[nancy][gratouille] add metrics

data/grid5000/sites/nancy/clusters/gratouille/gratouille.json

@@ -5,7 +5,278 @@
   "kavlan": false,
   "manufactured_at": "2017-06-07",
   "metrics": [
-
+    {
+      "description": "Front node ambient temperature reported by BMC, in celsius",
+      "name": "bmc_ambient_temp_celsius",
+      "period": 5000,
+      "scale_factor": 0.1,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.1",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Power consumption of node reported by BMC, in watt",
+      "name": "bmc_node_power_watt",
+      "period": 5000,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.600.30.1.6.1.{{ 1.3.6.1.4.1.674.10892.5.4.600.30.1.8.1 == System Board Pwr Consumption }}",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Voltage of PSU 1 reported by BMC, in volt",
+      "labels": {
+        "psu": "1"
+      },
+      "name": "bmc_psu_voltage_volt",
+      "optional_period": 5000,
+      "period": 0,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.600.12.1.16.1.1",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Voltage of PSU 2 reported by BMC, in volt",
+      "labels": {
+        "psu": "2"
+      },
+      "name": "bmc_psu_voltage_volt",
+      "optional_period": 5000,
+      "period": 0,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.600.12.1.16.1.2",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Current of PSU 1 reported by BMC, in amp",
+      "labels": {
+        "psu": "1"
+      },
+      "name": "bmc_psu_current_amp",
+      "optional_period": 5000,
+      "period": 0,
+      "scale_factor": 0.1,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.600.30.1.6.1.{{ 1.3.6.1.4.1.674.10892.5.4.600.30.1.8.1 == PS1 Current 1 }}",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Current of PSU 2 reported by BMC, in amp",
+      "labels": {
+        "psu": "2"
+      },
+      "name": "bmc_psu_current_amp",
+      "optional_period": 5000,
+      "period": 0,
+      "scale_factor": 0.1,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.600.30.1.6.1.{{ 1.3.6.1.4.1.674.10892.5.4.600.30.1.8.1 == PS2 Current 2 }}",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Cumulated power consumption of node reported by BMC, in watt-hour",
+      "name": "bmc_node_power_watthour_total",
+      "optional_period": 5000,
+      "period": 0,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.600.60.1.7.1.1",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Speed of Fan 1 reported by BMC, in rpm",
+      "labels": {
+        "fan": "1"
+      },
+      "name": "bmc_fan_speed_rpm",
+      "optional_period": 5000,
+      "period": 0,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.1",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Speed of Fan 2 reported by BMC, in rpm",
+      "labels": {
+        "fan": "2"
+      },
+      "name": "bmc_fan_speed_rpm",
+      "optional_period": 5000,
+      "period": 0,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.2",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Speed of Fan 3 reported by BMC, in rpm",
+      "labels": {
+        "fan": "3"
+      },
+      "name": "bmc_fan_speed_rpm",
+      "optional_period": 5000,
+      "period": 0,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.3",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Speed of Fan 4 reported by BMC, in rpm",
+      "labels": {
+        "fan": "4"
+      },
+      "name": "bmc_fan_speed_rpm",
+      "optional_period": 5000,
+      "period": 0,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.4",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Speed of Fan 5 reported by BMC, in rpm",
+      "labels": {
+        "fan": "5"
+      },
+      "name": "bmc_fan_speed_rpm",
+      "optional_period": 5000,
+      "period": 0,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.5",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Speed of Fan 6 reported by BMC, in rpm",
+      "labels": {
+        "fan": "6"
+      },
+      "name": "bmc_fan_speed_rpm",
+      "optional_period": 5000,
+      "period": 0,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.6",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Temperature of exhausted air reported by BMC, in celsius",
+      "name": "bmc_exhaust_temp_celsius",
+      "optional_period": 5000,
+      "period": 0,
+      "scale_factor": 0.1,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.2",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Temperature of CPU 1 reported by BMC, in celsius",
+      "labels": {
+        "cpu": "1"
+      },
+      "name": "bmc_cpu_temp_celsius",
+      "optional_period": 5000,
+      "period": 0,
+      "scale_factor": 0.1,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.3",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "description": "Temperature of CPU 2 reported by BMC, in celsius",
+      "labels": {
+        "cpu": "2"
+      },
+      "name": "bmc_cpu_temp_celsius",
+      "optional_period": 5000,
+      "period": 0,
+      "scale_factor": 0.1,
+      "source": {
+        "id": "1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.4",
+        "protocol": "snmp"
+      }
+    },
+    {
+      "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": "Dell PowerEdge R730",
   "priority": 201806,

input/grid5000/sites/nancy/clusters/gratouille/gratouille_metrics.yaml

+---
+metrics:
+
+  - name: bmc_ambient_temp_celsius
+    description: Front node ambient temperature reported by BMC, in celsius
+    period: 5000
+    scale_factor: 0.1
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.1
+
+  - name: bmc_node_power_watt
+    description: Power consumption of node reported by BMC, in watt
+    period: 5000
+    source:
+      protocol: snmp
+      id: >-
+        1.3.6.1.4.1.674.10892.5.4.600.30.1.6.1.{{
+        1.3.6.1.4.1.674.10892.5.4.600.30.1.8.1
+        == System Board Pwr Consumption }}
+
+  - name: bmc_psu_voltage_volt
+    labels: {"psu": "1"}
+    description: Voltage of PSU 1 reported by BMC, in volt
+    period: 0
+    optional_period: 5000
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.600.12.1.16.1.1
+
+  - name: bmc_psu_voltage_volt
+    labels: {"psu": "2"}
+    description: Voltage of PSU 2 reported by BMC, in volt
+    period: 0
+    optional_period: 5000
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.600.12.1.16.1.2
+
+  - name: bmc_psu_current_amp
+    labels: {"psu": "1"}
+    description: Current of PSU 1 reported by BMC, in amp
+    period: 0
+    optional_period: 5000
+    scale_factor: 0.1
+    source:
+      protocol: snmp
+      id: >-
+        1.3.6.1.4.1.674.10892.5.4.600.30.1.6.1.{{
+        1.3.6.1.4.1.674.10892.5.4.600.30.1.8.1 == PS1 Current 1 }}
+
+  - name: bmc_psu_current_amp
+    labels: {"psu": "2"}
+    description: Current of PSU 2 reported by BMC, in amp
+    period: 0
+    optional_period: 5000
+    scale_factor: 0.1
+    source:
+      protocol: snmp
+      id: >-
+        1.3.6.1.4.1.674.10892.5.4.600.30.1.6.1.{{
+        1.3.6.1.4.1.674.10892.5.4.600.30.1.8.1 == PS2 Current 2 }}
+
+  - name: bmc_node_power_watthour_total
+    description: Cumulated power consumption of node reported by BMC, in watt-hour
+    period: 0
+    optional_period: 5000
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.600.60.1.7.1.1
+
+  - name: bmc_fan_speed_rpm
+    labels: {"fan": "1"}
+    description: Speed of Fan 1 reported by BMC, in rpm
+    period: 0
+    optional_period: 5000
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.1
+
+  - name: bmc_fan_speed_rpm
+    labels: {"fan": "2"}
+    description: Speed of Fan 2 reported by BMC, in rpm
+    period: 0
+    optional_period: 5000
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.2
+
+  - name: bmc_fan_speed_rpm
+    labels: {"fan": "3"}
+    description: Speed of Fan 3 reported by BMC, in rpm
+    period: 0
+    optional_period: 5000
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.3
+
+  - name: bmc_fan_speed_rpm
+    labels: {"fan": "4"}
+    description: Speed of Fan 4 reported by BMC, in rpm
+    period: 0
+    optional_period: 5000
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.4
+
+  - name: bmc_fan_speed_rpm
+    labels: {"fan": "5"}
+    description: Speed of Fan 5 reported by BMC, in rpm
+    period: 0
+    optional_period: 5000
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.5
+
+  - name: bmc_fan_speed_rpm
+    labels: {"fan": "6"}
+    description: Speed of Fan 6 reported by BMC, in rpm
+    period: 0
+    optional_period: 5000
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.6
+
+  - name: bmc_exhaust_temp_celsius
+    description: Temperature of exhausted air reported by BMC, in celsius
+    period: 0
+    optional_period: 5000
+    scale_factor: 0.1
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.2
+
+  - name: bmc_cpu_temp_celsius
+    labels: {"cpu": "1"}
+    description: Temperature of CPU 1 reported by BMC, in celsius
+    period: 0
+    optional_period: 5000
+    scale_factor: 0.1
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.3
+
+  - name: bmc_cpu_temp_celsius
+    labels: {"cpu": "2"}
+    description: Temperature of CPU 2 reported by BMC, in celsius
+    period: 0
+    optional_period: 5000
+    scale_factor: 0.1
+    source:
+      protocol: snmp
+      id: 1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.4
+
+  - 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