背景：

参见前文：centos8+kubeadm1.20.5+cilium+hubble环境搭建，并降级到了1.21版本（Kubernetes 1.20.5 upgrade 1.21.0）。明天无聊查看一下集群呢忽然发现一个问题：

[root@sh-master-01 ~]# kubectl get pods -n default -o wideNAME                          READY   STATUS             RESTARTS   AGE    IP           NODE         NOMINATED NODE   READINESS GATEScsi-app                       1/1     Running            11         106d   10.0.4.204   sh-work-01   <none>           <none>nginx                         1/1     Running            0          13d    10.0.4.60    sh-work-01   <none>           <none>nginx-1-kkfvd                 1/1     Running            0          13d    10.0.5.223   sh-work-02   <none>           <none>nginx-1-klgpx                 1/1     Running            0          13d    10.0.4.163   sh-work-01   <none>           <none>nginx-1-s5mzp                 1/1     Running            0          13d    10.0.3.208   sh-work-03   <none>           <none>nginx-2-8cb2j                 1/1     Running            0          13d    10.0.3.218   sh-work-03   <none>           <none>nginx-2-l527j                 1/1     Running            0          13d    10.0.5.245   sh-work-02   <none>           <none>nginx-2-qnsrq                 1/1     Running            0          13d    10.0.4.77    sh-work-01   <none>           <none>php-apache-5b95f8f674-clzn5   1/1     Running            2          99d    10.0.3.64    sh-work-03   <none>           <none>pod-flag                      1/1     Running            316        13d    10.0.5.252   sh-work-02   <none>           <none>pod-nodeaffinity              1/1     Running            0          13d    10.0.4.118   sh-work-01   <none>           <none>pod-prefer                    1/1     Running            0          13d    10.0.5.181   sh-work-02   <none>           <none>pod-prefer1                   1/1     Running            0          13d    10.0.3.54    sh-work-03   <none>           <none>with-node-affinity            0/1     ImagePullBackOff   0          13d    10.0.4.126   sh-work-01   <none>           <none>with-pod-affinity             0/1     ImagePullBackOff   0          13d    10.0.5.30    sh-work-02   <none>           <none>with-pod-antiaffinity         1/1     Running            0          13d    10.0.4.159   sh-work-01   <none>           <none>

次要是看ip一栏，这.，我记得我的 serviceSubnet: 172.254.0.0/16 , podSubnet: 172.3.0.0/16啊是不是哪里搞错了了呢？

再看一眼service的网络情况：

[root@sh-master-01 ~]# kubectl get svc -n default -o wideNAME         TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)   AGE    SELECTORkubernetes   ClusterIP   172.254.0.1      <none>        443/TCP   106d   <none>php-apache   ClusterIP   172.254.13.205   <none>        80/TCP    99d    run=php-apache

这个ip段是对的呢！怎么会事件呢？

复盘解决问题

1. 确认问题

cilium是oprator的形式装置的，找到oprator的pod:

kubectl get pods -n kube-system -o wide

查看日志输入：

kubectl logs -f cilium-operator-f595b8f7d-7rvzz  -n kube-system

貌似找到了上面箭头指向的这一段ipv4CIDRs是 10.0.0.0/8!

2. 浏览官网文档

搭建集群我是参考的肥宅爽文的博客：02-02.kubeadm + Cilium 搭建kubernetes集群。博客下面写的比拟残缺的cilium搭建集群的貌似我俩写的了：centos8+kubeadm1.20.5+cilium+hubble环境搭建。初步狐疑就是helm装置cilium的时候配置网络组件并没有走config.yaml中的配置。认真看了一眼官网github仓库文件配置项：https://github.com/cilium/cilium/tree/v1.10.2/install/kubernetes/cilium。

注：我的版本是1.9.7，失常的找本人对应版本的配置文件看呢。差距不大我就没有去切换分支看了...
对照官网文档：在helm装置的时候没有指定ipam.operator.clusterPoolIPv4PodCIDR参数！网上的很多文章也都是草草来的，并没有具体的对ipv4的cidr进行指定那pod网络都是默认的10.0.0.0/8的cidr!我的cvm vpc网络默认是10.0.0.0/8的大网络，这样下来ip地址应该是会有抵触的！

3.找一个正确的参考

谷歌搜寻文档搜到亚马逊的一篇blog:https://aws.amazon.com/cn/blogs/containers/a-multi-cluster-shared-services-architecture-with-amazon-eks-using-cilium-clustermesh/
亚马逊的blog还是很好的。装置的过程很是具体能够参考一下：

至于我的集群只能upgrade了...

 helm upgrade cilium cilium/cilium --version 1.9.7   --namespace=kube-system --set ipam.operator.clusterPoolIPv4PodCIDR="172.3.0.0/16"

目测应该是这样的特别强调一下，抄他人装置的时候还是要去看一下官网文档的具体参数定义！国内能搜到的这些cilium的都大部分没有写podCIDR的配置的吧？参考他人博客的同时肯定记得再认真去看下官网文档参数！