K8S自己动手系列 - 1.3 - Taint & Affinity

  1. 云栖社区>
  2. 博客>
  3. 正文

K8S自己动手系列 - 1.3 - Taint & Affinity

荣滨 2019-06-17 20:13:41 浏览2002
展开阅读全文

Taint

please refer https://kubernetes.io/docs/concepts/configuration/taint-and-toleration/
三种taint:

  • NoSchedule
  • PreferNoSchedule
  • NoExecute

为节点打Taint

  ~ kubectl taint node worker02 role=nginx:NoSchedule
node/worker02 tainted
  ~ kubectl describe node worker02
Name:               worker02
...
Taints:             role=nginx:NoSchedule

  ~ kubectl get pod -o wide
NAME                     READY   STATUS    RESTARTS   AGE     IP            NODE       NOMINATED NODE   READINESS GATES
nginx-7cffb9df96-748j6   1/1     Running   0          5m21s   10.244.0.23   worker01   <none>           <none>
nginx-7cffb9df96-d2rt5   1/1     Running   0          5m22s   10.244.1.35   worker02   <none>           <none>

通过查看pod,发现已经运行的pod并未受影响
扩容pod,看下效果

  ~ kubectl scale --replicas=5 deployment/nginx
deployment.extensions/nginx scaled
  ~ kubectl get pod -o wide
NAME                     READY   STATUS              RESTARTS   AGE    IP            NODE       NOMINATED NODE   READINESS GATES
nginx-7cffb9df96-4rjhf   0/1     ContainerCreating   0          1s     <none>        worker01   <none>           <none>
nginx-7cffb9df96-748j6   1/1     Running             0          6m8s   10.244.0.23   worker01   <none>           <none>
nginx-7cffb9df96-89ngr   0/1     ContainerCreating   0          1s     <none>        worker01   <none>           <none>
nginx-7cffb9df96-d2rt5   1/1     Running             0          6m9s   10.244.1.35   worker02   <none>           <none>
nginx-7cffb9df96-zsgmd   0/1     ContainerCreating   0          1s     <none>        worker01   <none>           <none>

通过实验发现由于taint NoSchedule的影响,新pod不会调度到含有污点的节点上

移除节点Taint

  ~ kubectl taint node worker02 role:NoSchedule-
node/worker02 untainted

再次扩容,查看效果

  ~ kubectl scale --replicas=10 deployment/nginx
deployment.extensions/nginx scaled
  ~ kubectl get pod -o wide
NAME                     READY   STATUS    RESTARTS   AGE     IP            NODE       NOMINATED NODE   READINESS GATES
nginx-7cffb9df96-2wlln   1/1     Running   0          4s      10.244.1.38   worker02   <none>           <none>
nginx-7cffb9df96-4rjhf   1/1     Running   0          3m10s   10.244.0.26   worker01   <none>           <none>
nginx-7cffb9df96-748j6   1/1     Running   0          9m17s   10.244.0.23   worker01   <none>           <none>
nginx-7cffb9df96-89ngr   1/1     Running   0          3m10s   10.244.0.24   worker01   <none>           <none>
nginx-7cffb9df96-9vbdt   1/1     Running   0          4s      10.244.1.36   worker02   <none>           <none>
nginx-7cffb9df96-d2rt5   1/1     Running   0          9m18s   10.244.1.35   worker02   <none>           <none>
nginx-7cffb9df96-hqvfh   1/1     Running   0          4s      10.244.1.37   worker02   <none>           <none>
nginx-7cffb9df96-mhxmn   1/1     Running   0          4s      10.244.1.40   worker02   <none>           <none>
nginx-7cffb9df96-xdnzc   1/1     Running   0          4s      10.244.1.39   worker02   <none>           <none>
nginx-7cffb9df96-zsgmd   1/1     Running   0          3m10s   10.244.0.25   worker01   <none>           <none>

发现新增加的pod已经可以调度到worker02上了

容忍节点Taint

重新为节点增加污点

  ~ kubectl taint node worker02 role=nginx:NoSchedule
node/worker02 tainted

然后在deploy定义中增加:

tolerations:
- key: "role"
  operator: "Equal"
  value: "nginx"
  effect: "NoSchedule"

通过kubectl diff -f xxx.yaml可以查看变化情况

  lab02 kubectl diff -f nginx-deploy.yaml 
diff -u -N /tmp/LIVE-021075426/extensions.v1beta1.Deployment.default.nginx /tmp/MERGED-503027673/extensions.v1beta1.Deployment.default.nginx
--- /tmp/LIVE-021075426/extensions.v1beta1.Deployment.default.nginx    2019-06-09 15:53:54.120907334 +0800
+++ /tmp/MERGED-503027673/extensions.v1beta1.Deployment.default.nginx    2019-06-09 15:53:54.128907426 +0800
@@ -4,7 +4,7 @@
   annotations:
     deployment.kubernetes.io/revision: "8"
   creationTimestamp: "2019-06-08T04:27:19Z"
-  generation: 23
+  generation: 24
   labels:
     app: nginx
   name: nginx
@@ -14,7 +14,7 @@
   uid: b43106be-89a5-11e9-8ec2-080027a62701
 spec:
   progressDeadlineSeconds: 2147483647
-  replicas: 1
+  replicas: 2
   revisionHistoryLimit: 2147483647
   selector:
     matchLabels:
@@ -42,6 +42,11 @@
       schedulerName: default-scheduler
       securityContext: {}
       terminationGracePeriodSeconds: 30
+      tolerations:
+      - effect: NoSchedule
+        key: role
+        operator: Equal
+        value: nginx
 status:
   availableReplicas: 1
   conditions:
exit status 1

再次扩容,查看效果

  lab02 kubectl get pods -o wide
NAME                     READY   STATUS    RESTARTS   AGE   IP            NODE       NOMINATED NODE   READINESS GATES
nginx-859959cc7f-7sx2p   1/1     Running   0          4s    10.244.0.38   worker01   <none>           <none>
nginx-859959cc7f-8tpw7   1/1     Running   0          4s    10.244.0.39   worker01   <none>           <none>
nginx-859959cc7f-ffh7v   1/1     Running   0          4s    10.244.1.49   worker02   <none>           <none>
nginx-859959cc7f-j2xp7   1/1     Running   0          4s    10.244.0.37   worker01   <none>           <none>
nginx-859959cc7f-lht8b   1/1     Running   0          4s    10.244.1.47   worker02   <none>           <none>
nginx-859959cc7f-mcth6   1/1     Running   0          17s   10.244.0.36   worker01   <none>           <none>
nginx-859959cc7f-sx5ln   1/1     Running   0          4s    10.244.1.48   worker02   <none>           <none>
nginx-859959cc7f-tlxk5   1/1     Running   0          4s    10.244.1.46   worker02   <none>           <none>
nginx-859959cc7f-xkjtq   1/1     Running   0          4s    10.244.1.45   worker02   <none>           <none>
nginx-859959cc7f-xltkz   1/1     Running   0          17s   10.244.1.44   worker02   <none>           <none>

Affinity

上面通过节点污点的方式,我们可以避免pod被调度到含有污点的节点,同时也可以使某些pod容忍节点的污点,配合下面要说的affinity,可以实现特定节点为特定pod专用的目的

NodeSelector

在pod定义中增加如下:

      nodeSelector:
        kubernetes.io/hostname: worker02

执行查看效果

  lab02 kubectl apply -f nginx-deploy-toleration-nodeselector.yaml
deployment.extensions/nginx configured
  lab02 kubectl get pod -o wide
NAME                     READY   STATUS    RESTARTS   AGE   IP            NODE       NOMINATED NODE   READINESS GATES
nginx-564c745dcd-6dtz7   1/1     Running   0          25s   10.244.1.56   worker02   <none>           <none>
nginx-564c745dcd-fw8kt   1/1     Running   0          25s   10.244.1.55   worker02   <none>           <none>
nginx-564c745dcd-l8k7m   1/1     Running   0          22s   10.244.1.58   worker02   <none>           <none>
nginx-564c745dcd-mvqjd   1/1     Running   0          28s   10.244.1.53   worker02   <none>           <none>
nginx-564c745dcd-qn6r9   1/1     Running   0          29s   10.244.1.52   worker02   <none>           <none>
nginx-564c745dcd-rbl57   1/1     Running   0          27s   10.244.1.54   worker02   <none>           <none>
nginx-564c745dcd-s2jbm   1/1     Running   0          19s   10.244.1.59   worker02   <none>           <none>
nginx-564c745dcd-wnd45   1/1     Running   0          30s   10.244.1.50   worker02   <none>           <none>
nginx-564c745dcd-zlnfg   1/1     Running   0          30s   10.244.1.51   worker02   <none>           <none>
nginx-564c745dcd-zm7qx   1/1     Running   0          22s   10.244.1.57   worker02   <none>           <none>

NodeAffinity

NodeAffinity与NodeSelector有相似的作用,但是相比之下有更加灵活的表达语义
please refer: https://kubernetes.io/docs/concepts/configuration/assign-pod-node/#affinity-and-anti-affinity

网友评论

登录后评论
0/500
评论
荣滨
+ 关注