# Vendored plain-manifest equivalent of the utkuozdemir/nvidia_gpu_exporter Helm chart
# (pinned at chart app-version v1.14.0: https://github.com/utkuozdemir/nvidia_gpu_exporter),
# committed here instead of an ad hoc `helm install` so the exporter is an applied
# artifact like the rest of bootstrap/prereqs/ (see ../README.md).
#
# Why this exporter and not NVIDIA DCGM: DCGM has limited support on consumer GeForce
# cards under WSL2 (some DCGM_FI_* fields come back empty — see
# ../../phase-2-capstone/gpu-node/diy-cluster.md and
# ../../phase-2-capstone/tech-stack.md). This exporter just parses `nvidia-smi`, which
# works fine in WSL2, and exposes util / mem / temp / power on :9835/metrics.
#
# `hostNetwork: true` / `dnsPolicy: ClusterFirstWithHostNet` are baked in directly (not
# patched on after the fact) because cross-node ClusterIP routing is down under this
# cluster's WSL2 mirrored networking — see
# ../../docs/cluster-troubleshooting-log.md (§7) and bootstrap/platform.sh's `netfix`
# layer, which used to `kubectl patch` this in; vendoring makes that patch unnecessary.
apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: nvidia-gpu-exporter
  namespace: monitoring
  labels:
    app.kubernetes.io/name: nvidia-gpu-exporter
spec:
  selector:
    matchLabels:
      app.kubernetes.io/name: nvidia-gpu-exporter
  updateStrategy:
    type: RollingUpdate
  template:
    metadata:
      labels:
        app.kubernetes.io/name: nvidia-gpu-exporter
    spec:
      runtimeClassName: nvidia
      tolerations:
        - key: nvidia.com/gpu
          operator: Exists
          effect: NoSchedule
      hostNetwork: true
      dnsPolicy: ClusterFirstWithHostNet
      containers:
        - name: nvidia-gpu-exporter
          image: docker.io/utkuozdemir/nvidia_gpu_exporter:1.14.0
          env:
            - name: NVIDIA_VISIBLE_DEVICES
              value: "all"
            - name: NVIDIA_DRIVER_CAPABILITIES
              value: utility
          args:
            - --web.listen-address
            - ":9835"
            - --web.telemetry-path
            - /metrics
            - --nvidia-smi-command
            - nvidia-smi
            - --query-field-names
            - AUTO
          ports:
            - name: http
              containerPort: 9835
              protocol: TCP
          livenessProbe:
            httpGet:
              path: /-/healthy
              port: http
          readinessProbe:
            httpGet:
              path: /-/ready
              port: http
          resources:
            requests:
              cpu: 25m
              memory: 32Mi
            limits:
              cpu: 100m
              memory: 64Mi