registry.yaml

# Copyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# Component Registry - Declarative configuration for all AICR components
#
# This file defines the bundler configuration for each component.
# Adding a new component only requires adding an entry here - no Go code needed.
#
# Fields:
#   name:              Component identifier used in recipes (e.g., "gpu-operator")
#   displayName:       Human-readable name for templates and output
#   valueOverrideKeys: Alternative keys for --set flag (e.g., --set gpuoperator:key=value)
#   helm:              Default Helm chart settings (for Helm components)
#     defaultRepository: Helm repository URL
#     defaultChart:      Chart name (e.g., "nvidia/gpu-operator")
#     defaultVersion:    Default chart version if not specified in recipe
#   kustomize:         Default Kustomize settings (for Kustomize components)
#     defaultSource:     Git repository or OCI reference
#     defaultPath:       Path within the repository to the kustomization
#     defaultTag:        Git tag, branch, or commit
#   nodeScheduling:    Paths in Helm values where node selectors/tolerations are injected
#
# Note: A component must have either 'helm' OR 'kustomize' configuration, not both.
# Node scheduling paths define WHERE CLI flags like --system-node-selector are applied.
# The actual values come from CLI flags, not from this file.
#
apiVersion: aicr.nvidia.com/v1alpha1
kind: ComponentRegistry

components:
  - name: gpu-operator
    displayName: gpu-operator
    valueOverrideKeys:
      - gpuoperator
    healthCheck:
      assertFile: checks/gpu-operator/health-check.yaml
    helm:
      defaultRepository: https://helm.ngc.nvidia.com/nvidia
      defaultChart: nvidia/gpu-operator
      defaultNamespace: gpu-operator
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - operator.nodeSelector
          - node-feature-discovery.gc.nodeSelector
          - node-feature-discovery.master.nodeSelector
        tolerationPaths:
          - operator.tolerations
          - node-feature-discovery.gc.tolerations
          - node-feature-discovery.master.tolerations
      accelerated:
        nodeSelectorPaths:
          - daemonsets.nodeSelector
          # Note: NFD worker nodeSelector is intentionally excluded from
          # accelerated scheduling. NFD must run on all nodes (GPU, CPU,
          # and system) to detect hardware features — this matches the
          # gpu-operator default behavior where NFD workers also run on
          # control-plane nodes. NFD worker tolerations remain here so
          # it inherits the same tolerations as GPU daemonsets.
        tolerationPaths:
          - daemonsets.tolerations
          - node-feature-discovery.worker.tolerations

  - name: network-operator
    displayName: network-operator
    valueOverrideKeys:
      - networkoperator
    healthCheck:
      assertFile: checks/network-operator/health-check.yaml
    helm:
      defaultRepository: https://helm.ngc.nvidia.com/nvidia
      defaultChart: nvidia/network-operator
      defaultNamespace: nvidia-network-operator
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - operator.nodeSelector
        tolerationPaths:
          - operator.tolerations

  - name: aws-efa
    displayName: aws-efa
    valueOverrideKeys:
      - awsefa
      - efa
    healthCheck:
      assertFile: checks/aws-efa/health-check.yaml
    helm:
      defaultRepository: https://aws.github.io/eks-charts
      defaultChart: aws-efa-k8s-device-plugin
      defaultVersion: v0.5.3
      defaultNamespace: kube-system
    nodeScheduling:
      accelerated:
        nodeSelectorPaths:
          - nodeSelector
        tolerationPaths:
          - tolerations

  - name: cert-manager
    displayName: cert-manager
    valueOverrideKeys:
      - certmanager
    healthCheck:
      assertFile: checks/cert-manager/health-check.yaml
    helm:
      defaultRepository: https://charts.jetstack.io
      defaultChart: jetstack/cert-manager
      defaultVersion: v1.17.2
      defaultNamespace: cert-manager
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - nodeSelector
          - webhook.nodeSelector
          - cainjector.nodeSelector
          - startupapicheck.nodeSelector
        tolerationPaths:
          - tolerations
          - webhook.tolerations
          - cainjector.tolerations
          - startupapicheck.tolerations

  - name: skyhook-operator
    displayName: skyhook
    valueOverrideKeys:
      - skyhook
    healthCheck:
      assertFile: checks/skyhook-operator/health-check.yaml
    helm:
      defaultRepository: https://nvidia.github.io/skyhook
      defaultChart: skyhook-operator
      defaultNamespace: skyhook
    nodeScheduling:
      nodeCountPaths:
        - estimatedNodeCount
      system:
        nodeSelectorPaths:
          - controllerManager.selectors
        tolerationPaths:
          - controllerManager.tolerations
      accelerated:
        taintStrPaths:
          - controllerManager.manager.env.runtimeRequiredTaint

  - name: skyhook-customizations
    displayName: skyhook-customizations
    valueOverrideKeys:
      - skyhookcustomizations
    healthCheck:
      assertFile: checks/skyhook-customizations/health-check.yaml
    helm:
      # Manifest-only component - no external Helm chart, uses manifestFiles
      defaultRepository: ""
      defaultNamespace: skyhook
    nodeScheduling:
      accelerated:
        nodeSelectorPaths:
          - acceleratedNodeSelector
        tolerationPaths:
          - acceleratedTolerations
    podScheduling:
      workload:
        workloadSelectorPaths:
          - workloadSelector
    validations:
      - function: CheckWorkloadSelectorMissing
        severity: warning
        conditions:
          intent:
            - training
        message: "This may cause skyhook to evict running training jobs. Consider setting --workload-selector to prevent eviction."
      - function: CheckAcceleratedSelectorMissing
        severity: warning
        conditions:
          intent:
            - training
            - inference
        message: "Without this selector, the customization will run on all nodes. Consider setting --accelerated-node-selector to target specific nodes."

  - name: nvsentinel
    displayName: nvsentinel
    valueOverrideKeys:
      - nv-sentinel
    healthCheck:
      assertFile: checks/nvsentinel/health-check.yaml
    helm:
      defaultRepository: https://helm.ngc.nvidia.com/nvidia
      defaultChart: nvidia/nvsentinel
      defaultVersion: v0.10.1
      defaultNamespace: nvsentinel
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - global.systemNodeSelector
        tolerationPaths:
          - global.systemNodeTolerations

  - name: nvidia-dra-driver-gpu
    displayName: nvidia-dra-driver-gpu
    valueOverrideKeys:
      - dradriver
    healthCheck:
      assertFile: checks/nvidia-dra-driver-gpu/health-check.yaml
    helm:
      defaultRepository: https://helm.ngc.nvidia.com/nvidia
      defaultChart: nvidia/nvidia-dra-driver-gpu
      defaultNamespace: nvidia-dra-driver
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - controller.nodeSelector
        tolerationPaths:
          - controller.tolerations
      accelerated:
        nodeSelectorPaths:
          - kubeletPlugin.nodeSelector
        tolerationPaths:
          - kubeletPlugin.tolerations

  - name: kube-prometheus-stack
    displayName: kube-prometheus-stack
    valueOverrideKeys:
      - prometheus
      - kubeprometheusstack
    healthCheck:
      assertFile: checks/kube-prometheus-stack/health-check.yaml
    helm:
      defaultRepository: https://prometheus-community.github.io/helm-charts
      defaultChart: prometheus-community/kube-prometheus-stack
      defaultVersion: 82.8.0
      defaultNamespace: monitoring
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - prometheus.prometheusSpec.nodeSelector
          - alertmanager.alertmanagerSpec.nodeSelector
          - grafana.nodeSelector
          - prometheusOperator.nodeSelector
          - prometheusOperator.admissionWebhooks.patch.nodeSelector
          - kube-state-metrics.nodeSelector
          - prometheus-node-exporter.nodeSelector
          - thanosRuler.thanosRulerSpec.nodeSelector
        tolerationPaths:
          - prometheus.prometheusSpec.tolerations
          - alertmanager.alertmanagerSpec.tolerations
          - grafana.tolerations
          - prometheusOperator.tolerations
          - prometheusOperator.admissionWebhooks.patch.tolerations
          - kube-state-metrics.tolerations
          - prometheus-node-exporter.tolerations
          - thanosRuler.thanosRulerSpec.tolerations

  - name: prometheus-adapter
    displayName: prometheus-adapter
    valueOverrideKeys:
      - prometheusadapter
    healthCheck:
      assertFile: checks/prometheus-adapter/health-check.yaml
    helm:
      defaultRepository: https://prometheus-community.github.io/helm-charts
      defaultChart: prometheus-community/prometheus-adapter
      defaultVersion: 5.3.0
      defaultNamespace: monitoring
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - nodeSelector
        tolerationPaths:
          - tolerations

  - name: aws-ebs-csi-driver
    displayName: aws-ebs-csi-driver
    valueOverrideKeys:
      - ebscsidriver
      - awsebscsidriver
    healthCheck:
      assertFile: checks/aws-ebs-csi-driver/health-check.yaml
    helm:
      defaultRepository: https://kubernetes-sigs.github.io/aws-ebs-csi-driver
      defaultChart: aws-ebs-csi-driver/aws-ebs-csi-driver
      defaultVersion: 2.55.0
      defaultNamespace: kube-system
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - controller.nodeSelector
        tolerationPaths:
          - controller.tolerations
          - node.tolerations
      accelerated:
        # No nodeSelectorPaths — CSI node DaemonSet must run on ALL nodes
        # (not just GPU nodes) since any node may need EBS volumes.
        tolerationPaths:
          - node.tolerations

  - name: k8s-ephemeral-storage-metrics
    displayName: k8s-ephemeral-storage-metrics
    valueOverrideKeys:
      - k8sephemeralstoragemetrics
    healthCheck:
      assertFile: checks/k8s-ephemeral-storage-metrics/health-check.yaml
    helm:
      defaultRepository: https://jmcgrath207.github.io/k8s-ephemeral-storage-metrics/chart
      defaultChart: k8s-ephemeral-storage-metrics/k8s-ephemeral-storage-metrics
      defaultNamespace: monitoring
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - nodeSelector
        tolerationPaths:
          - tolerations

  - name: kai-scheduler
    displayName: kai-scheduler
    valueOverrideKeys:
      - kaischeduler
    healthCheck:
      assertFile: checks/kai-scheduler/health-check.yaml
    helm:
      defaultRepository: oci://ghcr.io/nvidia/kai-scheduler
      defaultChart: kai-scheduler
      defaultVersion: v0.12.17
      defaultNamespace: kai-scheduler
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - global.nodeSelector
        tolerationPaths:
          - global.tolerations

  - name: dynamo-crds
    displayName: dynamo-crds
    valueOverrideKeys:
      - dynamocrds
    helm:
      defaultRepository: https://helm.ngc.nvidia.com/nvidia/ai-dynamo
      defaultChart: dynamo-crds
      defaultVersion: "0.9.1"
      defaultNamespace: dynamo-system

  - name: dynamo-platform
    displayName: dynamo-platform
    valueOverrideKeys:
      - dynamoplatform
      - dynamo
    healthCheck:
      assertFile: checks/dynamo-platform/health-check.yaml
    helm:
      defaultRepository: https://helm.ngc.nvidia.com/nvidia/ai-dynamo
      defaultChart: dynamo-platform
      defaultVersion: "0.9.1"
      defaultNamespace: dynamo-system
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - dynamo-operator.controllerManager.nodeSelector
          - grove.nodeSelector
        tolerationPaths:
          - dynamo-operator.controllerManager.tolerations
          - grove.tolerations

  - name: kgateway-crds
    displayName: kgateway-crds
    valueOverrideKeys:
      - kgatewaycrds
    helm:
      defaultRepository: oci://cr.kgateway.dev/kgateway-dev/charts
      defaultChart: kgateway-crds
      defaultVersion: v2.0.0
      defaultNamespace: kgateway-system

  - name: kgateway
    displayName: kgateway
    valueOverrideKeys:
      - kgateway
    healthCheck:
      assertFile: checks/kgateway/health-check.yaml
    helm:
      defaultRepository: oci://cr.kgateway.dev/kgateway-dev/charts
      defaultChart: kgateway
      defaultVersion: v2.0.0
      defaultNamespace: kgateway-system
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - nodeSelector
        tolerationPaths:
          - tolerations

  - name: kubeflow-trainer
    displayName: kubeflow-trainer
    valueOverrideKeys:
      - kubeflowtrainer
      - trainer
    healthCheck:
      assertFile: checks/kubeflow-trainer/health-check.yaml
    helm:
      defaultRepository: oci://ghcr.io/kubeflow/charts
      defaultChart: kubeflow-trainer
      defaultVersion: 2.1.0
      defaultNamespace: kubeflow
    nodeScheduling:
      system:
        nodeSelectorPaths:
          - manager.nodeSelector
          - jobset.controller.nodeSelector
        tolerationPaths:
          - manager.tolerations
          - jobset.controller.tolerations