diff --git a/data/grid5000/sites/rennes/clusters/roazhon9/roazhon9.json b/data/grid5000/sites/rennes/clusters/roazhon9/roazhon9.json index a76b26cda0c2e387b37c1b14bcdadbc6a9b0cd31..50a5553f1bbfc82f4bee0b1cc3d365f085aeaa8e 100644 --- a/data/grid5000/sites/rennes/clusters/roazhon9/roazhon9.json +++ b/data/grid5000/sites/rennes/clusters/roazhon9/roazhon9.json @@ -4,206 +4,6 @@ "exotic": false, "kavlan": false, "metrics": [ - { - "description": "Front node ambient temperature reported by BMC, in celsius", - "name": "bmc_ambient_temp_celsius", - "period": 5000, - "scale_factor": 0.1, - "source": { - "id": "1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.3", - "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 1A reported by BMC, in rpm", - "labels": { - "fan": "1a" - }, - "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 2A reported by BMC, in rpm", - "labels": { - "fan": "2a" - }, - "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 3A reported by BMC, in rpm", - "labels": { - "fan": "3a" - }, - "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 4A reported by BMC, in rpm", - "labels": { - "fan": "4a" - }, - "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 5A reported by BMC, in rpm", - "labels": { - "fan": "5a" - }, - "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 6A reported by BMC, in rpm", - "labels": { - "fan": "6a" - }, - "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 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": "1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.1", - "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": "1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.2", - "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": "1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.4", - "protocol": "snmp" - } - }, { "description": "Default subset of metrics from Prometheus Node Exporter", "name": "prom_default_metrics", diff --git a/input/grid5000/sites/rennes/clusters/roazhon9/roazhon9_metrics.yaml b/input/grid5000/sites/rennes/clusters/roazhon9/roazhon9_metrics.yaml index 890bcec45188681cec3cf2c5158839b86532a937..1b7527d9fcc26e8892353ae61510b49c2fe579ba 100644 --- a/input/grid5000/sites/rennes/clusters/roazhon9/roazhon9_metrics.yaml +++ b/input/grid5000/sites/rennes/clusters/roazhon9/roazhon9_metrics.yaml @@ -1,159 +1,5 @@ --- 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: 1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.3 - - - 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": "1a"} - description: Speed of Fan 1A 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": "2a"} - description: Speed of Fan 2A 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": "3a"} - description: Speed of Fan 3A 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": "4a"} - description: Speed of Fan 4A 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": "5a"} - description: Speed of Fan 5A 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": "6a"} - description: Speed of Fan 6A 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_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: 1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.1 - - - 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: 1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.2 - - - 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: 1.3.6.1.4.1.674.10892.5.4.700.20.1.6.1.4 - - name: prom_default_metrics description: Default subset of metrics from Prometheus Node Exporter period: 0 @@ -193,27 +39,3 @@ metrics: source: protocol: prometheus port: 9100 - - # Uncomment if cluster has GPU - # - 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 -