: wudain3-pod
namespace: default
labels:
app: app1
spec:
containers:
- name: wudian3-pod
image: docker.io/library/tomcat
imagePullPolicy: IfNotPresent
tolerations: #容忍度
- key: "user"
operator: "Exists" #equal表示等于，exists代表存在
value: "" #根据字段，表示能容忍user=ws的污点
#如果operator为exist且value为空则代表容忍所有key相同的
effect: "NoExecute" #需要准确匹配容忍等级，如果不匹配则不会生效
tolerationSeconds: 1800 #effect为NoExecute时才能使用，表示容忍污染的时间，默认是0，即永远容忍
EOF
kubectl apply -f wudian3.yaml
#wudian3运行在node2上
kubectl get pods -owide | grep -i node2
wudain3-pod 1/1 Running 0 59s 10.10.234.73 ws-k8s-node2 <none> <none>
#清理环境
kubectl delete -f wudian.yaml
kubectl delete -f wudian2.yaml
kubectl delete -f wudian3.yaml
kubectl taint node ws-k8s-node1 user-
kubectl taint node ws-k8s-node2 user-
