K8S自己动手系列 - 1.3 - Taint & Affinity-低调大师

K8S自己动手系列 - 1.3 - Taint & Affinity

2019-06-16 470

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

微信关注我们

原文链接：https://yq.aliyun.com/articles/705679

转载内容版权归作者及来源网站所有！

低调大师中文资讯倾力打造互联网数据资讯、行业资源、电子商务、移动互联网、网络营销平台。持续更新报道IT业界、互联网、市场资讯、驱动更新,是最及时权威的产业资讯及硬件资讯报道平台。

云上快速搭建Serverless AI实验室

Serverless Kubernetes和ACK虚拟节点都已基于ECI提供GPU容器实例功能，让用户在云上低成本快速搭建serverless AI实验室，用户无需维护服务器和GPU基础运行环境，极大降低AI平台运维的负担，显著提升整体计算效率。如何使用GPU容器实例在pod的annotation中指定所需GPU的类型（P4/P100/V100等），同时在resource.limits中指定GPU的个数即可创建GPU容器实例。每个pod独占GPU，暂不支持vGPU，GPU实例的收费与ECS GPU类型收费一致，不产生额外费用，目前ECI提供多种规格的GPU类型。（请参考https://help.aliyun.com/document_detail/114581.html）示例 1. 创建Serverless Kubernetes集群

2019-06-17

644

说在最前 Pod作为Kubernetes的一等公民，是K8S调度的最小单元，也是使用Kubernetes的人打交道最多的K8S模型。既然Pod这么重要，搞清楚Pod的各种功能及操作就非常重要了，理解了Pod以后，各种Pod的控制器组件，例如：Deployment，StatefulSet，DaemonSet， Job，ReplicaSet理解起来才能事半功倍。正篇阅读官方文档对Pod的描述，其阐述了Pod在集群中的两种主要使用方式：一个Pod运行一个容器一个Pod运行多个需要在一起的容器由于前面的实验，我们已经使用过Deployment部署简单的nginx容器，这种就属于第一种，一个Pod运行一个容器的模型，所以本篇文章我们来实战一下，一个Pod运行多个相互关联的容器。实战场景说明本文以常用blog应用wordpress+mysql为例，将wordpress与mysql运行在同一个Pod内，并实现同Pod的相互访问。本文实验所有的源码保存在：https://github.com/zrbcool/blog-public/tree/master/k8s-hands-on/l...

2019-06-17

555

资源下载

更多资源

优质分享App

近一个月的开发和优化，本站点的第一个app全新上线。该app采用极致压缩，本体才4.36MB。系统里面做了大量数据访问、缓存优化。方便用户在手机上查看文章。后续会推出HarmonyOS的适配版本。

Nacos

Nacos /nɑ:kəʊs/ 是 Dynamic Naming and Configuration Service 的首字母简称，一个易于构建 AI Agent 应用的动态服务发现、配置管理和AI智能体管理平台。Nacos 致力于帮助您发现、配置和管理微服务及AI智能体应用。Nacos 提供了一组简单易用的特性集，帮助您快速实现动态服务发现、服务配置、服务元数据、流量管理。Nacos 帮助您更敏捷和容易地构建、交付和管理微服务平台。

Spring

Spring框架（Spring Framework）是由Rod Johnson于2002年提出的开源Java企业级应用框架，旨在通过使用JavaBean替代传统EJB实现方式降低企业级编程开发的复杂性。该框架基于简单性、可测试性和松耦合性设计理念，提供核心容器、应用上下文、数据访问集成等模块，支持整合Hibernate、Struts等第三方框架，其适用范围不仅限于服务器端开发，绝大多数Java应用均可从中受益。

Rocky Linux

Rocky Linux（中文名：洛基）是由Gregory Kurtzer于2020年12月发起的企业级Linux发行版，作为CentOS稳定版停止维护后与RHEL（Red Hat Enterprise Linux）完全兼容的开源替代方案，由社区拥有并管理，支持x86_64、aarch64等架构。其通过重新编译RHEL源代码提供长期稳定性，采用模块化包装和SELinux安全架构，默认包含GNOME桌面环境及XFS文件系统，支持十年生命周期更新。