容器编排系统K8s之节点污点和pod容忍度

　　前文我们了解了k8s上的kube-scheduler的工作方式，以及pod调度策略的定义；回顾请参考：https://www.cnblogs.com/qiuhom-1874/p/14243312.html；今天我们来聊一下k8s上的节点污点和pod容忍度相关话题；

　　节点污点是什么呢？

　　节点污点有点类似节点上的标签或注解信息，它们都是用来描述对应节点的元数据信息；污点定义的格式和标签、注解的定义方式很类似，都是用一个kv数据来表示，不同于节点标签，污点的键值数据中包含对应污点的effect，污点的effect是用于描述对应节点上的污点有什么作用；在k8s上污点有三个效用（effect），第一个效用是NoSchedule，表示拒绝pod调度到对应节点上运行；第二个效用是PreferSchedule，表示尽量不把pod调度到此节点上运行；第三个效用是NoExecute，表示拒绝将pod调度到此节点上运行；该效用相比NoSchedule要严苛一点；从上面的描述来看，对应污点就是来描述拒绝pod运行在对应节点的节点属性；

　　pod对节点污点的容忍度

　　从字面意思就能够理解，pod要想运行在对应有污点的节点上，对应pod就要容忍对应节点上的污点；我们把这种容忍节点污点的定义叫做pod对节点污点的容忍度；pod对节点污点的容忍度就是在对应pod中定义怎么去匹配节点污点；通常匹配节点污点的方式有两种，一种是等值匹配，一种是存在性匹配；所谓等值匹配表示对应pod的污点容忍度，必须和节点上的污点属性相等，所谓污点属性是指污点的key、value以及effect；即容忍度必须满足和对应污点的key，value和effect相同，这样表示等值匹配关系，其操作符为Equal；存在性匹配是指对应容忍度只需要匹配污点的key和effect即可，value不纳入匹配标准，即容忍度只要满足和对应污点的key和effect相同就表示对应容忍度和节点污点是存在性匹配，其操作符为Exists；

　　节点污点和pod容忍度的关系

　　提示：如上图所示，只有能够容忍对应节点污点的pod才能够被调度到对应节点运行，不能容忍节点污点的pod是一定不能调度到对应节点上运行（除节点污点为PreferNoSchedule）；

　　节点污点管理

　　给节点添加污点命令使用语法格式

Usage:
  kubectl taint NODE NAME KEY_1=VAL_1:TAINT_EFFECT_1 ... KEY_N=VAL_N:TAINT_EFFECT_N [options]

　　提示：给节点增加污点我们可以用kubectl taint node命令来增加节点污点，只需要指定对应节点名称和污点即可，污点可以指定多个，用空格隔开；

　　示例：给node01添加一个test=test:NoSchedule的污点

[root@master01 ~]# kubectl taint node node01.k8s.org test=test:NoSchedule
node/node01.k8s.org tainted
[root@master01 ~]#

　　查看节点污点

[root@master01 ~]# kubectl describe node node01.k8s.org |grep Taint
Taints:             test=test:NoSchedule
[root@master01 ~]#

　　删除污点

[root@master01 ~]# kubectl describe node node01.k8s.org |grep Taint
Taints:             test=test:NoSchedule
[root@master01 ~]# kubectl taint node node01.k8s.org test:NoSchedule-
node/node01.k8s.org untainted
[root@master01 ~]# kubectl describe node node01.k8s.org |grep Taint  
Taints:             <none>
[root@master01 ~]#

　　提示：删除污点可以指定对应节点上的污点的key和对应污点的effect，也可以直接在对应污点的key后面加“-”，表示删除对应名为对应key的所有污点；

　　pod容忍度定义

　　示例：创建一个pod，其容忍度为对应节点有 node-role.kubernetes.io/master:NoSchedule的污点

[root@master01 ~]# cat pod-demo-taints.yaml
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
  tolerations:
  - key: node-role.kubernetes.io/master
    operator: Exists
    effect: NoSchedule
[root@master01 ~]#

　　提示：定义pod对节点污点的容忍度需要用tolerations字段定义，该字段为一个列表对象；其中key是用来指定对应污点的key，这个key必须和对应节点污点上的key相等；operator字段用于指定对应的操作符，即描述容忍度怎么匹配污点，这个操作符只有两个，Equal和Exists；effect字段用于描述对应的效用，该字段的值通常有三个，NoSchedule、PreferNoSchedule、NoExecute；这个字段的值必须和对应的污点相同；上述清单表示，redis-demo这个pod能够容忍节点上有node-role.kubernetes.io/master:NoSchedule的污点；

　　应用清单

[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo created
[root@master01 ~]# kubectl get pods -o wide
NAME         READY   STATUS    RESTARTS   AGE   IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo   1/1     Running   0          7s    10.244.4.35   node04.k8s.org   <none>           <none>
[root@master01 ~]#

　　提示：可以看到对应pod运行在node04上；这里需要注意，定义pod容忍度只是表示对应pod可以运行在对应有污点的节点上，并非它一定运行在对应节点上；它也可以运行在那些没有污点的节点上；

　　验证：删除pod，给node01,node02,03,04都打上test:NoSchedule的污点，再次应用清单，看看对应pod是否能够正常运行？

[root@master01 ~]# kubectl delete -f pod-demo-taints.yaml
pod "redis-demo" deleted
[root@master01 ~]# kubectl taint node node01.k8s.org test:NoSchedule
node/node01.k8s.org tainted
[root@master01 ~]# kubectl taint node node02.k8s.org test:NoSchedule 
node/node02.k8s.org tainted
[root@master01 ~]# kubectl taint node node03.k8s.org test:NoSchedule 
node/node03.k8s.org tainted
[root@master01 ~]# kubectl taint node node04.k8s.org test:NoSchedule 
node/node04.k8s.org tainted
[root@master01 ~]# kubectl describe node node01.k8s.org |grep Taints
Taints:             test:NoSchedule
[root@master01 ~]# kubectl describe node node02.k8s.org |grep Taints 
Taints:             test:NoSchedule
[root@master01 ~]# kubectl describe node node03.k8s.org |grep Taints 
Taints:             test:NoSchedule
[root@master01 ~]# kubectl describe node node04.k8s.org |grep Taints 
Taints:             test:NoSchedule
[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo created
[root@master01 ~]# kubectl get pods -o wide
NAME         READY   STATUS    RESTARTS   AGE   IP            NODE               NOMINATED NODE   READINESS GATES
redis-demo   1/1     Running   0          18s   10.244.0.14   master01.k8s.org   <none>           <none>
[root@master01 ~]#

　　提示：可以看到对应pod，被调度到master节点上运行了；其原因是对应pod能够容忍master节点上的污点；对应其他node节点上的污点，它并不能容忍，所以只能运行在master节点；

　　删除对应pod中容忍度的定义，再次应用pod清单，看看对应pod是否会正常运行？

[root@master01 ~]# kubectl delete pod redis-demo 
pod "redis-demo" deleted
[root@master01 ~]# cat pod-demo-taints.yaml
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo created
[root@master01 ~]# kubectl get pods -o wide
NAME         READY   STATUS    RESTARTS   AGE   IP       NODE     NOMINATED NODE   READINESS GATES
redis-demo   0/1     Pending   0          6s    <none>   <none>   <none>           <none>
[root@master01 ~]#

　　提示：可以看到对应pod处于pending状态；其原因是对应pod没法容忍对应节点污点；即所有节点都排斥对应pod运行在对应节点上；

　　示例：定义等值匹配关系污点容忍度

[root@master01 ~]# cat pod-demo-taints.yaml
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
  tolerations:
  - key: test
    operator: Equal
    value: test
    effect: NoSchedule

[root@master01 ~]#

　　提示：定义等值匹配关系的容忍度，需要指定对应污点中的value属性；

　　删除原有pod，应用清单

[root@master01 ~]# kubectl delete pod redis-demo
pod "redis-demo" deleted
[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo created
[root@master01 ~]# kubectl get pods -o wide
NAME         READY   STATUS    RESTARTS   AGE   IP       NODE     NOMINATED NODE   READINESS GATES
redis-demo   0/1     Pending   0          4s    <none>   <none>   <none>           <none>
[root@master01 ~]#

　　提示：可以看到应用对应清单以后，pod处于pending状态，其原因是没有满足对应pod容忍度的节点，所以对应pod无法正常调度到节点上运行；

　　验证：修改node01节点的污点为test=test:NoSchedule

[root@master01 ~]# kubectl describe node node01.k8s.org |grep Taints
Taints:             test:NoSchedule
[root@master01 ~]# kubectl taint node node01.k8s.org test=test:NoSchedule --overwrite 
node/node01.k8s.org modified
[root@master01 ~]# kubectl describe node node01.k8s.org |grep Taints                 
Taints:             test=test:NoSchedule
[root@master01 ~]# kubectl get pods -o wide
NAME         READY   STATUS    RESTARTS   AGE     IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo   1/1     Running   0          4m46s   10.244.1.44   node01.k8s.org   <none>           <none>
[root@master01 ~]#

　　提示：可以看到把node01的污点修改为test=test:NoSchedule以后，对应pod就被调度到node01上运行；

　　验证：修改node01节点上的污点为test:NoSchedule，看看对应pod是否被驱离呢？

[root@master01 ~]# kubectl taint node node01.k8s.org test:NoSchedule --overwrite     
node/node01.k8s.org modified
[root@master01 ~]# kubectl describe node node01.k8s.org |grep Taints                 
Taints:             test:NoSchedule
[root@master01 ~]# kubectl get pods -o wide
NAME         READY   STATUS    RESTARTS   AGE     IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo   1/1     Running   0          7m27s   10.244.1.44   node01.k8s.org   <none>           <none>
[root@master01 ~]#

　　提示：可以看到对应节点污点修改为test:NoSchedule以后，对应pod也不会被驱离，说明效用为NoSchedule的污点只是在pod调度时起作用，对于调度完成的pod不起作用；

　　示例：定义pod容忍度为test:PreferNoSchedule

[root@master01 ~]# cat pod-demo-taints.yaml
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo1
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
  tolerations:
  - key: test
    operator: Exists
    effect: PreferNoSchedule

[root@master01 ~]#

　　应用清单

[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo1 created
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE   IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo    1/1     Running   0          11m   10.244.1.44   node01.k8s.org   <none>           <none>
redis-demo1   0/1     Pending   0          6s    <none>        <none>           <none>           <none>
[root@master01 ~]#

　　提示：可以看到对应pod处于pending状态，其原因是没有节点污点是test:PerferNoSchedule，所以对应pod不能被调度运行；

　　给node02节点添加test:PreferNoSchedule污点

[root@master01 ~]# kubectl describe node node02.k8s.org |grep Taints 
Taints:             test:NoSchedule
[root@master01 ~]# kubectl taint node node02.k8s.org test:PreferNoSchedule 
node/node02.k8s.org tainted
[root@master01 ~]# kubectl describe node node02.k8s.org |grep -A 1 Taints
Taints:             test:NoSchedule
                    test:PreferNoSchedule
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE     IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo    1/1     Running   0          18m     10.244.1.44   node01.k8s.org   <none>           <none>
redis-demo1   0/1     Pending   0          6m21s   <none>        <none>           <none>           <none>
[root@master01 ~]#

　　提示：可以看到对应node02上有两个污点，对应pod也没有正常运行起来，其原因是node02上有一个test:NoSchedule污点，对应pod容忍度不能容忍此类污点；

　　验证：修改node01,node03，node04上的节点污点为test:PreferNoSchedule，修改pod的容忍度为test:NoSchedule，再次应用清单，看看对应pod怎么调度

[root@master01 ~]# kubectl taint node node01.k8s.org test:NoSchedule-     
node/node01.k8s.org untainted
[root@master01 ~]# kubectl taint node node03.k8s.org test:NoSchedule- 
node/node03.k8s.org untainted
[root@master01 ~]# kubectl taint node node04.k8s.org test:NoSchedule- 
node/node04.k8s.org untainted
[root@master01 ~]# kubectl taint node node01.k8s.org test:PreferNoSchedule
node/node01.k8s.org tainted
[root@master01 ~]# kubectl taint node node03.k8s.org test:PreferNoSchedule  
node/node03.k8s.org tainted
[root@master01 ~]# kubectl taint node node04.k8s.org test:PreferNoSchedule 
node/node04.k8s.org tainted
[root@master01 ~]# kubectl describe node node01.k8s.org |grep -A 1 Taints 
Taints:             test:PreferNoSchedule
Unschedulable:      false
[root@master01 ~]# kubectl describe node node02.k8s.org |grep -A 1 Taints 
Taints:             test:NoSchedule
                    test:PreferNoSchedule
[root@master01 ~]# kubectl describe node node03.k8s.org |grep -A 1 Taints 
Taints:             test:PreferNoSchedule
Unschedulable:      false
[root@master01 ~]# kubectl describe node node04.k8s.org |grep -A 1 Taints 
Taints:             test:PreferNoSchedule
Unschedulable:      false
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE   IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo    1/1     Running   0          31m   10.244.1.44   node01.k8s.org   <none>           <none>
redis-demo1   1/1     Running   0          19m   10.244.1.45   node01.k8s.org   <none>           <none>
[root@master01 ~]# kubectl delete pod --all
pod "redis-demo" deleted
pod "redis-demo1" deleted
[root@master01 ~]# cat pod-demo-taints.yaml 
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo1
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
  tolerations:
  - key: test
    operator: Exists
    effect: NoSchedule

[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo1 created
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE   IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo1   1/1     Running   0          5s    10.244.4.36   node04.k8s.org   <none>           <none>
[root@master01 ~]#

　　提示：从上面的验证过程来看，当我们把node01，node03，node04节点上的污点删除以后，刚才创建的redis-demo1pod被调度到node01上运行了；其原因是node01上的污点第一个被删除；但我们把pod的容忍对修改成test:NoSchedule以后，再次应用清单，对应pod被调度到node04上运行；这意味着NoSchedule效用污点容忍度是可以正常容忍PreferNoSchedule污点；

　　示例：定义pod容忍度为test:NoExecute

[root@master01 ~]# cat pod-demo-taints.yaml
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo2
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
  tolerations:
  - key: test
    operator: Exists
    effect: NoExecute
[root@master01 ~]#

　　应用清单

[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo2 created
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE   IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo1   1/1     Running   0          35m   10.244.4.36   node04.k8s.org   <none>           <none>
redis-demo2   1/1     Running   0          5s    10.244.4.38   node04.k8s.org   <none>           <none>
[root@master01 ~]#

　　提示：可以看到对应pod被调度到node04上运行，说明容忍效用为NoExecute能够容忍污点效用为PreferNoSchedule的节点；

　　验证：更改所有node节点污点为test:NoSchedule，删除原有pod，再次应用清单，看看对应pod是否还会正常运行？

[root@master01 ~]# kubectl taint node node01.k8s.org test-
node/node01.k8s.org untainted
[root@master01 ~]# kubectl taint node node02.k8s.org test- 
node/node02.k8s.org untainted
[root@master01 ~]# kubectl taint node node03.k8s.org test- 
node/node03.k8s.org untainted
[root@master01 ~]# kubectl taint node node04.k8s.org test- 
node/node04.k8s.org untainted
[root@master01 ~]# kubectl taint node node01.k8s.org test:NoSchedule
node/node01.k8s.org tainted
[root@master01 ~]# kubectl taint node node02.k8s.org test:NoSchedule 
node/node02.k8s.org tainted
[root@master01 ~]# kubectl taint node node03.k8s.org test:NoSchedule 
node/node03.k8s.org tainted
[root@master01 ~]# kubectl taint node node04.k8s.org test:NoSchedule 
node/node04.k8s.org tainted
[root@master01 ~]# kubectl describe node node01.k8s.org |grep -A 1 Taints
Taints:             test:NoSchedule
Unschedulable:      false
[root@master01 ~]# kubectl describe node node02.k8s.org |grep -A 1 Taints 
Taints:             test:NoSchedule
Unschedulable:      false
[root@master01 ~]# kubectl describe node node03.k8s.org |grep -A 1 Taints 
Taints:             test:NoSchedule
Unschedulable:      false
[root@master01 ~]# kubectl describe node node04.k8s.org |grep -A 1 Taints 
Taints:             test:NoSchedule
Unschedulable:      false
[root@master01 ~]# kubectl delete pod --all
pod "redis-demo1" deleted
pod "redis-demo2" deleted
[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo2 created
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE   IP       NODE     NOMINATED NODE   READINESS GATES
redis-demo2   0/1     Pending   0          6s    <none>   <none>   <none>           <none>
[root@master01 ~]#

　　提示：可以看到对应pod处于pending状态，说明pod容忍效用为NoExecute，并不能容忍污点效用为NoSchedule；

　　删除pod，修改所有节点污点为test:NoExecute，把pod容忍度修改为NoScheudle，然后应用清单，看看对应pod怎么调度

[root@master01 ~]# kubectl delete pod --all
pod "redis-demo2" deleted
[root@master01 ~]# kubectl taint node node01.k8s.org test-               
node/node01.k8s.org untainted
[root@master01 ~]# kubectl taint node node02.k8s.org test- 
node/node02.k8s.org untainted
[root@master01 ~]# kubectl taint node node03.k8s.org test- 
node/node03.k8s.org untainted
[root@master01 ~]# kubectl taint node node04.k8s.org test- 
node/node04.k8s.org untainted
[root@master01 ~]# kubectl taint node node01.k8s.org test:NoExecute
node/node01.k8s.org tainted
[root@master01 ~]# kubectl taint node node02.k8s.org test:NoExecute 
node/node02.k8s.org tainted
[root@master01 ~]# kubectl taint node node03.k8s.org test:NoExecute 
node/node03.k8s.org tainted
[root@master01 ~]# kubectl taint node node04.k8s.org test:NoExecute 
node/node04.k8s.org tainted
[root@master01 ~]# kubectl describe node node01.k8s.org |grep -A 1 Taints
Taints:             test:NoExecute
Unschedulable:      false
[root@master01 ~]# kubectl describe node node02.k8s.org |grep -A 1 Taints 
Taints:             test:NoExecute
Unschedulable:      false
[root@master01 ~]# kubectl describe node node03.k8s.org |grep -A 1 Taints 
Taints:             test:NoExecute
Unschedulable:      false
[root@master01 ~]# kubectl describe node node04.k8s.org |grep -A 1 Taints 
Taints:             test:NoExecute
Unschedulable:      false
[root@master01 ~]# cat pod-demo-taints.yaml 
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo2
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
  tolerations:
  - key: test
    operator: Exists
    effect: NoSchedule
[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo2 created
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE   IP       NODE     NOMINATED NODE   READINESS GATES
redis-demo2   0/1     Pending   0          8s    <none>   <none>   <none>           <none>
[root@master01 ~]#

　　提示：从上面的演示来看，pod容忍度效用为NoSchedule也不能容忍污点效用为NoExecute；

　　删除pod，修改对应pod的容忍度为test:NoExecute

[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE    IP       NODE     NOMINATED NODE   READINESS GATES
redis-demo2   0/1     Pending   0          5m5s   <none>   <none>   <none>           <none>
[root@master01 ~]# kubectl delete pod --all
pod "redis-demo2" deleted
[root@master01 ~]# cat pod-demo-taints.yaml
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo2
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
  tolerations:
  - key: test
    operator: Exists
    effect: NoExecute
[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo2 created
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE   IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo2   1/1     Running   0          6s    10.244.4.43   node04.k8s.org   <none>           <none>
[root@master01 ~]#

　　修改node04节点污点为test:NoSchedule，看看对应pod是否可以正常运行？

[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE     IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo2   1/1     Running   0          4m38s   10.244.4.43   node04.k8s.org   <none>           <none>
[root@master01 ~]# kubectl taint node node04.k8s.org test-
node/node04.k8s.org untainted
[root@master01 ~]# kubectl get pods -o wide               
NAME          READY   STATUS    RESTARTS   AGE    IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo2   1/1     Running   0          8m2s   10.244.4.43   node04.k8s.org   <none>           <none>
[root@master01 ~]# kubectl taint node node04.k8s.org test:NoSchedule
node/node04.k8s.org tainted
[root@master01 ~]# kubectl describe node node04.k8s.org |grep -A 1 Taints
Taints:             test:NoSchedule
Unschedulable:      false
[root@master01 ~]# kubectl get pods -o wide                              
NAME          READY   STATUS    RESTARTS   AGE     IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo2   1/1     Running   0          8m25s   10.244.4.43   node04.k8s.org   <none>           <none>
[root@master01 ~]#

　　提示：从NoExecute更改为NoSchedule，对原有pod不会进行驱离；

　　修改pod的容忍度为test:NoSchedule，再次应用清单

[root@master01 ~]# cat pod-demo-taints.yaml
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo3
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
  tolerations:
  - key: test
    operator: Exists
    effect: NoSchedule
---
apiVersion: v1
kind: Pod
metadata:
  name: redis-demo4
  labels:
    app: db
spec:
  containers:
  - name: redis
    image: redis:4-alpine
    ports:
    - name: redis
      containerPort: 6379
  tolerations:
  - key: test
    operator: Exists
    effect: NoSchedule
[root@master01 ~]# kubectl apply -f pod-demo-taints.yaml
pod/redis-demo3 created
pod/redis-demo4 created
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE   IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo2   1/1     Running   0          14m   10.244.4.43   node04.k8s.org   <none>           <none>
redis-demo3   1/1     Running   0          4s    10.244.4.45   node04.k8s.org   <none>           <none>
redis-demo4   1/1     Running   0          4s    10.244.4.46   node04.k8s.org   <none>           <none>
[root@master01 ~]#

　　提示：可以看到后面两个pod都被调度node04上运行；其原因是对应pod的容忍度test:NoSchedule只能容忍node04上的污点test:NoSchedule；

　　修改node04的污点为NoExecute，看看对应pod是否会被驱离？

[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE     IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo2   1/1     Running   0          17m     10.244.4.43   node04.k8s.org   <none>           <none>
redis-demo3   1/1     Running   0          2m32s   10.244.4.45   node04.k8s.org   <none>           <none>
redis-demo4   1/1     Running   0          2m32s   10.244.4.46   node04.k8s.org   <none>           <none>
[root@master01 ~]# kubectl describe node node04.k8s.org |grep -A 1 Taints
Taints:             test:NoSchedule
Unschedulable:      false
[root@master01 ~]# kubectl taint node node04.k8s.org test-
node/node04.k8s.org untainted
[root@master01 ~]# kubectl taint node node04.k8s.org test:NoExecute
node/node04.k8s.org tainted
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS        RESTARTS   AGE     IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo2   1/1     Running       0          18m     10.244.4.43   node04.k8s.org   <none>           <none>
redis-demo3   0/1     Terminating   0          3m43s   10.244.4.45   node04.k8s.org   <none>           <none>
redis-demo4   0/1     Terminating   0          3m43s   10.244.4.46   node04.k8s.org   <none>           <none>
[root@master01 ~]# kubectl get pods -o wide
NAME          READY   STATUS    RESTARTS   AGE   IP            NODE             NOMINATED NODE   READINESS GATES
redis-demo2   1/1     Running   0          18m   10.244.4.43   node04.k8s.org   <none>           <none>
[root@master01 ~]#

　　提示：可以看到修改node04的污点为test:NoExecute以后，对应pod容忍污点效用为不是NoExecute的pod被驱离了；说明污点效用为NoExecute，它会驱离不能容忍该污点效用的所有pod；

　　创建一个deploy，其指定容器的容忍度为test:NoExecute，并指定其驱离延迟施加为10秒

[root@master01 ~]# cat deploy-demo-taint.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-demo
spec:
  replicas: 3
  selector:
     matchLabels:
      app: redis
  template:
    metadata:
      labels:
        app: redis
    spec:
      containers:
      - name: redis
        image: redis:4-alpine
        ports:
        - name: redis
          containerPort: 6379
      tolerations:
      - key: test
        operator: Exists
        effect: NoExecute
        tolerationSeconds: 10
   
[root@master01 ~]#

　　提示：tolerationSeconds字段用于指定其驱离宽限其时长；该字段只能用在其容忍污点效用为NoExecute的容忍度中使用；其他污点效用不能使用该字段来指定其容忍宽限时长；

　　应用配置清单

[root@master01 ~]# kubectl apply -f deploy-demo-taint.yaml
deployment.apps/deploy-demo created
[root@master01 ~]# kubectl get pods -o wide -w
NAME                           READY   STATUS    RESTARTS   AGE   IP            NODE             NOMINATED NODE   READINESS GATES
deploy-demo-79b89f9847-9zk8j   1/1     Running   0          7s    10.244.2.71   node02.k8s.org   <none>           <none>
deploy-demo-79b89f9847-h8zlc   1/1     Running   0          7s    10.244.3.61   node03.k8s.org   <none>           <none>
deploy-demo-79b89f9847-shscr   1/1     Running   0          7s    10.244.1.62   node01.k8s.org   <none>           <none>
redis-demo2                    1/1     Running   0          54m   10.244.4.43   node04.k8s.org   <none>           <none>
deploy-demo-79b89f9847-h8zlc   1/1     Terminating   0          10s   10.244.3.61   node03.k8s.org   <none>           <none>
deploy-demo-79b89f9847-shscr   1/1     Terminating   0          10s   10.244.1.62   node01.k8s.org   <none>           <none>
deploy-demo-79b89f9847-2x8w6   0/1     Pending       0          0s    <none>        <none>           <none>           <none>
deploy-demo-79b89f9847-2x8w6   0/1     Pending       0          0s    <none>        node03.k8s.org   <none>           <none>
deploy-demo-79b89f9847-lhltv   0/1     Pending       0          0s    <none>        <none>           <none>           <none>
deploy-demo-79b89f9847-9zk8j   1/1     Terminating   0          10s   10.244.2.71   node02.k8s.org   <none>           <none>
deploy-demo-79b89f9847-2x8w6   0/1     ContainerCreating   0          0s    <none>        node03.k8s.org   <none>           <none>
deploy-demo-79b89f9847-lhltv   0/1     Pending             0          0s    <none>        node02.k8s.org   <none>           <none>
deploy-demo-79b89f9847-lhltv   0/1     ContainerCreating   0          0s    <none>        node02.k8s.org   <none>           <none>
deploy-demo-79b89f9847-w8xjw   0/1     Pending             0          0s    <none>        <none>           <none>           <none>
deploy-demo-79b89f9847-w8xjw   0/1     Pending             0          0s    <none>        node01.k8s.org   <none>           <none>
deploy-demo-79b89f9847-w8xjw   0/1     ContainerCreating   0          0s    <none>        node01.k8s.org   <none>           <none>
deploy-demo-79b89f9847-shscr   1/1     Terminating         0          10s   10.244.1.62   node01.k8s.org   <none>           <none>
deploy-demo-79b89f9847-h8zlc   1/1     Terminating         0          10s   10.244.3.61   node03.k8s.org   <none>           <none>
deploy-demo-79b89f9847-9zk8j   1/1     Terminating         0          10s   10.244.2.71   node02.k8s.org   <none>           <none>
deploy-demo-79b89f9847-shscr   0/1     Terminating         0          11s   10.244.1.62   node01.k8s.org   <none>           <none>
deploy-demo-79b89f9847-2x8w6   0/1     ContainerCreating   0          1s    <none>        node03.k8s.org   <none>           <none>
deploy-demo-79b89f9847-lhltv   0/1     ContainerCreating   0          1s    <none>        node02.k8s.org   <none>           <none>
deploy-demo-79b89f9847-w8xjw   0/1     ContainerCreating   0          1s    <none>        node01.k8s.org   <none>           <none>
deploy-demo-79b89f9847-h8zlc   0/1     Terminating         0          11s   10.244.3.61   node03.k8s.org   <none>           <none>
deploy-demo-79b89f9847-2x8w6   1/1     Running             0          1s    10.244.3.62   node03.k8s.org   <none>           <none>
deploy-demo-79b89f9847-9zk8j   0/1     Terminating         0          11s   10.244.2.71   node02.k8s.org   <none>           <none>
deploy-demo-79b89f9847-lhltv   1/1     Running             0          1s    10.244.2.72   node02.k8s.org   <none>           <none>
deploy-demo-79b89f9847-w8xjw   1/1     Running             0          2s    10.244.1.63   node01.k8s.org   <none>           <none>
deploy-demo-79b89f9847-h8zlc   0/1     Terminating         0          15s   10.244.3.61   node03.k8s.org   <none>           <none>
deploy-demo-79b89f9847-h8zlc   0/1     Terminating         0          15s   10.244.3.61   node03.k8s.org   <none>           <none>
^C[root@master01 ~]#

　　提示：可以看到对应pod只能在对应节点上运行10秒，随后就被驱离，因为我们创建的是一个deploy，对应pod被驱离以后，对应deploy又会重建；

　　总结：对于污点效用为NoSchedule来说，它只会拒绝新建的pod，不会对原有pod进行驱离；如果对应pod能够容忍该污点，则对应pod就有可能运行在对应节点上；如果不能容忍，则对应pod一定不会调度到对应节点运行；对于污点效用为PreferNoSchedule来说，它也不会驱离已存在pod，它只有在所有节点都不满足对应pod容忍度时，对应pod可以勉强运行在此类污点效用的节点上；对于污点效用为NoExecute来说，默认不指定其容忍宽限时长，表示能够一直容忍，如果指定了其宽限时长，则到了宽限时长对应pod将会被驱离；对应之前被调度到该节点上的pod，在节点污点效用变为NoExecute后，该节点会立即驱离所有不能容忍污点效用为NoExecute的pod；

posted @ 2021-01-09 21:14 Linux-1874 阅读(2718) 评论(0) 编辑收藏举报

刷新页面返回顶部

1874

容器编排系统K8s之节点污点和pod容忍度

公告