feat: GPU-Auto-Erkennung (nvidia_smi, rocm-smi); Testserver-Aufraeumen dokumentiert

telegraf/base.conf

@@ -22,6 +22,7 @@
     Content-Type = "application/x-protobuf"
     Content-Encoding = "snappy"
     X-Prometheus-Remote-Write-Version = "0.1.0"
+    Authorization = "Bearer ${TELEGRAF_TOKEN}"
 
 # ---- Logs → Loki (imac) ----
 [[outputs.loki]]
@@ -31,6 +32,7 @@
   namepass = ["docker_log", "apache_log"]
   sanitize_label_names = true
   metric_name_label = ""
+  http_headers = { "Authorization" = "Bearer ${TELEGRAF_TOKEN}" }
 
 # ---- Host-Metriken ----
 [[inputs.cpu]]

telegraf/gpu.conf

@@ -0,0 +1,22 @@
+# GPU-Metriken (Auto-Erkennung: NVIDIA / AMD / Intel)
+#
+# Der Installer bindet diese Datei nur ein, wenn auf dem Host eine GPU
+# erkannt wurde. Fehlt der Treiber, laeuft der Rest unveraendert weiter.
+
+# --- NVIDIA (nvidia-smi) ---
+[[inputs.nvidia_smi]]
+  timeout = "10s"
+  # Liefert u.a.: utilization_gpu, utilization_memory, memory_total, memory_used,
+  # memory_free, temperature_gpu, power_draw, power_limit, fan_speed,
+  # clocks_current_graphics, clocks_current_memory, pstate, name (Tag)
+  [inputs.nvidia_smi.tagdrop]
+    compute_mode = ["*"]
+
+# --- AMD (rocm-smi) ---
+# Aktivieren, wenn `rocm-smi` vorhanden ist. Der Installer prueft das.
+# [[inputs.exec]]
+#   commands = ["/etc/telegraf/gpu-rocm.sh"]
+#   interval = "30s"
+#   timeout = "20s"
+#   data_format = "influx"
+#   name_override = "gpu"