Skip to content

installation.md

Latest commit

 

History

History
142 lines (106 loc) · 6.02 KB

installation.md

File metadata and controls

142 lines (106 loc) · 6.02 KB
Raw

Installation

NodeWright is installed via Helm as an OCI artifact from GitHub Container Registry.

Prerequisites

  • Kubernetes cluster (tested on v1.30+)
  • Helm 3.8+ installed (required for native OCI support)
  • Container registry access credentials (if using private registries)

Configure Image Pull Secrets (if needed)

If you're using private container registries, create the namespace and secret before installing so the release can reference it:

kubectl create namespace nodewright

kubectl create secret generic node-init-secret \
  --from-file=.dockerconfigjson=${HOME}/.docker/config.json \
  --type=kubernetes.io/dockerconfigjson \
  --namespace nodewright

NodeWright currently uses a single shared image pull secret for all packages, and agent/operator containers. If you need access to multiple registries, combine the credentials into one dockerconfigjson secret with multiple registry auths.

Install NodeWright

# The chart is distributed as an OCI artifact on GitHub Container Registry.
# Helm 3.8+ supports OCI natively — no `helm repo add` needed.
helm install nodewright oci://ghcr.io/nvidia/nodewright/charts/nodewright \
  --version v0.18.0 \
  --namespace nodewright \
  --create-namespace \
  --set imagePullSecret=node-init-secret

Omit --set imagePullSecret=node-init-secret if you're pulling from public registries only.

Where things live: chart at oci://ghcr.io/nvidia/nodewright/charts/nodewright, operator image at ghcr.io/nvidia/nodewright/operator, agent image at ghcr.io/nvidia/nodewright/agent. Migrating from helm repo add skyhook https://helm.ngc.nvidia.com/...? Run helm repo remove skyhook, then point your existing release at the OCI chart. Because the release already exists, this is an upgrade, not an install — helm install fails on a name that is already in use:

helm upgrade <release-name> oci://ghcr.io/nvidia/nodewright/charts/nodewright \
  --version v0.18.0 --namespace <existing-namespace>

Keeping the old release name (e.g. skyhook) is fine — the chart works either way.

Already installed in the skyhook namespace? Stay there. The documented namespace for new installs moved from skyhook to nodewright, but Kubernetes namespaces cannot be renamed in place and Helm cannot move a release between namespaces, so there is nothing to migrate and no deadline. kubectl nodewright finds the operator in either namespace automatically. See Install namespace.

Verify Installation

# Check that the operator is running
kubectl get pods -n nodewright

# Wait for the deployment to be available
kubectl wait --for=condition=Available deployment -l control-plane=controller-manager -n nodewright --timeout=300s

# Wait for the operator pod to be ready
kubectl wait --for=condition=Ready pod -l control-plane=controller-manager -n nodewright --timeout=300s

# Verify the CRDs are installed
kubectl get crd | grep nodewright

Uninstalling

By default, the Helm chart includes a pre-delete hook that automatically cleans up all NodeWright and DeploymentPolicy resources before uninstalling:

helm uninstall nodewright --namespace nodewright

The pre-delete hook will:

  • Delete all NodeWright resources, and legacy Skyhook resources if any remain
  • Delete DeploymentPolicy resources in both API groups
  • Complete quickly if no resources exist
  • Wait for finalizers to be processed if resources exist

Each delete is best-effort — a failing one does not fail the hook. The job as a whole is bounded by cleanup.jobTimeoutSeconds (default 120), and that bound is a hard deadline: if a finalizer keeps the job waiting past it, the job is killed and marked Failed, which fails the pre-delete hook and so fails helm uninstall. If that happens, clean up by hand using the commands below and re-run the uninstall with --no-hooks.

Cleanup options

Set these at install time with helm install, or on an existing release with helm upgrade. When changing a setting on a release you already have, pass both of these or you will change more than you meant to:

  • --version pinned to the chart version you are already running. Without it, helm upgrade against an OCI chart resolves to the newest published version, so flipping a cleanup flag would also upgrade the operator.
  • --reuse-values, so the values you set at install time (an imagePullSecret, say) survive. Without it Helm starts from the chart defaults and applies only your --set flags.
# Disable automatic cleanup and manage resources manually
helm upgrade nodewright oci://ghcr.io/nvidia/nodewright/charts/nodewright \
  --version v0.18.0 --reuse-values \
  --namespace nodewright --set cleanup.enabled=false
# Adjust the cleanup job timeout
helm upgrade nodewright oci://ghcr.io/nvidia/nodewright/charts/nodewright \
  --version v0.18.0 --reuse-values \
  --namespace nodewright --set cleanup.jobTimeoutSeconds=180

Manual cleanup

If you disabled automatic cleanup, or need to clean up by hand:

# Delete all NodeWright and DeploymentPolicy resources first. This mirrors what
# the pre-delete hook does: a migrated cluster can still hold legacy resources,
# and their finalizers block uninstall just as the current ones do.
kubectl delete nodewrights --all --ignore-not-found
kubectl delete skyhooks --all --ignore-not-found
kubectl delete deploymentpolicies.nodewright.nvidia.com --all --ignore-not-found
kubectl delete deploymentpolicies.skyhook.nvidia.com --all --ignore-not-found

# Then uninstall the chart
helm uninstall nodewright --namespace nodewright

Why cleanup matters: uninstalling while NodeWright CRs with finalizers still exist can leave resources in a broken state that causes problems on reinstall.

For more details on explicit package uninstall, see Uninstall.

Related