From 324ed4e20811e464bea2aad46484e0194f0fa4d0 Mon Sep 17 00:00:00 2001
From: BERARD Benjamin <benjamin.berard@inria.fr>
Date: Tue, 19 Oct 2021 19:09:41 +0200
Subject: [PATCH] [nancy/gruss] Add kwollect information
---
.../sites/nancy/clusters/gruss/gruss.json | 663 +++++++++++++++++-
.../nancy/network_equipments/sgruss.json | 74 ++
.../nancy/clusters/gruss/gruss_metrics.yaml | 430 ++++++++++++
.../grid5000/sites/nancy/networks/sgruss.yaml | 49 ++
4 files changed, 1215 insertions(+), 1 deletion(-)
create mode 100644 input/grid5000/sites/nancy/clusters/gruss/gruss_metrics.yaml
diff --git a/data/grid5000/sites/nancy/clusters/gruss/gruss.json b/data/grid5000/sites/nancy/clusters/gruss/gruss.json
index a75e3ae0ad6..961b37e8324 100644
--- a/data/grid5000/sites/nancy/clusters/gruss/gruss.json
+++ b/data/grid5000/sites/nancy/clusters/gruss/gruss.json
@@ -3,7 +3,668 @@
"exotic": false,
"kavlan": true,
"metrics": [
-
+ {
+ "description": "Temperature of Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 1,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 2,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature reported by BMC, in celsius",
+ "name": "bmc_ambient_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 3,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan1A reported by BMC, in rpm",
+ "labels": {
+ "id": "fan1a"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 4,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan2A reported by BMC, in rpm",
+ "labels": {
+ "id": "fan2a"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 5,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan3A reported by BMC, in rpm",
+ "labels": {
+ "id": "fan3a"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 6,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan4A reported by BMC, in rpm",
+ "labels": {
+ "id": "fan4a"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 7,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan5A reported by BMC, in rpm",
+ "labels": {
+ "id": "fan5a"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 8,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan6A reported by BMC, in rpm",
+ "labels": {
+ "id": "fan6a"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 9,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan1B reported by BMC, in rpm",
+ "labels": {
+ "id": "fan1b"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 10,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan2B reported by BMC, in rpm",
+ "labels": {
+ "id": "fan2b"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 11,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan3B reported by BMC, in rpm",
+ "labels": {
+ "id": "fan3b"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 12,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan4B reported by BMC, in rpm",
+ "labels": {
+ "id": "fan4b"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 13,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan5B reported by BMC, in rpm",
+ "labels": {
+ "id": "fan5b"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 14,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan6B reported by BMC, in rpm",
+ "labels": {
+ "id": "fan6b"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 15,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Current of Current 1 reported by BMC, in amp",
+ "labels": {
+ "id": "current1"
+ },
+ "name": "bmc_other_current_amp",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 33,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Current of Current 2 reported by BMC, in amp",
+ "labels": {
+ "id": "current2"
+ },
+ "name": "bmc_other_current_amp",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 34,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Voltage of Voltage 1 reported by BMC, in volt",
+ "labels": {
+ "id": "voltage1"
+ },
+ "name": "bmc_other_voltage_volt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 35,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Voltage of Voltage 2 reported by BMC, in volt",
+ "labels": {
+ "id": "voltage2"
+ },
+ "name": "bmc_other_voltage_volt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 36,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Power consumption of Pwr Consumption reported by BMC, in watt",
+ "labels": {
+ "id": "pwrconsumption"
+ },
+ "name": "bmc_other_power_watt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 43,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Usage of CPU Usage reported by BMC, in percent",
+ "labels": {
+ "id": "cpuusage"
+ },
+ "name": "bmc_other_usage_percent",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 49,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU1 Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "gpu1temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 161,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU2 Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "gpu2temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 162,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU3 Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "gpu3temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 163,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU4 Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "gpu4temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 164,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU5 Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "gpu5temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 165,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU6 Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "gpu6temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 166,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU7 Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "gpu7temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 167,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature of GPU8 Temp reported by BMC, in celsius",
+ "labels": {
+ "id": "gpu8temp"
+ },
+ "name": "bmc_other_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 168,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Temperature reported by BMC, in celsius",
+ "name": "bmc_exhaust_temp_celsius",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 173,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Voltage of VCORE VR reported by BMC, in volt",
+ "labels": {
+ "id": "vcorevr"
+ },
+ "name": "bmc_other_voltage_volt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 225,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Voltage of VCORE VR reported by BMC, in volt",
+ "labels": {
+ "id": "vcorevr"
+ },
+ "name": "bmc_other_voltage_volt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 226,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Voltage of MEMABCD VR reported by BMC, in volt",
+ "labels": {
+ "id": "memabcdvr"
+ },
+ "name": "bmc_other_voltage_volt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 227,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Voltage of MEMEFGH VR reported by BMC, in volt",
+ "labels": {
+ "id": "memefghvr"
+ },
+ "name": "bmc_other_voltage_volt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 228,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Voltage of MEMABCD VR reported by BMC, in volt",
+ "labels": {
+ "id": "memabcdvr"
+ },
+ "name": "bmc_other_voltage_volt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 229,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Voltage of MEMEFGH VR reported by BMC, in volt",
+ "labels": {
+ "id": "memefghvr"
+ },
+ "name": "bmc_other_voltage_volt",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 230,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan7A reported by BMC, in rpm",
+ "labels": {
+ "id": "fan7a"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 256,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan8A reported by BMC, in rpm",
+ "labels": {
+ "id": "fan8a"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 257,
+ "protocol": "ipmisensor"
+ }
+ },
+ {
+ "description": "Speed of Fan9A reported by BMC, in rpm",
+ "labels": {
+ "id": "fan9a"
+ },
+ "name": "bmc_other_speed_rpm",
+ "optional_period": 5000,
+ "period": 0,
+ "source": {
+ "id": 258,
+ "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": {
+ "interface": "eth2"
+ },
+ "name": "network_ifacein_bytes_total",
+ "period": 15000,
+ "source": {
+ "protocol": "network_equipment"
+ }
+ },
+ {
+ "description": "Output byte counter for the network device port",
+ "labels": {
+ "interface": "eth2"
+ },
+ "name": "network_ifaceout_bytes_total",
+ "period": 15000,
+ "source": {
+ "protocol": "network_equipment"
+ }
+ },
+ {
+ "description": "Input packet counter for the network device port",
+ "labels": {
+ "interface": "eth2"
+ },
+ "name": "network_ifacein_packets_total",
+ "period": 15000,
+ "source": {
+ "protocol": "network_equipment"
+ }
+ },
+ {
+ "description": "Output packet counter for the network device port",
+ "labels": {
+ "interface": "eth2"
+ },
+ "name": "network_ifaceout_packets_total",
+ "period": 15000,
+ "source": {
+ "protocol": "network_equipment"
+ }
+ },
+ {
+ "description": "Input counter of discarded packets for the network device port",
+ "labels": {
+ "interface": "eth2"
+ },
+ "name": "network_ifacein_packets_discard_total",
+ "period": 15000,
+ "source": {
+ "protocol": "network_equipment"
+ }
+ },
+ {
+ "description": "Output counter of discarded packets for the network device port",
+ "labels": {
+ "interface": "eth2"
+ },
+ "name": "network_ifaceout_packets_discard_total",
+ "period": 15000,
+ "source": {
+ "protocol": "network_equipment"
+ }
+ },
+ {
+ "description": "Input counter of packet errors for the network device port",
+ "labels": {
+ "interface": "eth2"
+ },
+ "name": "network_ifacein_packets_error_total",
+ "period": 15000,
+ "source": {
+ "protocol": "network_equipment"
+ }
+ },
+ {
+ "description": "Output counter of packet errors for the network device port",
+ "labels": {
+ "interface": "eth2"
+ },
+ "name": "network_ifaceout_packets_error_total",
+ "period": 15000,
+ "source": {
+ "protocol": "network_equipment"
+ }
+ }
],
"model": "Dell PowerEdge R7525",
"queues": [
diff --git a/data/grid5000/sites/nancy/network_equipments/sgruss.json b/data/grid5000/sites/nancy/network_equipments/sgruss.json
index 03d3b42ebec..e9acf6fe0d5 100644
--- a/data/grid5000/sites/nancy/network_equipments/sgruss.json
+++ b/data/grid5000/sites/nancy/network_equipments/sgruss.json
@@ -100,6 +100,80 @@
"snmp_pattern": "ethernet1/%LINECARD%/%PORT%"
}
],
+ "metrics": [
+ {
+ "description": "Input byte counter for the network device port",
+ "name": "network_ifacein_bytes_total",
+ "period": 15000,
+ "source": {
+ "id": "1.3.6.1.2.1.2.2.1.10.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}",
+ "protocol": "snmp"
+ }
+ },
+ {
+ "description": "Output byte counter for the network device port",
+ "name": "network_ifaceout_bytes_total",
+ "period": 15000,
+ "source": {
+ "id": "1.3.6.1.2.1.2.2.1.16.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}",
+ "protocol": "snmp"
+ }
+ },
+ {
+ "description": "Input packet counter for the network device port",
+ "name": "network_ifacein_packets_total",
+ "period": 15000,
+ "source": {
+ "id": "1.3.6.1.2.1.2.2.1.11.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}",
+ "protocol": "snmp"
+ }
+ },
+ {
+ "description": "Output packet counter for the network device port",
+ "name": "network_ifaceout_packets_total",
+ "period": 15000,
+ "source": {
+ "id": "1.3.6.1.2.1.2.2.1.17.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}",
+ "protocol": "snmp"
+ }
+ },
+ {
+ "description": "Input counter of discarded packets for the network device port",
+ "name": "network_ifacein_packets_discard_total",
+ "period": 15000,
+ "source": {
+ "id": "1.3.6.1.2.1.2.2.1.13.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}",
+ "protocol": "snmp"
+ }
+ },
+ {
+ "description": "Output counter of discarded packets for the network device port",
+ "name": "network_ifaceout_packets_discard_total",
+ "period": 15000,
+ "source": {
+ "id": "1.3.6.1.2.1.2.2.1.19.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}",
+ "protocol": "snmp"
+ }
+ },
+ {
+ "description": "Input counter of packet errors for the network device port",
+ "name": "network_ifacein_packets_error_total",
+ "period": 15000,
+ "source": {
+ "id": "1.3.6.1.2.1.2.2.1.14.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}",
+ "protocol": "snmp"
+ }
+ },
+ {
+ "description": "Output counter of packet errors for the network device port",
+ "name": "network_ifaceout_packets_error_total",
+ "period": 15000,
+ "source": {
+ "id": "1.3.6.1.2.1.2.2.1.20.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}",
+ "protocol": "snmp"
+ }
+ }
+ ],
"model": "Dell S5224F-ON",
"site": "nancy",
"snmp_community": "public",
diff --git a/input/grid5000/sites/nancy/clusters/gruss/gruss_metrics.yaml b/input/grid5000/sites/nancy/clusters/gruss/gruss_metrics.yaml
new file mode 100644
index 00000000000..c0e6f07ad03
--- /dev/null
+++ b/input/grid5000/sites/nancy/clusters/gruss/gruss_metrics.yaml
@@ -0,0 +1,430 @@
+---
+metrics:
+
+ # - name: bmc_ambient_temp_celsius
+ # description: Front node ambient temperature reported by BMC, in celsius
+ # period: 5000
+ # source:
+ # protocol: ipmisensor
+ # id: FIXME #Inlet Temp
+
+ #- name: bmc_node_power_watt
+ # description: Power consumption of node reported by BMC, in watt
+ # period: 5000
+ # source:
+ # protocol: ipmisensor
+ # id: FIXME #Pwr Consumption
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "temp"}
+ description: Temperature of Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 1 #Temp
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "temp"}
+ description: Temperature of Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 2 #Temp
+
+ - name: bmc_ambient_temp_celsius
+ description: Temperature reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 3 #Inlet Temp
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan1a"}
+ description: Speed of Fan1A reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 4 #Fan1A
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan2a"}
+ description: Speed of Fan2A reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 5 #Fan2A
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan3a"}
+ description: Speed of Fan3A reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 6 #Fan3A
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan4a"}
+ description: Speed of Fan4A reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 7 #Fan4A
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan5a"}
+ description: Speed of Fan5A reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 8 #Fan5A
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan6a"}
+ description: Speed of Fan6A reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 9 #Fan6A
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan1b"}
+ description: Speed of Fan1B reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 10 #Fan1B
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan2b"}
+ description: Speed of Fan2B reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 11 #Fan2B
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan3b"}
+ description: Speed of Fan3B reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 12 #Fan3B
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan4b"}
+ description: Speed of Fan4B reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 13 #Fan4B
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan5b"}
+ description: Speed of Fan5B reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 14 #Fan5B
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan6b"}
+ description: Speed of Fan6B reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 15 #Fan6B
+
+ - name: bmc_other_current_amp
+ labels: {"id": "current1"}
+ description: Current of Current 1 reported by BMC, in amp
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 33 #Current 1
+
+ - name: bmc_other_current_amp
+ labels: {"id": "current2"}
+ description: Current of Current 2 reported by BMC, in amp
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 34 #Current 2
+
+ - name: bmc_other_voltage_volt
+ labels: {"id": "voltage1"}
+ description: Voltage of Voltage 1 reported by BMC, in volt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 35 #Voltage 1
+
+ - name: bmc_other_voltage_volt
+ labels: {"id": "voltage2"}
+ description: Voltage of Voltage 2 reported by BMC, in volt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 36 #Voltage 2
+
+ - name: bmc_other_power_watt
+ labels: {"id": "pwrconsumption"}
+ description: Power consumption of Pwr Consumption reported by BMC, in watt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 43 #Pwr Consumption
+
+ - name: bmc_other_usage_percent
+ labels: {"id": "cpuusage"}
+ description: Usage of CPU Usage reported by BMC, in percent
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 49 #CPU Usage
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "gpu1temp"}
+ description: Temperature of GPU1 Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 161 #GPU1 Temp
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "gpu2temp"}
+ description: Temperature of GPU2 Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 162 #GPU2 Temp
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "gpu3temp"}
+ description: Temperature of GPU3 Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 163 #GPU3 Temp
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "gpu4temp"}
+ description: Temperature of GPU4 Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 164 #GPU4 Temp
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "gpu5temp"}
+ description: Temperature of GPU5 Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 165 #GPU5 Temp
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "gpu6temp"}
+ description: Temperature of GPU6 Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 166 #GPU6 Temp
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "gpu7temp"}
+ description: Temperature of GPU7 Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 167 #GPU7 Temp
+
+ - name: bmc_other_temp_celsius
+ labels: {"id": "gpu8temp"}
+ description: Temperature of GPU8 Temp reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 168 #GPU8 Temp
+
+ - name: bmc_exhaust_temp_celsius
+ description: Temperature reported by BMC, in celsius
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 173 #Exhaust Temp
+
+ - name: bmc_other_voltage_volt
+ labels: {"id": "vcorevr"}
+ description: Voltage of VCORE VR reported by BMC, in volt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 225 #VCORE VR
+
+ - name: bmc_other_voltage_volt
+ labels: {"id": "vcorevr"}
+ description: Voltage of VCORE VR reported by BMC, in volt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 226 #VCORE VR
+
+ - name: bmc_other_voltage_volt
+ labels: {"id": "memabcdvr"}
+ description: Voltage of MEMABCD VR reported by BMC, in volt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 227 #MEMABCD VR
+
+ - name: bmc_other_voltage_volt
+ labels: {"id": "memefghvr"}
+ description: Voltage of MEMEFGH VR reported by BMC, in volt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 228 #MEMEFGH VR
+
+ - name: bmc_other_voltage_volt
+ labels: {"id": "memabcdvr"}
+ description: Voltage of MEMABCD VR reported by BMC, in volt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 229 #MEMABCD VR
+
+ - name: bmc_other_voltage_volt
+ labels: {"id": "memefghvr"}
+ description: Voltage of MEMEFGH VR reported by BMC, in volt
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 230 #MEMEFGH VR
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan7a"}
+ description: Speed of Fan7A reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 256 #Fan7A
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan8a"}
+ description: Speed of Fan8A reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 257 #Fan8A
+
+ - name: bmc_other_speed_rpm
+ labels: {"id": "fan9a"}
+ description: Speed of Fan9A reported by BMC, in rpm
+ period: 0
+ optional_period: 5000
+ source:
+ protocol: ipmisensor
+ id: 258 #Fan9A
+
+ - 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
+
+
diff --git a/input/grid5000/sites/nancy/networks/sgruss.yaml b/input/grid5000/sites/nancy/networks/sgruss.yaml
index 3cb02bb79ca..2c1762061d5 100644
--- a/input/grid5000/sites/nancy/networks/sgruss.yaml
+++ b/input/grid5000/sites/nancy/networks/sgruss.yaml
@@ -5,6 +5,55 @@ sgruss:
kind: switch
site: nancy
snmp_community: public
+ metrics:
+ - name: network_ifacein_bytes_total
+ description: Input byte counter for the network device port
+ period: 15000
+ source:
+ protocol: snmp
+ id: 1.3.6.1.2.1.2.2.1.10.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}
+ - name: network_ifaceout_bytes_total
+ description: Output byte counter for the network device port
+ period: 15000
+ source:
+ protocol: snmp
+ id: 1.3.6.1.2.1.2.2.1.16.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}
+ - name: network_ifacein_packets_total
+ description: Input packet counter for the network device port
+ period: 15000
+ source:
+ protocol: snmp
+ id: 1.3.6.1.2.1.2.2.1.11.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}
+ - name: network_ifaceout_packets_total
+ description: Output packet counter for the network device port
+ period: 15000
+ source:
+ protocol: snmp
+ id: 1.3.6.1.2.1.2.2.1.17.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}
+ - name: network_ifacein_packets_discard_total
+ description: Input counter of discarded packets for the network device port
+ period: 15000
+ source:
+ protocol: snmp
+ id: 1.3.6.1.2.1.2.2.1.13.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}
+ - name: network_ifaceout_packets_discard_total
+ description: Output counter of discarded packets for the network device port
+ period: 15000
+ source:
+ protocol: snmp
+ id: 1.3.6.1.2.1.2.2.1.19.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}
+ - name: network_ifacein_packets_error_total
+ description: Input counter of packet errors for the network device port
+ period: 15000
+ source:
+ protocol: snmp
+ id: 1.3.6.1.2.1.2.2.1.14.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}
+ - name: network_ifaceout_packets_error_total
+ description: Output counter of packet errors for the network device port
+ period: 15000
+ source:
+ protocol: snmp
+ id: 1.3.6.1.2.1.2.2.1.20.{{ 1.3.6.1.2.1.2.2.1.2 == %SNMP_IFACE% }}
ip: 172.17.79.212
backplane_bps: 2160000000000
linecards:
--
GitLab