Installation

Follow this guide to install the Node Readiness Controller in your Kubernetes cluster.

Prerequisites

If you plan to use the install-full.yaml option (which includes secure metrics and the validating admission webhook), you must first have cert-manager installed in your cluster.

Deployment Options

Option 1: Official Release (Recommended)

First, to install the CRDs, apply the crds.yaml manifest:

# Replace with the desired version
VERSION=v0.5.0

kubectl apply -f https://github.com/kubernetes-sigs/node-readiness-controller/releases/download/${VERSION}/crds.yaml
kubectl wait --for condition=established --timeout=30s crd/nodereadinessrules.readiness.node.x-k8s.io

2. Install the Controller

Choose one of the two following manifests based on your requirements:

Standard Installation (Minimal): The simplest way to deploy the controller with no external dependencies.

kubectl apply -f https://github.com/kubernetes-sigs/node-readiness-controller/releases/download/${VERSION}/install.yaml

Full Installation (Production Ready): Includes secure metrics (TLS-protected) and validating webhooks for rule conflict prevention. Requires cert-manager to be installed in your cluster.

kubectl apply -f https://github.com/kubernetes-sigs/node-readiness-controller/releases/download/${VERSION}/install-full.yaml

This will deploy the controller into the nrr-system namespace on any available node in your cluster.

Controller priority

The controller is deployed with system-cluster-critical priority to prevent eviction during node resource pressure.

If it gets evicted during resource pressure, nodes can’t transition to Ready state, blocking all workload scheduling cluster-wide.

This is the priority class used by other critical cluster components (eg: core-dns).

Images

The official releases use multi-arch images (AMD64, Arm64) and are available at registry.k8s.io/node-readiness-controller/node-readiness-controller

REPO="registry.k8s.io/node-readiness-controller/node-readiness-controller"
TAG=$(skopeo list-tags docker://$REPO | jq .'Tags[-1]' | tr -d '"')
docker pull $REPO:$TAG

Option 2: Helm Chart

The official Helm chart is published to the OCI registry at registry.k8s.io/node-readiness-controller/charts/node-readiness-controller.

# Chart version, which is independent of the controller release tag above.
CHART_VERSION=0.5.0

helm install node-readiness-controller \
  oci://registry.k8s.io/node-readiness-controller/charts/node-readiness-controller \
  --version ${CHART_VERSION} \
  --namespace nrr-system --create-namespace

Requires Helm 3.8+ (native OCI support). This deploys the controller with the same defaults as the standard manifest: leader election on, metrics off, and the validating webhook off.

helm show values lists everything the chart exposes, and redirecting it gives you a starting point to edit:

helm show values oci://registry.k8s.io/node-readiness-controller/charts/node-readiness-controller --version ${CHART_VERSION} > custom-values.yaml

Override them with --set, or from a file with -f:

helm install node-readiness-controller \
  oci://registry.k8s.io/node-readiness-controller/charts/node-readiness-controller \
  --version ${CHART_VERSION} \
  --namespace nrr-system --create-namespace \
  --set metrics.enabled=true

helm install node-readiness-controller \
  oci://registry.k8s.io/node-readiness-controller/charts/node-readiness-controller \
  --version ${CHART_VERSION} \
  --namespace nrr-system --create-namespace \
  -f custom-values.yaml

Optional components

Everything beyond the core controller is opt-in, matching the kustomize components.

The webhook rejects rules whose taint key and effect collide with an existing rule over an overlapping node selector, so it is worth enabling in production.

helm install node-readiness-controller \
  oci://registry.k8s.io/node-readiness-controller/charts/node-readiness-controller \
  --version ${CHART_VERSION} \
  --namespace nrr-system --create-namespace \
  --set certManager.enabled=true \
  --set webhook.enabled=true \
  --set validatingWebhook.enabled=true \
  --set metrics.enabled=true \
  --set metrics.secure=true

Both webhook.enabled and validatingWebhook.enabled are needed. The first runs the webhook server in the controller and mounts its certificate, the second registers the ValidatingWebhookConfiguration with the API server.

Tuning for larger clusters

The values under controller map to the manager’s own flags, and each one is only passed to the container when you move it off its default:

Raising the two concurrency values is the usual response to readiness taints lagging behind node joins on a large cluster. enableNodeStateMetrics adds the per-rule node_readiness_nodes_by_state gauge at the cost of extra API reads on node updates, so turn it on when you want the fleet view and can afford the reads.

Upgrading

Upgrade the release in place using the OCI chart. Values you set at install time are carried over, so only pass the ones you are changing:

helm upgrade node-readiness-controller \
  oci://registry.k8s.io/node-readiness-controller/charts/node-readiness-controller \
  --version ${CHART_VERSION} \
  --namespace nrr-system \
  -f custom-values.yaml

helm upgrade --install works too if you want one command that handles both the first install and later upgrades.

Check what changed before applying it to a live cluster:

helm diff upgrade node-readiness-controller \
  oci://registry.k8s.io/node-readiness-controller/charts/node-readiness-controller \
  --version ${CHART_VERSION} \
  --namespace nrr-system   # needs the helm-diff plugin

Read the CRD note below first. Helm will not update the CRD for you, so a chart bump that changes the schema needs that step done by hand.

CRD upgrades

Helm installs the CRD from the chart’s crds/ directory on first install only. It does not upgrade or remove it on helm upgrade or helm uninstall.

Before moving to a chart version that changes the NodeReadinessRule schema, apply the CRD yourself. Use the controller release that the chart version ships, which helm show chart reports as its appVersion:

RELEASE=$(helm show chart \
  oci://registry.k8s.io/node-readiness-controller/charts/node-readiness-controller \
  --version ${CHART_VERSION} | awk '/^appVersion:/ {print $2}')

kubectl apply -f https://github.com/kubernetes-sigs/node-readiness-controller/releases/download/${RELEASE}/crds.yaml

Skipping this leaves the old schema in place, and rules using newly added fields are rejected by the API server even though the controller supports them.

Option 3: Advanced Deployment (Kustomize)

If you need deeper customization, you can use Kustomize directly from the source.

# 1. Install CRDs
kubectl apply -k config/crd

# 2. Deploy Controller with default configuration
kubectl apply -k config/default

You can enable optional components (Metrics, TLS, Webhook) by creating a kustomization.yaml that includes the relevant components from the config/ directory. For reference on how these components can be combined, see the deploy-with-metrics, deploy-with-tls, deploy-with-webhook, and deploy-full targets in the projects Makefile.

Option 4: Deploy as a Static Pod (Control Plane)

Running the controller as a Static Pod on control-plane nodes is useful for self-managed clusters (e.g., kubeadm) where you want the controller to be available alongside core components like the API server.

Refer to the examples/static-pod/node-readiness-controller.yaml for a detailed example on deploying the controller as a static pod in a kind cluster.

Deployment Steps

1. Prepare the Manifest: Refer the example manifest in examples/static-pod/node-readiness-controller.yaml. This manifest handles kubeconfig with a initContainer and necessary flags for leader election.

2. Deploy to Nodes: Use Ansible / Terraform to copy the manifest to the /etc/kubernetes/manifests/ directory on each control-plane node. The Kubelet will automatically detect and start the pod.

3. Install CRDs:

   kubectl apply -k config/crd
   

   This is typically handled via a bootstrap script or post-install job in a kubeadm setup.

Verification

After installation, verify that the controller is running successfully.

Note

Replace ${NAMESPACE} with the namespace where the controller is deployed (typically nrr-system for standard deployments, or kube-system for static pods).

1. Check Pod Status:

   kubectl get pods -n ${NAMESPACE} -l component=node-readiness-controller
   

   You should see the controller pods in Running status.

2. Check Logs:

   kubectl logs -n ${NAMESPACE} -l component=node-readiness-controller
   

   Look for “Starting EventSource” or “Starting Controller” messages indicating the manager is active.

3. Verify CRDs:

   kubectl get crd nodereadinessrules.readiness.node.x-k8s.io
   

4. Verify High Availability: In an HA cluster, verify that one instance has acquired the leader lease:

   # The lease namespace should match the controller's namespace (configured via --leader-election-namespace)
   kubectl get lease -n ${NAMESPACE} ba65f13e.readiness.node.x-k8s.io
   

Uninstallation

Important

Follow this order to avoid “stuck” resources.

The controller uses a finalizer (readiness.node.x-k8s.io/cleanup-taints) on NodeReadinessRule resources to ensure taints are safely removed from nodes before a rule is deleted.

Caution

You must delete all rule objects before deleting the controller.

1. Delete all Rules:

   kubectl delete nodereadinessrules --all
   

   Wait for this command to complete. This ensures the running controller removes its taints from your nodes.

2. Uninstall Controller:

   # If installed via release manifest
   kubectl delete -f https://github.com/kubernetes-sigs/node-readiness-controller/releases/download/${VERSION}/install.yaml
   
   # Or if using the full manifest
   kubectl delete -f https://github.com/kubernetes-sigs/node-readiness-controller/releases/download/${VERSION}/install-full.yaml
   
   # OR if using Kustomize
   kubectl delete -k config/default
   
   # OR if using Helm
   helm uninstall node-readiness-controller --namespace nrr-system
   
   # OR if using Static Pods
   # Remove the manifest from /etc/kubernetes/manifests/ on all control-plane nodes
   

3. Uninstall CRDs (Optional):

   kubectl delete -k config/crd
   

   Helm does not remove the CRD it installed from the chart’s crds/ directory, so delete it explicitly if you used the chart.

Recovering from Stuck Resources

If you accidentally deleted the controller before the rules, the NodeReadinessRule objects will get stuck in a Terminating state because the controller is needed to cleanup the taints and finalizers.

To force-delete them (this will require you to manually clean up the managed taints if any on your nodes):

# Patch the finalizer to remove it
kubectl patch nodereadinessrule <rule-name> -p '{"metadata":{"finalizers":[]}}' --type=merge

Troubleshooting Deployment

RBAC Permissions If the controller logs show “Forbidden” errors, verify the ClusterRole bindings:

kubectl describe clusterrole nrr-manager-role

It requires nodes (update/patch) and nodereadinessrules (all) access.

Debug Logging To enable verbose logging for deeper investigation:

kubectl patch deployment -n nrr-system nrr-controller-manager \
  -p '{"spec":{"template":{"spec":{"containers":[{"name":"manager","args":["--zap-log-level=debug"]}]}}}}'