feat: monitoring — node_exporter + ROCm sysfs collector + alertmanager webhook

Ansible roles:
- node_exporter : déploie node_exporter v1.9.1 sur storage-01 et gpu-01
  avec textfile collector + services systemd. ROCm activé sur gpu-01 uniquement
  via group_vars. Collecte métriques GPU via sysfs (temp, VRAM, utilisation)
  — pas de dépendance rocm-smi (remplacé dans ROCm 7.x)
- alertmanager_webhook : service Python sur storage-01 (:9093/webhook)
  reçoit alertes AlertManager et les route vers ask-agent monitor → Hermes

Playbooks: node_exporter + alertmanager_webhook ajoutés à storage-01 et gpu-01

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

ansible/roles/node_exporter/templates/node_exporter.service.j2

@@ -0,0 +1,18 @@
+[Unit]
+Description=Prometheus Node Exporter
+After=network.target
+
+[Service]
+User={{ node_exporter_user }}
+Group={{ node_exporter_user }}
+ExecStart=/usr/local/bin/node_exporter \
+  --web.listen-address=":{{ node_exporter_port }}" \
+  --collector.textfile.directory={{ node_exporter_textfile_dir }} \
+  --collector.systemd \
+  --collector.processes
+
+Restart=on-failure
+RestartSec=5s
+
+[Install]
+WantedBy=multi-user.target

ansible/roles/node_exporter/templates/rocm_scraper.service.j2

@@ -0,0 +1,7 @@
+[Unit]
+Description=ROCm metrics scraper for Prometheus textfile collector
+After=multi-user.target
+
+[Service]
+Type=oneshot
+ExecStart=/usr/local/bin/rocm_scraper.sh

ansible/roles/node_exporter/templates/rocm_scraper.sh.j2

@@ -0,0 +1,33 @@
+#!/bin/bash
+# Collecte les métriques GPU AMD via sysfs (pas de rocm-smi requis)
+OUTPUT="{{ node_exporter_textfile_dir }}/rocm.prom"
+TMPFILE="${OUTPUT}.tmp"
+
+{
+  echo "# HELP rocm_gpu_temperature_celsius GPU temperature in Celsius"
+  echo "# TYPE rocm_gpu_temperature_celsius gauge"
+  echo "# HELP rocm_gpu_utilization_percent GPU utilization percentage"
+  echo "# TYPE rocm_gpu_utilization_percent gauge"
+  echo "# HELP rocm_vram_used_bytes VRAM used in bytes"
+  echo "# TYPE rocm_vram_used_bytes gauge"
+  echo "# HELP rocm_vram_total_bytes VRAM total in bytes"
+  echo "# TYPE rocm_vram_total_bytes gauge"
+
+  for card in /sys/class/drm/card[0-9]; do
+    [ -d "$card/device" ] || continue
+    idx=$(basename "$card" | tr -d 'card')
+    labels="gpu=\"${idx}\",model=\"gfx1031\""
+
+    temp_raw=$(cat "$card/device/hwmon/hwmon"*/temp1_input 2>/dev/null | head -1)
+    [ -n "$temp_raw" ] && echo "rocm_gpu_temperature_celsius{${labels}} $(echo "scale=1; ${temp_raw}/1000" | bc)"
+
+    util=$(cat "$card/device/gpu_busy_percent" 2>/dev/null)
+    [ -n "$util" ] && echo "rocm_gpu_utilization_percent{${labels}} ${util}"
+
+    vram_used=$(cat "$card/device/mem_info_vram_used" 2>/dev/null)
+    [ -n "$vram_used" ] && echo "rocm_vram_used_bytes{${labels}} ${vram_used}"
+
+    vram_total=$(cat "$card/device/mem_info_vram_total" 2>/dev/null)
+    [ -n "$vram_total" ] && echo "rocm_vram_total_bytes{${labels}} ${vram_total}"
+  done
+} > "$TMPFILE" && mv "$TMPFILE" "$OUTPUT"

ansible/roles/node_exporter/templates/rocm_scraper.timer.j2

@@ -0,0 +1,11 @@
+[Unit]
+Description=ROCm metrics scraper timer
+Requires=rocm_scraper.service
+
+[Timer]
+OnBootSec=30s
+OnUnitActiveSec={{ rocm_scrape_interval }}s
+AccuracySec=5s
+
+[Install]
+WantedBy=timers.target