From dbf02d5f83d903df174b7ff32ea35f50deaa8bb0 Mon Sep 17 00:00:00 2001
From: Baptiste Jonglez <baptiste.jonglez@imag.fr>
Date: Tue, 12 Jan 2021 16:57:28 +0100
Subject: [PATCH] [grenoble] [drac] Add kwollect metrics
---
.../sites/grenoble/clusters/drac/drac.json | 661 ++++++++++++++++++
.../grenoble/clusters/drac/drac_metrics.yaml | 550 +++++++++++++++
2 files changed, 1211 insertions(+)
create mode 100644 input/grid5000/sites/grenoble/clusters/drac/drac_metrics.yaml
diff --git a/data/grid5000/sites/grenoble/clusters/drac/drac.json b/data/grid5000/sites/grenoble/clusters/drac/drac.json
index 6d952ecc58..7f9957a1c5 100644
--- a/data/grid5000/sites/grenoble/clusters/drac/drac.json
+++ b/data/grid5000/sites/grenoble/clusters/drac/drac.json
@@ -3,6 +3,667 @@
"exotic": true,
"kavlan": true,
"metrics": [
+ {
+ "description": "Front node ambiant temperature reported by BMC, in celsius",
+ "name": "bmc_temp_ambient_celsius",
+ "period": 5000,
+ "source": {
+ "id": 175,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of node reported by BMC, in watt",
+ "name": "bmc_node_power_watt",
+ "period": 5000,
+ "source": {
+ "id": 212,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of CPU Diode 1 reported by BMC, in celsius",
+ "name": "bmc_temp_cpudiode1_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 2,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of CPU Diode 2 reported by BMC, in celsius",
+ "name": "bmc_temp_cpudiode2_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 3,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of CPU1 reported by BMC, in celsius",
+ "name": "bmc_temp_cpu1_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 61,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of CPU2 reported by BMC, in celsius",
+ "name": "bmc_temp_cpu2_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 62,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM1 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm1_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 63,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM2 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm2_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 64,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM3 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm3_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 65,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM4 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm4_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 66,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM5 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm5_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 68,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM6 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm6_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 69,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM7 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm7_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 70,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM8 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm8_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 71,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM9 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm9_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 73,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM10 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm10_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 74,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM11 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm11_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 75,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM12 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm12_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 76,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM13 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm13_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 78,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM14 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm14_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 79,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM15 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm15_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 80,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM16 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm16_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 81,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM17 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm17_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 83,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM18 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm18_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 84,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM19 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm19_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 85,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM20 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm20_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 86,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM21 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm21_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 88,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM22 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm22_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 89,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM23 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm23_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 90,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM24 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm24_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 91,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM25 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm25_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 93,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM26 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm26_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 94,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM27 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm27_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 95,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM28 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm28_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 96,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM29 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm29_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 98,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM30 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm30_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 99,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM31 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm31_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 100,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of DIMM32 reported by BMC, in celsius",
+ "name": "bmc_temp_dimm32_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 101,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of Fan reported by BMC, in watt",
+ "name": "bmc_power_fan_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 103,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of Mem Proc0 reported by BMC, in watt",
+ "name": "bmc_power_memproc0_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 104,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of Mem Proc1 reported by BMC, in watt",
+ "name": "bmc_power_memproc1_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 105,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of PCIE Proc0 reported by BMC, in watt",
+ "name": "bmc_power_pcieproc0_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 106,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of Mem Cache reported by BMC, in watt",
+ "name": "bmc_power_memcache_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 107,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of Proc0 reported by BMC, in watt",
+ "name": "bmc_power_proc0_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 108,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Voltage of CPU VDD reported by BMC, in volt",
+ "name": "bmc_voltage_cpuvdd_volt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 207,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Current of CPU VDD reported by BMC, in amp",
+ "name": "bmc_current_cpuvdd_amp",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 208,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of Proc1 reported by BMC, in watt",
+ "name": "bmc_power_proc1_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 210,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of PCIE Proc1 reported by BMC, in watt",
+ "name": "bmc_power_pcieproc1_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 211,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of GPU reported by BMC, in watt",
+ "name": "bmc_power_gpu_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 213,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU 1 reported by BMC, in celsius",
+ "name": "bmc_temp_gpu1_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 216,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU 2 reported by BMC, in celsius",
+ "name": "bmc_temp_gpu2_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 217,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU 3 reported by BMC, in celsius",
+ "name": "bmc_temp_gpu3_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 218,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU 4 reported by BMC, in celsius",
+ "name": "bmc_temp_gpu4_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 219,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan 1 reported by BMC, in rpm",
+ "name": "bmc_fanspeed_fan1_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 227,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan 2 reported by BMC, in rpm",
+ "name": "bmc_fanspeed_fan2_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 228,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan 3 reported by BMC, in rpm",
+ "name": "bmc_fanspeed_fan3_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 229,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan 4 reported by BMC, in rpm",
+ "name": "bmc_fanspeed_fan4_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 230,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of CPU 1 VDD reported by BMC, in celsius",
+ "name": "bmc_temp_cpu1vdd_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 251,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of CPU 2 VDD reported by BMC, in celsius",
+ "name": "bmc_temp_cpu2vdd_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 252,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "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_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_GPU_UTIL",
+ "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"
+ }
+ },
{
"description": "Input byte counter for the network device port",
"labels": {
diff --git a/input/grid5000/sites/grenoble/clusters/drac/drac_metrics.yaml b/input/grid5000/sites/grenoble/clusters/drac/drac_metrics.yaml
new file mode 100644
index 0000000000..c8b5e9b9c2
--- /dev/null
+++ b/input/grid5000/sites/grenoble/clusters/drac/drac_metrics.yaml
@@ -0,0 +1,550 @@
+---
+metrics:
+
+ - name: network_ifacein_bytes_total
+ description: >-
+ Input byte counter for the network device port connected to the node
+ period: 1000
+ source:
+ protocol: network_equipment
+
+ - name: network_ifaceout_bytes_total
+ description: >-
+ Output byte counter for the network device port connected to the node
+ period: 1000
+ source:
+ protocol: network_equipment
+
+ - name: bmc_temp_ambient_celsius
+ description: >-
+ Front node ambiant temperature reported by BMC, in celsius
+ period: 5000
+ source:
+ protocol: ipmisensor
+ id: 175 #Ambient Temp
+
+ - name: bmc_node_power_watt
+ description: Power consumption of node reported by BMC, in watt
+ period: 5000
+ source:
+ protocol: ipmisensor
+ id: 212 #System Power
+
+ - name: bmc_temp_cpudiode1_celsius
+ description: Temperature of CPU Diode 1 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 2 #CPU Diode 1
+
+ - name: bmc_temp_cpudiode2_celsius
+ description: Temperature of CPU Diode 2 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 3 #CPU Diode 2
+
+ - name: bmc_temp_cpu1_celsius
+ description: Temperature of CPU1 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 61 #CPU1 Temp
+
+ - name: bmc_temp_cpu2_celsius
+ description: Temperature of CPU2 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 62 #CPU2 Temp
+
+ - name: bmc_temp_dimm1_celsius
+ description: Temperature of DIMM1 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 63 #DIMM1 Temp
+
+ - name: bmc_temp_dimm2_celsius
+ description: Temperature of DIMM2 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 64 #DIMM2 Temp
+
+ - name: bmc_temp_dimm3_celsius
+ description: Temperature of DIMM3 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 65 #DIMM3 Temp
+
+ - name: bmc_temp_dimm4_celsius
+ description: Temperature of DIMM4 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 66 #DIMM4 Temp
+
+ - name: bmc_temp_dimm5_celsius
+ description: Temperature of DIMM5 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 68 #DIMM5 Temp
+
+ - name: bmc_temp_dimm6_celsius
+ description: Temperature of DIMM6 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 69 #DIMM6 Temp
+
+ - name: bmc_temp_dimm7_celsius
+ description: Temperature of DIMM7 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 70 #DIMM7 Temp
+
+ - name: bmc_temp_dimm8_celsius
+ description: Temperature of DIMM8 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 71 #DIMM8 Temp
+
+ - name: bmc_temp_dimm9_celsius
+ description: Temperature of DIMM9 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 73 #DIMM9 Temp
+
+ - name: bmc_temp_dimm10_celsius
+ description: Temperature of DIMM10 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 74 #DIMM10 Temp
+
+ - name: bmc_temp_dimm11_celsius
+ description: Temperature of DIMM11 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 75 #DIMM11 Temp
+
+ - name: bmc_temp_dimm12_celsius
+ description: Temperature of DIMM12 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 76 #DIMM12 Temp
+
+ - name: bmc_temp_dimm13_celsius
+ description: Temperature of DIMM13 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 78 #DIMM13 Temp
+
+ - name: bmc_temp_dimm14_celsius
+ description: Temperature of DIMM14 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 79 #DIMM14 Temp
+
+ - name: bmc_temp_dimm15_celsius
+ description: Temperature of DIMM15 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 80 #DIMM15 Temp
+
+ - name: bmc_temp_dimm16_celsius
+ description: Temperature of DIMM16 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 81 #DIMM16 Temp
+
+ - name: bmc_temp_dimm17_celsius
+ description: Temperature of DIMM17 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 83 #DIMM17 Temp
+
+ - name: bmc_temp_dimm18_celsius
+ description: Temperature of DIMM18 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 84 #DIMM18 Temp
+
+ - name: bmc_temp_dimm19_celsius
+ description: Temperature of DIMM19 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 85 #DIMM19 Temp
+
+ - name: bmc_temp_dimm20_celsius
+ description: Temperature of DIMM20 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 86 #DIMM20 Temp
+
+ - name: bmc_temp_dimm21_celsius
+ description: Temperature of DIMM21 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 88 #DIMM21 Temp
+
+ - name: bmc_temp_dimm22_celsius
+ description: Temperature of DIMM22 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 89 #DIMM22 Temp
+
+ - name: bmc_temp_dimm23_celsius
+ description: Temperature of DIMM23 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 90 #DIMM23 Temp
+
+ - name: bmc_temp_dimm24_celsius
+ description: Temperature of DIMM24 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 91 #DIMM24 Temp
+
+ - name: bmc_temp_dimm25_celsius
+ description: Temperature of DIMM25 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 93 #DIMM25 Temp
+
+ - name: bmc_temp_dimm26_celsius
+ description: Temperature of DIMM26 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 94 #DIMM26 Temp
+
+ - name: bmc_temp_dimm27_celsius
+ description: Temperature of DIMM27 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 95 #DIMM27 Temp
+
+ - name: bmc_temp_dimm28_celsius
+ description: Temperature of DIMM28 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 96 #DIMM28 Temp
+
+ - name: bmc_temp_dimm29_celsius
+ description: Temperature of DIMM29 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 98 #DIMM29 Temp
+
+ - name: bmc_temp_dimm30_celsius
+ description: Temperature of DIMM30 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 99 #DIMM30 Temp
+
+ - name: bmc_temp_dimm31_celsius
+ description: Temperature of DIMM31 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 100 #DIMM31 Temp
+
+ - name: bmc_temp_dimm32_celsius
+ description: Temperature of DIMM32 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 101 #DIMM32 Temp
+
+ - name: bmc_power_fan_watt
+ description: Power consumption of Fan reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 103 #Fan Power
+
+ - name: bmc_power_memproc0_watt
+ description: Power consumption of Mem Proc0 reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 104 #Mem Proc0 Pwr
+
+ - name: bmc_power_memproc1_watt
+ description: Power consumption of Mem Proc1 reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 105 #Mem Proc1 Pwr
+
+ - name: bmc_power_pcieproc0_watt
+ description: Power consumption of PCIE Proc0 reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 106 #PCIE Proc0 Pwr
+
+ - name: bmc_power_memcache_watt
+ description: Power consumption of Mem Cache reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 107 #Mem Cache Power
+
+ - name: bmc_power_proc0_watt
+ description: Power consumption of Proc0 reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 108 #Proc0 Power
+
+ - name: bmc_voltage_cpuvdd_volt
+ description: Voltage of CPU VDD reported by BMC, in volt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 207 #CPU VDD Volt
+
+ - name: bmc_current_cpuvdd_amp
+ description: Current of CPU VDD reported by BMC, in amp
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 208 #CPU VDD Curr
+
+ - name: bmc_power_proc1_watt
+ description: Power consumption of Proc1 reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 210 #Proc1 Power
+
+ - name: bmc_power_pcieproc1_watt
+ description: Power consumption of PCIE Proc1 reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 211 #PCIE Proc1 Power
+
+ - name: bmc_power_gpu_watt
+ description: Power consumption of GPU reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 213 #GPU Power
+
+ - name: bmc_temp_gpu1_celsius
+ description: Temperature of GPU 1 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 216 #GPU Temp 1
+
+ - name: bmc_temp_gpu2_celsius
+ description: Temperature of GPU 2 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 217 #GPU Temp 2
+
+ - name: bmc_temp_gpu3_celsius
+ description: Temperature of GPU 3 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 218 #GPU Temp 3
+
+ - name: bmc_temp_gpu4_celsius
+ description: Temperature of GPU 4 reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 219 #GPU Temp 4
+
+ - name: bmc_fanspeed_fan1_rpm
+ description: Speed of Fan 1 reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 227 #Fan 1
+
+ - name: bmc_fanspeed_fan2_rpm
+ description: Speed of Fan 2 reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 228 #Fan 2
+
+ - name: bmc_fanspeed_fan3_rpm
+ description: Speed of Fan 3 reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 229 #Fan 3
+
+ - name: bmc_fanspeed_fan4_rpm
+ description: Speed of Fan 4 reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 230 #Fan 4
+
+ - name: bmc_temp_cpu1vdd_celsius
+ description: Temperature of CPU 1 VDD reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 251 #CPU 1 VDD Temp
+
+ - name: bmc_temp_cpu2vdd_celsius
+ description: Temperature of CPU 2 VDD reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 252 #CPU 2 VDD Temp
+
+ - 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_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_GPU_UTIL
+ - 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
--
GitLab