Merge branch 'esterel42Metrics' into 'master'

[Sophia][Esterel42] Add metrics

See merge request !1389

data/grid5000/sites/sophia/clusters/esterel42/esterel42.json

@@ -5,7 +5,78 @@
   "kavlan": true,
   "manufactured_at": "2025-05-30",
   "metrics": [
-
+    {
+      "description": "Default subset of metrics from Prometheus Node Exporter",
+      "name": "prom_default_metrics",
+      "optional_period": 15000,
+      "period": 0,
+      "source": {
+        "id": [
+          "node_boot_time_seconds",
+          "node_cpu_scaling_frequency_hertz",
+          "node_cpu_seconds_total",
+          "node_filesystem_free_bytes",
+          "node_filesystem_size_bytes",
+          "node_load1",
+          "node_load15",
+          "node_load5",
+          "node_memory_Buffers_bytes",
+          "node_memory_Cached_bytes",
+          "node_memory_MemAvailable_bytes",
+          "node_memory_MemFree_bytes",
+          "node_memory_MemTotal_bytes",
+          "node_memory_Shmem_bytes",
+          "node_memory_SwapFree_bytes",
+          "node_memory_SwapTotal_bytes",
+          "node_network_receive_bytes_total",
+          "node_network_receive_packets_total",
+          "node_network_transmit_bytes_total",
+          "node_network_transmit_packets_total",
+          "node_procs_blocked",
+          "node_procs_running",
+          "kwollect_custom"
+        ],
+        "port": 9100,
+        "protocol": "prometheus"
+      }
+    },
+    {
+      "description": "All metrics from Prometheus Node Exporter",
+      "name": "prom_all_metrics",
+      "optional_period": 15000,
+      "period": 0,
+      "source": {
+        "port": 9100,
+        "protocol": "prometheus"
+      }
+    },
+    {
+      "description": "Default subset of metrics from Prometheus Nvidia DCGM Exporter",
+      "name": "prom_nvgpu_default_metrics",
+      "optional_period": 15000,
+      "period": 0,
+      "source": {
+        "id": [
+          "DCGM_FI_DEV_SM_CLOCK",
+          "DCGM_FI_DEV_MEM_CLOCK",
+          "DCGM_FI_DEV_GPU_TEMP",
+          "DCGM_FI_DEV_POWER_USAGE",
+          "DCGM_FI_DEV_MEM_COPY_UTIL"
+        ],
+        "port": 9400,
+        "protocol": "prometheus"
+      }
+    },
+    {
+      "description": "All metrics from Prometheus Nvidia DCGM Exporter",
+      "name": "prom_nvgpu_all_metrics",
+      "optional_period": 15000,
+      "period": 0,
+      "source": {
+        "port": 9400,
+        "protocol": "prometheus"
+      }
+    }
   ],
   "model": "DL380a Gen12",
   "nodes_count": 1,

input/grid5000/sites/sophia/clusters/esterel42/esterel42_metrics.yaml

+---
+metrics:
+
+  - name: prom_default_metrics
+    description: Default subset of metrics from Prometheus Node Exporter
+    period: 0
+    optional_period: 15000
+    source:
+      protocol: prometheus
+      port: 9100
+      id:
+        - node_boot_time_seconds
+        - node_cpu_scaling_frequency_hertz
+        - node_cpu_seconds_total
+        - node_filesystem_free_bytes
+        - node_filesystem_size_bytes
+        - node_load1
+        - node_load15
+        - node_load5
+        - node_memory_Buffers_bytes
+        - node_memory_Cached_bytes
+        - node_memory_MemAvailable_bytes
+        - node_memory_MemFree_bytes
+        - node_memory_MemTotal_bytes
+        - node_memory_Shmem_bytes
+        - node_memory_SwapFree_bytes
+        - node_memory_SwapTotal_bytes
+        - node_network_receive_bytes_total
+        - node_network_receive_packets_total
+        - node_network_transmit_bytes_total
+        - node_network_transmit_packets_total
+        - node_procs_blocked
+        - node_procs_running
+        - kwollect_custom
+
+  - name: prom_all_metrics
+    description: All metrics from Prometheus Node Exporter
+    period: 0
+    optional_period: 15000
+    source:
+      protocol: prometheus
+      port: 9100
+
+  - name: prom_nvgpu_default_metrics
+    description: Default subset of metrics from Prometheus Nvidia DCGM Exporter
+    period: 0
+    optional_period: 15000
+    source:
+      protocol: prometheus
+      port: 9400
+      id:
+        - DCGM_FI_DEV_SM_CLOCK
+        - DCGM_FI_DEV_MEM_CLOCK
+        - DCGM_FI_DEV_GPU_TEMP
+        - DCGM_FI_DEV_POWER_USAGE
+        - DCGM_FI_DEV_MEM_COPY_UTIL
+
+  - name: prom_nvgpu_all_metrics
+    description: All metrics from Prometheus Nvidia DCGM Exporter
+    period: 0
+    optional_period: 15000
+    source:
+      protocol: prometheus
+      port: 9400