pod污点taint 与容忍度tolerations详解

一.系统环境

服务器版本docker软件版本Kubernetes(k8s)集群版本CPU架构CentOS Linux release 7.4.1708 (Core)Docker version 20.10.12v1.21.9x86_64

Kubernetes集群架构：k8scloude1作为master节点，k8scloude2，k8scloude3作为worker节点

服务器操作系统版本CPU架构进程功能描述k8scloude1/192.168.110.130CentOS Linux release 7.4.1708 (Core)x86_64docker，kube-apiserver，etcd，kube-scheduler，kube-controller-manager，kubelet，kube-proxy，coredns，calicok8s master节点k8scloude2/192.168.110.129CentOS Linux release 7.4.1708 (Core)x86_64docker，kubelet，kube-proxy，calicok8s worker节点k8scloude3/192.168.110.128CentOS Linux release 7.4.1708 (Core)x86_64docker，kubelet，kube-proxy，calicok8s worker节点

二.前言

本文介绍污点taint 与容忍度tolerations，可以影响pod的调度。

使用污点taint 与容忍度tolerations的前提是已经有一套可以正常运行的Kubernetes集群，关于Kubernetes(k8s)集群的安装部署，可以查看博客《Centos7 安装部署Kubernetes(k8s)集群》

三.污点taint

3.1 污点taint概览

节点亲和性是 Pod 的一种属性，它使 Pod 被吸引到一类特定的节点（这可能出于一种偏好，也可能是硬性要求）。污点（Taint）则相反——它使节点能够排斥一类特定的 Pod。

3.2 给节点添加污点taint

给节点增加一个污点的语法如下：给节点 node1 增加一个污点，它的键名是 key1，键值是 value1，效果是 NoSchedule。这表示只有拥有和这个污点相匹配的容忍度的 Pod 才能够被分配到 node1 这个节点。

#污点的格式：键=值:NoSchedule
kubectl taint nodes node1 key1=value1:NoSchedule
#只有键没有值的话，格式为：键:NoSchedule
kubectl taint nodes node1 key1:NoSchedule

移除污点语法如下：

kubectl taint nodes node1 key1=value1:NoSchedule-

节点的描述信息里有一个Taints字段，Taints字段表示节点有没有污点

[root@k8scloude1 deploy]# kubectl get nodes -o wide
NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME
k8scloude1 Ready control-plane,master 8d v1.21.0 192.168.110.130 <none> CentOS Linux 7 (Core) 3.10.0-693.el7.x86_64 docker://20.10.12
k8scloude2 Ready <none> 8d v1.21.0 192.168.110.129 <none> CentOS Linux 7 (Core) 3.10.0-693.el7.x86_64 docker://20.10.12
k8scloude3 Ready <none> 8d v1.21.0 192.168.110.128 <none> CentOS Linux 7 (Core) 3.10.0-693.el7.x86_64 docker://20.10.12
[root@k8scloude1 deploy]# kubectl describe nodes k8scloude1
Name: k8scloude1
Roles: control-plane,master
Labels: beta.kubernetes.io/arch=amd64
beta.kubernetes.io/os=linux
kubernetes.io/arch=amd64
kubernetes.io/hostname=k8scloude1
kubernetes.io/os=linux
node-role.kubernetes.io/control-plane=
node-role.kubernetes.io/master=
node.kubernetes.io/exclude-from-external-load-balancers=
Annotations: kubeadm.alpha.kubernetes.io/cri-socket: /var/run/dockershim.sock
node.alpha.kubernetes.io/ttl: 0
projectcalico.org/IPv4Address: 192.168.110.130/24
projectcalico.org/IPv4IPIPTunnelAddr: 10.244.158.64
volumes.kubernetes.io/controller-managed-attach-detach: true
CreationTimestamp: Sun, 09 Jan 2022 16:19:06 +0800
Taints: node-role.kubernetes.io/master:NoSchedule
Unschedulable: false
……

查看节点是否有污点，Taints: node-role.kubernetes.io/master:NoSchedule表示k8s集群的master节点有污点，这是默认就存在的污点，这也是master节点为什么不能运行应用pod的原因。

[root@k8scloude1 deploy]# kubectl describe nodes k8scloude2 | grep -i Taints
Taints: <none>
[root@k8scloude1 deploy]# kubectl describe nodes k8scloude1 | grep -i Taints
Taints: node-role.kubernetes.io/master:NoSchedule
[root@k8scloude1 deploy]# kubectl describe nodes k8scloude3 | grep -i Taints
Taints: <none>

创建pod，nodeSelector:kubernetes.io/hostname: k8scloude1表示pod运行在标签为kubernetes.io/hostname=k8scloude1的节点上。

关于pod的调度详细内容，请查看博客《pod(八)：pod的调度——将 Pod 指派给节点》

[root@k8scloude1 pod]# vim schedulepod4.yaml
[root@k8scloude1 pod]# cat schedulepod4.yaml
apiVersion: v1
kind: Pod
metadata:
creationTimestamp: null
labels:
run: pod1
name: pod1
namespace: pod
spec:
nodeSelector:
kubernetes.io/hostname: k8scloude1
containers:
– image: nginx
imagePullPolicy: IfNotPresent
name: pod1
resources: {}
ports:
– name: http
containerPort: 80
protocol: TCP
hostPort: 80
dnsPolicy: ClusterFirst
restartPolicy: Always
status: {}

标签为kubernetes.io/hostname=k8scloude1的节点为k8scloude1节点

[root@k8scloude1 pod]# kubectl get nodes -l kubernetes.io/hostname=k8scloude1
NAME STATUS ROLES AGE VERSION
k8scloude1 Ready control-plane,master 8d v1.21.0

创建pod，因为k8scloude1上有污点，pod1不能运行在k8scloude1上，所以pod1状态为Pending

[root@k8scloude1 pod]# kubectl apply -f schedulepod4.yaml
pod/pod1 created
#因为k8scloude1上有污点，pod1不能运行在k8scloude1上，所以pod1状态为Pending
[root@k8scloude1 pod]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod1 0/1 Pending 0 9s <none> <none> <none> <none>
[root@k8scloude1 pod]# kubectl delete pod pod1 –force
warning: Immediate deletion does not wait for confirmation that the running resource has been terminated. The resource may continue to run on the cluster indefinitely.
pod \”pod1\” force deleted
[root@k8scloude1 pod]# kubectl get pod -o wide
No resources found in pod namespace.

四.容忍度tolerations

4.1 容忍度tolerations概览

容忍度（Toleration）是应用于 Pod 上的。容忍度允许调度器调度带有对应污点的 Pod。容忍度允许调度但并不保证调度：作为其功能的一部分，调度器也会评估其他参数。

污点和容忍度（Toleration）相互配合，可以用来避免 Pod 被分配到不合适的节点上。每个节点上都可以应用一个或多个污点，这表示对于那些不能容忍这些污点的 Pod，是不会被该节点接受的。

4.2 设置容忍度tolerations

只有拥有和这个污点相匹配的容忍度的 Pod 才能够被分配到 node节点。

查看k8scloude1节点的污点

[root@k8scloude1 pod]# kubectl describe nodes k8scloude1 | grep -i taint
Taints: node-role.kubernetes.io/master:NoSchedule

你可以在 Pod 规约中为 Pod 设置容忍度，创建pod，tolerations参数表示可以容忍污点：node-role.kubernetes.io/master:NoSchedule ，nodeSelector:kubernetes.io/hostname: k8scloude1表示pod运行在标签为kubernetes.io/hostname=k8scloude1的节点上。

[root@k8scloude1 pod]# vim schedulepod4.yaml
[root@k8scloude1 pod]# cat schedulepod4.yaml
apiVersion: v1
kind: Pod
metadata:
creationTimestamp: null
labels:
run: pod1
name: pod1
namespace: pod
spec:
tolerations:
– key: \”node-role.kubernetes.io/master\”
operator: \”Equal\”
value: \”\”
effect: \”NoSchedule\”
nodeSelector:
kubernetes.io/hostname: k8scloude1
containers:
– image: nginx
imagePullPolicy: IfNotPresent
name: pod1
resources: {}
ports:
– name: http
containerPort: 80
protocol: TCP
hostPort: 80
dnsPolicy: ClusterFirst
restartPolicy: Always
status: {}
[root@k8scloude1 pod]# kubectl get pods -o wide
No resources found in pod namespace.
[root@k8scloude1 pod]# kubectl apply -f schedulepod4.yaml
pod/pod1 created

查看pod，即使k8scloude1节点有污点，pod还是正常运行。

taint污点和cordon，drain的区别：某个节点上有污点，可以设置tolerations容忍度，让pod运行在该节点，某个节点被cordon，drain，则该节点不能被分配出去运行pod。

关于cordon，drain的详细信息，请查看博客《cordon节点，drain驱逐节点，delete 节点》

[root@k8scloude1 pod]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod1 1/1 Running 0 4s 10.244.158.84 k8scloude1 <none> <none>
[root@k8scloude1 pod]# kubectl delete pod pod1 –force
warning: Immediate deletion does not wait for confirmation that the running resource has been terminated. The resource may continue to run on the cluster indefinitely.
pod \”pod1\” force deleted
[root@k8scloude1 pod]# kubectl get pods -o wide
No resources found in pod namespace.

注意，tolerations容忍度有两种写法，任选一种即可:

tolerations:
– key: \”key1\”
operator: \”Equal\”
value: \”value1\”
effect: \”NoSchedule\”
tolerations:
– key: \”key1\”
operator: \”Exists\”
effect: \”NoSchedule\”

给k8scloude2节点打标签

[root@k8scloude1 pod]# kubectl label nodes k8scloude2 taint=T
node/k8scloude2 labeled
[root@k8scloude1 pod]# kubectl get node –show-labels
NAME STATUS ROLES AGE VERSION LABELS
k8scloude1 Ready control-plane,master 8d v1.21.0 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8scloude1,kubernetes.io/os=linux,node-role.kubernetes.io/control-plane=,node-role.kubernetes.io/master=,node.kubernetes.io/exclude-from-external-load-balancers=
k8scloude2 Ready <none> 8d v1.21.0 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8scloude2,kubernetes.io/os=linux,taint=T
k8scloude3 Ready <none> 8d v1.21.0 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8scloude3,kubernetes.io/os=linux

对k8scloude2设置污点

#污点taint的格式：键=值:NoSchedule
[root@k8scloude1 pod]# kubectl taint node k8scloude2 wudian=true:NoSchedule
node/k8scloude2 tainted
[root@k8scloude1 pod]# kubectl describe nodes k8scloude2 | grep -i Taints
Taints: wudian=true:NoSchedule

创建pod，tolerations参数表示容忍污点wudian=true:NoSchedule，nodeSelector:taint: T参数表示pod运行在标签为nodeSelector=taint: T的节点。

[root@k8scloude1 pod]# vim schedulepod4.yaml
[root@k8scloude1 pod]# cat schedulepod4.yaml
apiVersion: v1
kind: Pod
metadata:
creationTimestamp: null
labels:
run: pod1
name: pod1
namespace: pod
spec:
tolerations:
– key: \”wudian\”
operator: \”Equal\”
value: \”true\”
effect: \”NoSchedule\”
nodeSelector:
taint: T
containers:
– image: nginx
imagePullPolicy: IfNotPresent
name: pod1
resources: {}
ports:
– name: http
containerPort: 80
protocol: TCP
hostPort: 80
dnsPolicy: ClusterFirst
restartPolicy: Always
status: {}
[root@k8scloude1 pod]# kubectl get pod -o wide
No resources found in pod namespace.
[root@k8scloude1 pod]# kubectl apply -f schedulepod4.yaml
pod/pod1 created

查看pod，k8scloude2节点就算有污点也能运行pod

[root@k8scloude1 pod]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod1 1/1 Running 0 8s 10.244.112.177 k8scloude2 <none> <none>
[root@k8scloude1 pod]# kubectl delete pod pod1 –force
warning: Immediate deletion does not wait for confirmation that the running resource has been terminated. The resource may continue to run on the cluster indefinitely.
pod \”pod1\” force deleted
[root@k8scloude1 pod]# kubectl get pods -o wide
No resources found in pod namespace.

污点容忍的另一种写法：operator: "Exists"，没有value值。

[root@k8scloude1 pod]# vim schedulepod4.yaml
[root@k8scloude1 pod]# cat schedulepod4.yaml
apiVersion: v1
kind: Pod
metadata:
creationTimestamp: null
labels:
run: pod1
name: pod1
namespace: pod
spec:
tolerations:
– key: \”wudian\”
operator: \”Exists\”
effect: \”NoSchedule\”
nodeSelector:
taint: T
containers:
– image: nginx
imagePullPolicy: IfNotPresent
name: pod1
resources: {}
ports:
– name: http
containerPort: 80
protocol: TCP
hostPort: 80
dnsPolicy: ClusterFirst
restartPolicy: Always
status: {}
[root@k8scloude1 pod]# kubectl apply -f schedulepod4.yaml
pod/pod1 created

查看pod，k8scloude2节点就算有污点也能运行pod

[root@k8scloude1 pod]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod1 1/1 Running 0 10s 10.244.112.178 k8scloude2 <none> <none>
[root@k8scloude1 pod]# kubectl delete pod pod1 –force
warning: Immediate deletion does not wait for confirmation that the running resource has been terminated. The resource may continue to run on the cluster indefinitely.
pod \”pod1\” force deleted
[root@k8scloude1 pod]# kubectl get pods -o wide
No resources found in pod namespace.

给k8scloude2节点再添加一个污点

[root@k8scloude1 pod]# kubectl describe nodes k8scloude2 | grep Taints
Taints: wudian=true:NoSchedule
[root@k8scloude1 pod]# kubectl taint node k8scloude2 zang=shide:NoSchedule
node/k8scloude2 tainted
[root@k8scloude1 pod]# kubectl describe nodes k8scloude2 | grep Taints
Taints: wudian=true:NoSchedule
[root@k8scloude1 pod]# kubectl describe nodes k8scloude2 | grep -A2 Taints
Taints: wudian=true:NoSchedule
zang=shide:NoSchedule
Unschedulable: false
[root@k8scloude1 pod]# kubectl describe nodes k8scloude2 | grep -A1 Taints
Taints: wudian=true:NoSchedule
zang=shide:NoSchedule

创建pod，tolerations参数表示容忍2个污点：wudian=true:NoSchedule和zang=shide:NoSchedule，nodeSelector:taint: T参数表示pod运行在标签为nodeSelector=taint: T的节点。

[root@k8scloude1 pod]# vim schedulepod4.yaml
[root@k8scloude1 pod]# cat schedulepod4.yaml
apiVersion: v1
kind: Pod
metadata:
creationTimestamp: null
labels:
run: pod1
name: pod1
namespace: pod
spec:
tolerations:
– key: \”wudian\”
operator: \”Equal\”
value: \”true\”
effect: \”NoSchedule\”
– key: \”zang\”
operator: \”Equal\”
value: \”shide\”
effect: \”NoSchedule\”
nodeSelector:
taint: T
containers:
– image: nginx
imagePullPolicy: IfNotPresent
name: pod1
resources: {}
ports:
– name: http
containerPort: 80
protocol: TCP
hostPort: 80
dnsPolicy: ClusterFirst
restartPolicy: Always
status: {}
[root@k8scloude1 pod]# kubectl apply -f schedulepod4.yaml
pod/pod1 created

查看pod，k8scloude2节点就算有2个污点也能运行pod

[root@k8scloude1 pod]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod1 1/1 Running 0 6s 10.244.112.179 k8scloude2 <none> <none>
[root@k8scloude1 pod]# kubectl delete pod pod1 –force
warning: Immediate deletion does not wait for confirmation that the running resource has been terminated. The resource may continue to run on the cluster indefinitely.
pod \”pod1\” force deleted

创建pod，tolerations参数表示容忍污点：wudian=true:NoSchedule，nodeSelector:taint: T参数表示pod运行在标签为nodeSelector=taint: T的节点。

[root@k8scloude1 pod]# vim schedulepod4.yaml
[root@k8scloude1 pod]# cat schedulepod4.yaml
apiVersion: v1
kind: Pod
metadata:
creationTimestamp: null
labels:
run: pod1
name: pod1
namespace: pod
spec:
tolerations:
– key: \”wudian\”
operator: \”Equal\”
value: \”true\”
effect: \”NoSchedule\”
nodeSelector:
taint: T
containers:
– image: nginx
imagePullPolicy: IfNotPresent
name: pod1
resources: {}
ports:
– name: http
containerPort: 80
protocol: TCP
hostPort: 80
dnsPolicy: ClusterFirst
restartPolicy: Always
status: {}
[root@k8scloude1 pod]# kubectl apply -f schedulepod4.yaml
pod/pod1 created

查看pod，一个节点有两个污点值，但是yaml文件只容忍一个，所以pod创建不成功。

[root@k8scloude1 pod]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod1 0/1 Pending 0 8s <none> <none> <none> <none>
[root@k8scloude1 pod]# kubectl delete pod pod1 –force
warning: Immediate deletion does not wait for confirmation that the running resource has been terminated. The resource may continue to run on the cluster indefinitely.
pod \”pod1\” force deleted
[root@k8scloude1 pod]# kubectl get pods -o wide
No resources found in pod namespace.

取消k8scloude2污点

[root@k8scloude1 pod]# kubectl describe nodes k8scloude2 | grep -A2 Taints
Taints: wudian=true:NoSchedule
zang=shide:NoSchedule
Unschedulable: false
#取消污点
[root@k8scloude1 pod]# kubectl taint node k8scloude2 zang-
node/k8scloude2 untainted
[root@k8scloude1 pod]# kubectl taint node k8scloude2 wudian-
node/k8scloude2 untainted
[root@k8scloude1 pod]# kubectl describe nodes k8scloude1 | grep -A2 Taints
Taints: node-role.kubernetes.io/master:NoSchedule
Unschedulable: false
Lease:
[root@k8scloude1 pod]# kubectl describe nodes k8scloude2 | grep -A2 Taints
Taints: <none>
Unschedulable: false
Lease:
[root@k8scloude1 pod]# kubectl describe nodes k8scloude3 | grep -A2 Taints
Taints: <none>
Unschedulable: false
Lease: