Mentions légales du service

Skip to content
Snippets Groups Projects
Verified Commit 7a6eb2b1 authored by Philippe Virouleau's avatar Philippe Virouleau
Browse files

[abacus16] Generate kwollect metrics

parent d9018a99
No related branches found
No related tags found
No related merge requests found
Pipeline #879687 passed with warnings
Stage: lint
Stage: validate
Stage: generate
Stage: deploy
Stage: checks
Stage: .post
Hide whitespace changes
Inline Side-by-side
data/grid5000/sites/rennes/clusters/abacus16/abacus16.json
+ 402
− 1
View file @ 7a6eb2b1
...@@ -4,7 +4,408 @@ ...@@ -4,7 +4,408 @@
"exotic": false, "exotic": false,
"kavlan": false, "kavlan": false,
"metrics": [ "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 1B reported by BMC, in rpm",
"labels": {
"fan": "1b"
},
"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 2A reported by BMC, in rpm",
"labels": {
"fan": "2a"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.3",
"protocol": "snmp"
}
},
{
"description": "Speed of Fan 2B reported by BMC, in rpm",
"labels": {
"fan": "2b"
},
"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 3A reported by BMC, in rpm",
"labels": {
"fan": "3a"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.5",
"protocol": "snmp"
}
},
{
"description": "Speed of Fan 3B reported by BMC, in rpm",
"labels": {
"fan": "3b"
},
"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": "Speed of Fan 4A reported by BMC, in rpm",
"labels": {
"fan": "4a"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.7",
"protocol": "snmp"
}
},
{
"description": "Speed of Fan 4B reported by BMC, in rpm",
"labels": {
"fan": "4b"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.8",
"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": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.9",
"protocol": "snmp"
}
},
{
"description": "Speed of Fan 5B reported by BMC, in rpm",
"labels": {
"fan": "5b"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.10",
"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": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.11",
"protocol": "snmp"
}
},
{
"description": "Speed of Fan 6B reported by BMC, in rpm",
"labels": {
"fan": "6b"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.12",
"protocol": "snmp"
}
},
{
"description": "Speed of Fan 7A reported by BMC, in rpm",
"labels": {
"fan": "7a"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.13",
"protocol": "snmp"
}
},
{
"description": "Speed of Fan 7B reported by BMC, in rpm",
"labels": {
"fan": "7b"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.14",
"protocol": "snmp"
}
},
{
"description": "Speed of Fan 8A reported by BMC, in rpm",
"labels": {
"fan": "8a"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.15",
"protocol": "snmp"
}
},
{
"description": "Speed of Fan 8B reported by BMC, in rpm",
"labels": {
"fan": "8b"
},
"name": "bmc_fan_speed_rpm",
"optional_period": 5000,
"period": 0,
"source": {
"id": "1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.16",
"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",
"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 C4140", "model": "Dell PowerEdge C4140",
"queues": [ "queues": [
......
input/grid5000/sites/rennes/clusters/abacus16/abacus16_metrics.yaml 0 → 100644
+ 308
− 0
View file @ 7a6eb2b1
---
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": "1b"}
description: Speed of Fan 1B 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": "2a"}
description: Speed of Fan 2A reported by BMC, in rpm
period: 0
optional_period: 5000
source:
protocol: snmp
id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.3
- name: bmc_fan_speed_rpm
labels: {"fan": "2b"}
description: Speed of Fan 2B 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": "3a"}
description: Speed of Fan 3A reported by BMC, in rpm
period: 0
optional_period: 5000
source:
protocol: snmp
id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.5
- name: bmc_fan_speed_rpm
labels: {"fan": "3b"}
description: Speed of Fan 3B 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_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: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.7
- name: bmc_fan_speed_rpm
labels: {"fan": "4b"}
description: Speed of Fan 4B reported by BMC, in rpm
period: 0
optional_period: 5000
source:
protocol: snmp
id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.8
- 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: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.9
- name: bmc_fan_speed_rpm
labels: {"fan": "5b"}
description: Speed of Fan 5B reported by BMC, in rpm
period: 0
optional_period: 5000
source:
protocol: snmp
id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.10
- 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: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.11
- name: bmc_fan_speed_rpm
labels: {"fan": "6b"}
description: Speed of Fan 6B reported by BMC, in rpm
period: 0
optional_period: 5000
source:
protocol: snmp
id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.12
- name: bmc_fan_speed_rpm
labels: {"fan": "7a"}
description: Speed of Fan 7A reported by BMC, in rpm
period: 0
optional_period: 5000
source:
protocol: snmp
id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.13
- name: bmc_fan_speed_rpm
labels: {"fan": "7b"}
description: Speed of Fan 7B reported by BMC, in rpm
period: 0
optional_period: 5000
source:
protocol: snmp
id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.14
- name: bmc_fan_speed_rpm
labels: {"fan": "8a"}
description: Speed of Fan 8A reported by BMC, in rpm
period: 0
optional_period: 5000
source:
protocol: snmp
id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.15
- name: bmc_fan_speed_rpm
labels: {"fan": "8b"}
description: Speed of Fan 8B reported by BMC, in rpm
period: 0
optional_period: 5000
source:
protocol: snmp
id: 1.3.6.1.4.1.674.10892.5.4.700.12.1.6.1.16
- 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
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
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment