Merge branch 'production-esterel3' into 'master'

[sophia][esterel3] Add kwollect metrics

See merge request !1162

data/grid5000/sites/sophia/clusters/esterel3/esterel3.json

@@ -49,6 +49,33 @@
         "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": "Dell PowerEdge T630",

input/grid5000/sites/sophia/clusters/esterel3/esterel3_metrics.yaml

 ---
 metrics:
+
   - name: prom_default_metrics
     description: Default subset of metrics from Prometheus Node Exporter
     period: 0
@@ -31,6 +32,7 @@ metrics:
         - node_procs_blocked
         - node_procs_running
         - kwollect_custom
+
   - name: prom_all_metrics
     description: All metrics from Prometheus Node Exporter
     period: 0
@@ -38,3 +40,25 @@ metrics:
     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
\ No newline at end of file