Expiration time of each certificate
/etc/kubernetes/pki/apiserver.crt #1年有效期
/etc/kubernetes/pki/front-proxy-ca.crt #10年有效期
/etc/kubernetes/pki/ca.crt #10年有效期
/etc/kubernetes/pki/apiserver-etcd-client.crt #1年有效期
/etc/kubernetes/pki/front-proxy-client.crt #1年有效期
/etc/kubernetes/pki/etcd/server.crt #1年有效期
/etc/kubernetes/pki/etcd/ca.crt #10年有效期
/etc/kubernetes/pki/etcd/peer.crt #1年有效期
/etc/kubernetes/pki/etcd/healthcheck-client.crt #1年有效期
/etc/kubernetes/pki/apiserver-kubelet-client.crt #1年有效期
Can be used to check the expiration time of the certificate:
method 1
kubeadm alpha certs check-expiration
Method 2
openssl x509 -noout -dates -in /etc/kubernetes/pki/apiserver.crt
kubeadm alpha certs 命令详解:
Available Commands:
certificate-key 生成证书和key
check-expiration 检测证书过期时间
renew 续订Kubernetes集群的证书
kubeadm alpha certs命令仅支持v1.15及其以上的版本。
kubeadm alpha certs check-expiration
This command displays the expiration time/remaining time of the client certificate in the /etc/kubernetes/pki folder and the client certificate embedded in the KUBECONFIG file used by kubeadm.
If you are running a highly available cluster, this command needs to be executed on all control panel nodes.
Next, let's update our cluster certificate. The following operations are performed on the master node. First, back up the original certificate:
mkdir /etc/kubernetes.bak
cp -r /etc/kubernetes/pki/ /etc/kubernetes.bak
cp /etc/kubernetes/*.conf /etc/kubernetes.bak
然后备份 etcd 数据目录:
cp -r /var/lib/etcd /var/lib/etcd.bak
Update certificate with Kubernetes certificate API
In addition to the one-click manual certificate update mentioned above, you can also use the Kubernetes certificate API to perform manual certificate update. For the online environment, we may not take the risk of frequently updating the cluster or updating the certificate. These are risky after all. Therefore, we hope that the validity period of the generated certificate is long enough. Although this is not recommended from a security perspective, it is not recommended. In some scenarios, a sufficiently long certificate validity period is also very necessary. Many administrators manually change the source code of kubeadm to 10 years and then recompile to create a cluster. Although this method can achieve the purpose, it is not recommended to use this method, especially when you want to update the cluster. It must be updated with the new version. In fact, Kubernetes provides an API method to help us generate a long enough certificate validity period.
To use the built-in API method to sign, first we need to configure the --experimental-cluster-signing-duration parameter of the kube-controller-manager component and adjust it to 10 years. Here is the cluster installed by kubeadm, so we directly modify the static The Pod's yaml file can be:
vim /etc/kubernetes/manifests/kube-controller-manager.yaml
......
spec:
containers:
- command:
- kube-controller-manager
#设置证书有效期为 10 年
- --experimental-cluster-signing-duration=87600h
- --client-ca-file=/etc/kubernetes/pki/ca.crt
......
After the modification is completed, kube-controller-manager will automatically restart to take effect. Then we need to use the following command to create a certificate signing request for the Kubernetes certificate API. If you set up an external signer such as cert-manager, certificate signing requests (CSRs) will be automatically approved. Otherwise, you must manually approve the certificate using the kubectl certificate command. The following kubeadm command outputs the name of the certificate to be approved, and then waits for approval to occur:
kubeadm alpha certs renew all --use-api --config kubeadm.yaml &
The output is similar to the following:
[1] 2890
[certs] Certificate request "kubeadm-cert-kubernetes-admin-pn99f" created
#然后接下来我们需要去手动批准证书:
kubectl get csr
NAME AGE REQUESTOR CONDITION
kubeadm-cert-kubernetes-admin-pn99f 64s kubernetes-admin Pending
Manually approve the certificate
kubectl certificate approve kubeadm-cert-kubernetes-admin-pn99f
certificatesigningrequest.certificates.k8s.io/kubeadm-cert-kubernetes-admin-pn99f approved
用同样的方式为处于 Pending 状态的 csr 执行批准操作,直到所有的 csr 都批准完成为止。最后所有的 csr 列表状态如下所示:
kubectl get csr
NAME AGE REQUESTOR CONDITION
kubeadm-cert-front-proxy-client-llhrj 30s kubernetes-admin Approved,Issued
kubeadm-cert-kube-apiserver-2s6kf 2m43s kubernetes-admin Approved,Issued
kubeadm-cert-kube-apiserver-etcd-client-t9pkx 2m7s kubernetes-admin Approved,Issued
kubeadm-cert-kube-apiserver-kubelet-client-pjbjm 108s kubernetes-admin Approved,Issued
kubeadm-cert-kube-etcd-healthcheck-client-8dcn8 64s kubernetes-admin Approved,Issued
kubeadm-cert-kubernetes-admin-pn99f 4m29s kubernetes-admin Approved,Issued
kubeadm-cert-system:kube-controller-manager-mr86h 79s kubernetes-admin Approved,Issued
kubeadm-cert-system:kube-scheduler-t8lnw 17s kubernetes-admin Approved,Issued
kubeadm-cert-ydzs-master-cqh4s 52s kubernetes-admin Approved,Issued
kubeadm-cert-ydzs-master-lvbr5 41s kubernetes-admin Approved,Issued
Check the validity period of the certificate after approval:
kubeadm alpha certs check-expiration
CERTIFICATE EXPIRES RESIDUAL TIME EXTERNALLY MANAGED
admin.conf Nov 05, 2029 11:53 UTC 9y no
apiserver Nov 05, 2029 11:54 UTC 9y no
apiserver-etcd-client Nov 05, 2029 11:53 UTC 9y no
apiserver-kubelet-client Nov 05, 2029 11:54 UTC 9y no
controller-manager.conf Nov 05, 2029 11:54 UTC 9y no
etcd-healthcheck-client Nov 05, 2029 11:53 UTC 9y no
etcd-peer Nov 05, 2029 11:53 UTC 9y no
etcd-server Nov 05, 2029 11:54 UTC 9y no
front-proxy-client Nov 05, 2029 11:54 UTC 9y no
scheduler.conf Nov 05, 2029 11:53 UTC 9y no
We can see that it has been extended for 10 years. This is because the validity period of the CA certificate is only 10 years.
But now we can't directly restart several components of the control panel. This is because the etcd corresponding to the cluster installed with kubeadm uses the /etc/kubernetes/pki/etcd/ca.crt certificate to perform the previous step by default. The certificate approved by the command kubectl certificate approve is issued with the default /etc/kubernetes/pki/ca.crt certificate, so we need to replace the CA certificate in etcd:
Copy the static Pod resource list first
cp -r /etc/kubernetes/manifests/ /etc/kubernetes/manifests.bak
vi /etc/kubernetes/manifests/etcd.yaml
......
spec:
containers:
- command:
- etcd
# 修改为 CA 文件
- --peer-trusted-ca-file=/etc/kubernetes/pki/ca.crt
- --trusted-ca-file=/etc/kubernetes/pki/ca.crt
......
volumeMounts:
- mountPath: /var/lib/etcd
name: etcd-data
- mountPath: /etc/kubernetes/pki # 更改证书目录
name: etcd-certs
volumes:
- hostPath:
path: /etc/kubernetes/pki # 将 pki 目录挂载到 etcd 中去
type: DirectoryOrCreate
name: etcd-certs
- hostPath:
path: /var/lib/etcd
type: DirectoryOrCreate
name: etcd-data
......
Since kube-apiserver needs to connect to the etcd cluster, the corresponding etcd ca file also needs to be revised:
vim /etc/kubernetes/manifests/kube-apiserver.yaml
......
spec:
containers:
- command:
- kube-apiserver
# 将etcd ca文件修改为默认的ca.crt文件
- --etcd-cafile=/etc/kubernetes/pki/ca.crt
......
In addition, you need to replace the requestheader-client-ca-file file, the default is the /etc/kubernetes/pki/front-proxy-ca.crt file, and now you need to replace it with the default CA file, otherwise use the aggregation API, such as After installing the metrics-server and executing the kubectl top command, an error will be reported:
cp /etc/kubernetes/pki/ca.crt /etc/kubernetes/pki/front-proxy-ca.crt
cp /etc/kubernetes/pki/ca.key /etc/kubernetes/pki/front-proxy-ca.key
Since it is a static Pod, the above components will automatically restart to take effect after the modification is completed. Since our current version of kubelet has automatic certificate rotation enabled by default, the kubelet certificate does not need to be managed anymore, so I will update the certificate to a validity period of 10. Be sure to back up the certificate directory before operation to prevent rollback of operation errors.