| name | ocp-node-feature-discovery |
| metadata | {"author":"rhoai3-demo","version":"1.0.0","platform-family":"ocp","platform-baseline":"repo","ocp-baseline":"repo","skill-group":"OpenShift Platform"} |
| description | Use when documenting, reviewing, or rebuilding OpenShift Container Platform Node Feature Discovery guidance from official OCP documentation: specialized hardware discovery, NFD Operator installation, NodeFeatureDiscovery custom resources, operand.image handling, NFD worker configuration, feature labels, NodeFeatureRule labels and taints, NFD Topology Updater, NodeResourceTopology, PCI, USB, kernel, and custom feature sources, and accelerator node label validation, and the Red Hat CoP NFD GitOps catalog pattern. Do NOT use for provisioning GPU workers; use ocp-machine-management. Do NOT use for NVIDIA GPU Operator, ClusterPolicy, or RHOAI hardware profiles; use rhoai-nvidia-gpu-accelerators. Do NOT run live NFD changes without the env safety guard and explicit approval.
|
OCP Node Feature Discovery
Use this skill to ground OpenShift Node Feature Discovery (NFD) installation,
configuration, feature-label review, and topology updater guidance in the
official OpenShift Container Platform Specialized hardware and driver
enablement guide for the active baseline in docs/PLATFORM_BASELINE.md.
Source Grounding
Read references/source-capture.md before using product behavior. Official Red
Hat documentation is product authority. This skill covers NFD and NFD-created
hardware metadata; it does not replace Machine API, MachineSet, NVIDIA GPU
Operator, Kernel Module Management, Driver Toolkit, or RHOAI accelerator
skills.
Demo NFD Posture
For this AWS-hosted RHOAI demo:
- Use NFD as the OpenShift hardware-discovery layer that publishes feature
labels used by accelerator and workload-placement workflows.
- Treat NFD as a prerequisite gate for NVIDIA GPU enablement, not as the source
of
nvidia.com/gpu capacity. GPU capacity is exposed by the NVIDIA GPU
Operator and device plugin and is handled in rhoai-nvidia-gpu-accelerators.
- Use the Red Hat CoP
nfd catalog as a local curation pattern only. Reuse its
operator/channel layout, aggregate overlay shape, and NVIDIA-focused instance
overlay idea; do not commit remote Kustomize references.
- For the NVIDIA-only demo path, prefer an
only-nvidia NFD instance overlay
that limits PCI label publication to the accelerator-relevant class/vendor
signal after the active CRD schema and official docs are verified.
- Verify feature labels from the live cluster or official CRD/schema before
writing placement rules, hardware-profile assumptions, or README claims.
- Use
NodeFeatureRule only for intentional vendor-specific or
application-specific labels and taints. Do not create broad scheduling taints
unless the scheduling impact is documented.
- Enable NFD Topology Updater only when the implementation requires
topology-aware placement or NUMA/resource-zone information.
NFD Model
Use the official docs to frame:
- NFD Operator: installs and manages Node Feature Discovery in OpenShift.
- NodeFeatureDiscovery: the custom resource that configures NFD operands,
worker behavior, enabled feature sources, label filtering, and optional
topology updater workers.
- Feature labels: node labels generated from hardware and OS discovery
sources such as PCI, USB, kernel, CPU, and custom hooks.
- NodeFeatureRule: rule-based custom labeling and optional tainting based
on discovered node features.
- NFD Topology Updater: per-node daemon that reports per-zone resource
availability and causes
NodeResourceTopology custom resources to be
created.
Workflow
- Confirm the active OpenShift baseline in
docs/PLATFORM_BASELINE.md.
- Read
references/official-doc-extraction.md.
- Read
references/gitops-catalog-nfd-pattern.md when rebuilding NFD
operator, instance, aggregate overlays, or NVIDIA-only discovery posture.
- Identify whether the task concerns:
- NFD Operator installation or subscription review
NodeFeatureDiscovery custom resource configuration
- feature sources, label filtering, or no-publish behavior
- PCI, USB, kernel, CPU, or custom feature labels
NodeFeatureRule labels or taints
- NFD Topology Updater and
NodeResourceTopology
- GPU accelerator discovery handoff to RHOAI/NVIDIA skills
- For GitOps manifests, verify CR fields against the active cluster schema or
official docs before committing.
- For live operations, use the repo environment guard and pair this skill with
env-troubleshoot, env-manage-resources, or env-deploy-and-evaluate.
- Validate the output with
references/validation-checklist.md.
Related Skills
- Use
ocp-machine-management for GPU worker MachineSets and node creation.
- Use
ocp-nodes for node scheduling, labels, taints, tolerations, and
placement mechanics after hardware labels exist.
- Use
ocp-machine-configuration for Machine Config Operator, KubeletConfig,
ContainerRuntimeConfig, and node operating-system configuration.
- Use
rhoai-nvidia-gpu-accelerators for NVIDIA GPU Operator, ClusterPolicy,
RHOAI accelerator profiles, and GPU serving intent.
- Create separate future
ocp-driver-toolkit or ocp-kernel-module-management
skills for Driver Toolkit and Kernel Module Management guidance from the same
OCP documentation family.
- Use
project-gitops-authoring and project-manifest-review for manifests.
References
references/source-capture.md
references/official-doc-extraction.md
references/gitops-catalog-nfd-pattern.md
references/validation-checklist.md
examples/nfd-review-patterns.md