Certificate Rotation

Rotate certificates for Kubernetes cluster components in RKE2 and K3s clusters

Certificate rotation allows you to renew TLS certificates for Kubernetes cluster components before they expire. Rancher automates certificate rotation for RKE2 and K3s clusters managed through the CAPI (Cluster API) provisioning system.

Overview

Kubernetes clusters use TLS certificates for:

• API server authentication
• etcd cluster communication
• kubelet authentication
• Service account token signing
• Controller manager and scheduler secure connections

These certificates have expiration dates and must be rotated periodically to maintain cluster security and operation.

Supported Clusters

Certificate rotation is available for:

• RKE2 clusters (ClusterDriverRke2)
• K3s clusters (ClusterDriverK3s)
• Clusters managed through CAPI RKEControlPlane

Certificate rotation is not available for:

• Imported clusters
• Hosted Kubernetes services (EKS, GKE, AKS)
• Legacy RKE clusters (deprecated)

These cluster types must use their native certificate management tools.

Cluster Action

The certificate rotation action is defined in /home/daytona/workspace/source/pkg/apis/management.cattle.io/v3/cluster_types.go:35:

ClusterActionRotateCertificates = "rotateCertificates"

This action triggers the certificate rotation workflow through the CAPI planner.

Certificate Rotation Input

The rotation input type (defined in /home/daytona/workspace/source/pkg/apis/management.cattle.io/v3/cluster_types.go:372):

type RotateCertificateInput struct {
    CACertificates bool      // Rotate the root CA certificates
    Services       []string  // Specific services to rotate, or empty for all
}

Service Options

Valid service names for selective rotation:

• etcd - etcd peer and client certificates
• kubelet - kubelet serving and client certificates
• kube-apiserver - API server serving certificate
• kube-proxy - kube-proxy client certificate
• kube-scheduler - scheduler client certificate
• kube-controller-manager - controller manager client certificate
• rke2-server / k3s-server - Server component certificates
• rke2-controller / k3s-controller - Controller certificates
• admin - Admin client certificate
• cloud-controller - Cloud controller client certificate
• auth-proxy - Authentication proxy certificate

If Services is empty, all certificates are rotated.

Performing Certificate Rotation

Via Rancher UI

Via Rancher API

Trigger rotation via API:

curl -X POST \
  'https://{RANCHER_URL}/v3/clusters/{CLUSTER_ID}?action=rotateCertificates' \
  -H 'Authorization: Bearer {TOKEN}' \
  -H 'Content-Type: application/json' \
  -d '{
    "caCertificates": false,
    "services": ["kube-apiserver", "kubelet"]
  }'

Rotate all certificates:

curl -X POST \
  'https://{RANCHER_URL}/v3/clusters/{CLUSTER_ID}?action=rotateCertificates' \
  -H 'Authorization: Bearer {TOKEN}' \
  -H 'Content-Type: application/json' \
  -d '{
    "caCertificates": false,
    "services": []
  }'

Rotation Process

The rotation workflow is implemented in /home/daytona/workspace/source/pkg/capr/planner/certificaterotation.go:17.

Rotation Plans

The planner generates node-specific rotation plans (from /home/daytona/workspace/source/pkg/capr/planner/certificaterotation.go:90):

Control Plane Node Plan

# Stop service
systemctl stop rke2-server

# Rotate certificates
rke2 certificate rotate -s kube-apiserver -s etcd

# Remove auto-generated certs
rm -f /var/lib/rancher/rke2/server/tls/kube-controller-manager.crt
rm -f /var/lib/rancher/rke2/server/tls/kube-controller-manager.key
rm -f /var/lib/rancher/rke2/server/tls/kube-scheduler.crt  
rm -f /var/lib/rancher/rke2/server/tls/kube-scheduler.key

# Remove static manifests
rm -f /var/lib/rancher/rke2/agent/pod-manifests/kube-controller-manager.yaml
rm -f /var/lib/rancher/rke2/agent/pod-manifests/kube-scheduler.yaml

# Restart service
systemctl start rke2-server

Worker Node Plan

# Restart with new certificates
systemctl restart rke2-agent

Worker nodes only require a service restart to pick up rotated CA certificates.

Downtime Considerations

High Availability Clusters

• Control plane: Rolling restart with no downtime if 3+ control plane nodes
• Workloads: Continue running on worker nodes during control plane rotation
• API availability: May experience brief interruptions during API server rotation

Single-Node Clusters

• API unavailable: API server is unavailable during rotation
• Workload impact: Workloads continue running but cannot be managed
• Duration: Typically 2-5 minutes per node

Service-Specific Impact

Rotating specific services reduces downtime:

• kubelet only: Minimal impact, no control plane restart
• etcd only: Brief etcd cluster interruption
• kube-apiserver only: API unavailable during rotation

CA Certificate Rotation

Rotating CA certificates (caCertificates: true) requires a full cluster restart and causes significant downtime. Plan maintenance windows accordingly.

CA rotation:

• Generates new root CA certificates
• Re-issues all service certificates with the new CA
• Requires all components to restart and trust the new CA
• Takes longer than standard certificate rotation
• Should be performed during scheduled maintenance

Certificate Expiration

Rancher tracks certificate expiration in the cluster status:

type ClusterStatus struct {
    ...
    CertificatesExpiration map[string]CertExpiration
}

type CertExpiration struct {
    ExpirationDate string
}

View certificate expiration:

kubectl get cluster {CLUSTER_NAME} -o jsonpath='{.status.certificatesExpiration}'

Plan rotation before certificates expire:

• 30 days before expiration: Schedule rotation
• 7 days before expiration: Perform rotation
• After expiration: Cluster components will fail to authenticate

Troubleshooting

Rotation Fails to Start

• Cluster not initialized: Verify Status.ControlPlaneInitialized is true
• No init node: Cluster must have an identified init node
• Previous rotation in progress: Wait for current rotation to complete

Check rotation status:

kubectl get rkecontrolplane {CLUSTER_NAME} -n fleet-default -o yaml

Node Fails Health Check

• Service failed to start: Check systemd logs
  journalctl -u rke2-server -n 100
  
• Certificate mismatch: Old and new certificates may conflict
• etcd cluster issues: Verify etcd cluster health

Cluster Stuck During Rotation

• CAPI cluster paused: Verify CAPI cluster is paused during rotation
• Plan not applied: Check node plan status in RKE2 control plane
• Network issues: Ensure nodes can communicate during rotation

Force rotation completion:

kubectl patch rkecontrolplane {CLUSTER_NAME} -n fleet-default --type=merge \
  -p '{"spec":{"rotateCertificates":{"generation":0}}}'

Kubeconfig Stops Working

After rotation, kubeconfig files may need updating:

1. Generate new kubeconfig: See Kubeconfig Access
2. Update local kubeconfig: kubectl config use-context {CLUSTER_NAME}
3. Distribute new kubeconfig to users

Best Practices

• Regular rotation: Rotate certificates every 6-12 months
• Before expiration: Rotate at least 30 days before expiration
• Selective rotation: Rotate specific services to minimize impact
• Maintenance windows: Schedule CA rotation during planned downtime
• Testing: Test rotation in non-production clusters first
• Monitoring: Track certificate expiration dates
• Documentation: Keep runbooks for rotation procedures

Related Resources

• Kubeconfig Access
• Backup and Restore
• Cluster API Reference