# Copyright 2026 NVIDIA CORPORATION
# SPDX-License-Identifier: Apache-2.0
#
# GPU Operator values for the NVSentinel thermal-margin demo (mock GPUs on Kind).
#
# This mirrors tests/e2e/go/assets/gpu-operator-values.yaml with ONE key
# difference: the standalone DCGM (nv-hostengine) DaemonSet is ENABLED so it
# publishes the "nvidia-dcgm" Service on :5555 that NVSentinel's GPU Health
# Monitor polls (global.dcgm.mode=operator-service). The repo's e2e disables it
# because dcgm-exporter embeds its own hostengine; NVSentinel needs the shared
# service instead.

# The mock replaces the kernel driver + container toolkit.
driver:
  enabled: false
toolkit:
  enabled: false

# Standalone DCGM (nv-hostengine) — REQUIRED for NVSentinel operator-service mode.
# Point it at the mock NVML so nv-hostengine loads libnvidia-ml from the mock
# driver root staged on the host by the nvml-mock DaemonSet.
dcgm:
  enabled: true
  env:
    - name: NVIDIA_DRIVER_ROOT
      value: "/var/lib/nvml-mock/driver"

# dcgm-exporter kept on for an independent cross-check of DCGM_FI_DEV_MEMORY_TEMP / margin fields.
dcgmExporter:
  enabled: true
  env:
    - name: NVIDIA_DRIVER_ROOT
      value: "/var/lib/nvml-mock/driver"
    - name: DCGM_EXPORTER_COLLECT_INTERVAL
      value: "5000"

# MIG unsupported by the mock.
mig:
  strategy: none
migManager:
  enabled: false

nodeStatusExporter:
  enabled: false

# CDI mode — toolkit reads /var/run/cdi/nvidia.yaml generated by nvml-mock.
cdi:
  enabled: true
  default: true

devicePlugin:
  enabled: true
  config:
    name: ""
  env:
    - name: NVIDIA_DRIVER_ROOT
      value: "/var/lib/nvml-mock/driver"

gfd:
  enabled: true
  env:
    - name: NVIDIA_DRIVER_ROOT
      value: "/var/lib/nvml-mock/driver"

validator:
  driver:
    env:
      - name: DRIVER_INSTALL_DIR
        value: "/run/nvidia/driver"
      - name: LD_LIBRARY_PATH
        value: "/run/nvidia/driver/usr/lib64"
      - name: DISABLE_DEV_CHAR_SYMLINK_CREATION
        value: "true"
  toolkit:
    env:
      - name: NVIDIA_VISIBLE_DEVICES
        value: "all"
  cuda:
    env:
      - name: WITH_WORKLOAD
        value: "false"
  plugin:
    env:
      - name: LD_LIBRARY_PATH
        value: "/run/nvidia/driver/usr/lib64"