Certificates used by Kubernetes control plane components such as the Kubernetes API server have a lifetime of 1 year. Replicated has several mechanisms to ensure these certificates are rotated before they expire.
Replicated 2.43+ provides automatic certificate rotation for the Kubernetes control plane.
Replicated will schedule a Job on each master Node once a week to check the expiry of certificates in
If a Node is found to have any certificate expiring in less than 100 days, all certificates on that Node will be rotated.
No Job is scheduled on worker Nodes because the kubelet is able to automatically rotate its own certificate before it expires.
(The kubelet is the only Kubernetes component running on workers that uses a certificate.)
These weekly certificate rotation Jobs and their Pods will be automatically deleted after completion unless there is a failure or the Replicated log level is set to “debug”.
The default weekly schedule can be changed by setting the
KubernetesCertRotationSchedule param to a different cron expression and restarting the Replicated pod.
On the first node where the Replicated install script was run there is a known issue where the kubelet will continue to use the certificate generated during installation rather than its automatically renewed certificate.
The certificate rotation Job will check for this issue and fix
/etc/kubernetes/kubelet.conf to point to the renewed certificates.
All certificates are automatically upgraded whenever the Replicated install script is re-run and triggers an upgrade of the installed version of Kubernetes. For example, upgrading Replicated from 2.37.1 to 2.38.0 would upgrade Kubernetes from 1.13.5 to 1.15.3, which would cause all certificates to be rotated.
Use the following commands on all master nodes to monitor and rotate certificates as needed:
kubeadm alpha certs check-expiration
kubeadm alpha certs renew all