diff --git a/data/grid5000/accesses/refrepo.json b/data/grid5000/accesses/refrepo.json index 8c5d56cbb1da0c900de3a75230892c640b1be405..a81903bf3451ebc52d326363cd8330ac6c3dd6df 100644 --- a/data/grid5000/accesses/refrepo.json +++ b/data/grid5000/accesses/refrepo.json @@ -45264,7 +45264,7 @@ "grdix": { "created_at": "Fri, 27 Sep 2024 00:00:00 GMT", "manufactured_at": "2024-09-02", - "model": "ProLiant DL385 Gen11", + "model": "ProLiant DL365 Gen11", "nodes": { "grdix-1": { "architecture": { diff --git a/data/grid5000/sites/nancy/clusters/grdix/grdix.json b/data/grid5000/sites/nancy/clusters/grdix/grdix.json index 0945e54c4df49de26219a59d482897d412fcfcfb..cf3ccd0ed04711ab35505559cc730050872a313f 100644 --- a/data/grid5000/sites/nancy/clusters/grdix/grdix.json +++ b/data/grid5000/sites/nancy/clusters/grdix/grdix.json @@ -5,9 +5,333 @@ "kavlan": true, "manufactured_at": "2024-09-02", "metrics": [ - + { + "description": "Power consumption of Psu 1 reported by BMC, in watt", + "labels": { + "psu": "1" + }, + "name": "bmc_psu_power_watt", + "optional_period": 5000, + "period": 0, + "source": { + "id": 51, + "protocol": "ipmisensor" + } + }, + { + "description": "Power consumption of Psu 2 reported by BMC, in watt", + "labels": { + "psu": "2" + }, + "name": "bmc_psu_power_watt", + "optional_period": 5000, + "period": 0, + "source": { + "id": 53, + "protocol": "ipmisensor" + } + }, + { + "description": "Power consumption reported by BMC, in watt", + "name": "bmc_node_power_watt", + "optional_period": 5000, + "period": 0, + "source": { + "id": 54, + "protocol": "ipmisensor" + } + }, + { + "description": "Power consumption of PS 1 Output reported by BMC, in watt", + "labels": { + "id": "ps1output" + }, + "name": "bmc_other_power_watt", + "optional_period": 5000, + "period": 0, + "source": { + "id": 62, + "protocol": "ipmisensor" + } + }, + { + "description": "Voltage of PS_Volt_Out_01 reported by BMC, in volt", + "labels": { + "id": "ps_volt_out_01" + }, + "name": "bmc_other_voltage_volt", + "optional_period": 5000, + "period": 0, + "source": { + "id": 63, + "protocol": "ipmisensor" + } + }, + { + "description": "Voltage of PS_Volt_In_01 reported by BMC, in volt", + "labels": { + "id": "ps_volt_in_01" + }, + "name": "bmc_other_voltage_volt", + "optional_period": 5000, + "period": 0, + "source": { + "id": 64, + "protocol": "ipmisensor" + } + }, + { + "description": "Current of PS_Curr_Out_01 reported by BMC, in amp", + "labels": { + "id": "ps_curr_out_01" + }, + "name": "bmc_other_current_amp", + "optional_period": 5000, + "period": 0, + "source": { + "id": 65, + "protocol": "ipmisensor" + } + }, + { + "description": "Current of PS_Curr_In_01 reported by BMC, in amp", + "labels": { + "id": "ps_curr_in_01" + }, + "name": "bmc_other_current_amp", + "optional_period": 5000, + "period": 0, + "source": { + "id": 66, + "protocol": "ipmisensor" + } + }, + { + "description": "Power consumption of PS 2 Output reported by BMC, in watt", + "labels": { + "id": "ps2output" + }, + "name": "bmc_other_power_watt", + "optional_period": 5000, + "period": 0, + "source": { + "id": 67, + "protocol": "ipmisensor" + } + }, + { + "description": "Voltage of PS_Volt_Out_02 reported by BMC, in volt", + "labels": { + "id": "ps_volt_out_02" + }, + "name": "bmc_other_voltage_volt", + "optional_period": 5000, + "period": 0, + "source": { + "id": 68, + "protocol": "ipmisensor" + } + }, + { + "description": "Voltage of PS_Volt_In_02 reported by BMC, in volt", + "labels": { + "id": "ps_volt_in_02" + }, + "name": "bmc_other_voltage_volt", + "optional_period": 5000, + "period": 0, + "source": { + "id": 69, + "protocol": "ipmisensor" + } + }, + { + "description": "Current of PS_Curr_Out_02 reported by BMC, in amp", + "labels": { + "id": "ps_curr_out_02" + }, + "name": "bmc_other_current_amp", + "optional_period": 5000, + "period": 0, + "source": { + "id": 70, + "protocol": "ipmisensor" + } + }, + { + "description": "Current of PS_Curr_In_02 reported by BMC, in amp", + "labels": { + "id": "ps_curr_in_02" + }, + "name": "bmc_other_current_amp", + "optional_period": 5000, + "period": 0, + "source": { + "id": 71, + "protocol": "ipmisensor" + } + }, + { + "description": "Temperature of 30.1-OCP 2-I/O m reported by BMC, in celsius", + "labels": { + "id": "30.1-ocp2-i/om" + }, + "name": "bmc_other_temp_celsius", + "optional_period": 5000, + "period": 0, + "source": { + "id": 72, + "protocol": "ipmisensor" + } + }, + { + "description": "Temperature of 32.1-PCI 1-I/O m reported by BMC, in celsius", + "labels": { + "id": "32.1-pci1-i/om" + }, + "name": "bmc_other_temp_celsius", + "optional_period": 5000, + "period": 0, + "source": { + "id": 73, + "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_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": "Input byte counter for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifacein_bytes_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Output byte counter for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifaceout_bytes_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Input packet counter for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifacein_packets_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Output packet counter for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifaceout_packets_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Input counter of discarded packets for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifacein_packets_discard_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Output counter of discarded packets for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifaceout_packets_discard_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Input counter of packet errors for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifacein_packets_error_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Output counter of packet errors for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifaceout_packets_error_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + } ], - "model": "ProLiant DL385 Gen11", + "model": "ProLiant DL365 Gen11", "queues": [ "admin", "testing" diff --git a/data/grid5000/sites/nancy/clusters/gres/gres.json b/data/grid5000/sites/nancy/clusters/gres/gres.json index 622a04abe31e04577b6bf73b09abb1e0ebc7fe7f..9d5f060f2aade7b8bfca6f4122e9a936381a6fee 100644 --- a/data/grid5000/sites/nancy/clusters/gres/gres.json +++ b/data/grid5000/sites/nancy/clusters/gres/gres.json @@ -5,7 +5,94 @@ "kavlan": true, "manufactured_at": "2024-08-07", "metrics": [ - + { + "description": "Input byte counter for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifacein_bytes_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Output byte counter for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifaceout_bytes_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Input packet counter for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifacein_packets_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Output packet counter for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifaceout_packets_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Input counter of discarded packets for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifacein_packets_discard_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Output counter of discarded packets for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifaceout_packets_discard_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Input counter of packet errors for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifacein_packets_error_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + }, + { + "description": "Output counter of packet errors for the network device port", + "labels": { + "interface": "eth0" + }, + "name": "network_ifaceout_packets_error_total", + "period": 5000, + "source": { + "protocol": "network_equipment" + } + } ], "model": "ProLiant DL385 Gen11", "queues": [ diff --git a/data/grid5000/sites/nancy/network_equipments/sgrdix.json b/data/grid5000/sites/nancy/network_equipments/sgrdix.json index d45b740c35e4ca92aa10f68522fd6271d79f8866..c144ca03d07a522e4dd61cae99f236cd13ff5ea2 100644 --- a/data/grid5000/sites/nancy/network_equipments/sgrdix.json +++ b/data/grid5000/sites/nancy/network_equipments/sgrdix.json @@ -211,6 +211,80 @@ } ], "managed_by_us": true, + "metrics": [ + { + "description": "Input byte counter for the network device port", + "name": "network_ifacein_bytes_total", + "period": 5000, + "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": 5000, + "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": 5000, + "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": 5000, + "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": 5000, + "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": 5000, + "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": 5000, + "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": 5000, + "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": "Aruba 8325-48Y8C", "serial": "TW47KM00BX", "site": "nancy", diff --git a/input/grid5000/sites/nancy/clusters/grdix/grdix.yaml b/input/grid5000/sites/nancy/clusters/grdix/grdix.yaml index 452356548fb697b2b0da322127fe52464b7d8817..46550e3aa75e44f51a2981c3bea7bb7dcac134e4 100644 --- a/input/grid5000/sites/nancy/clusters/grdix/grdix.yaml +++ b/input/grid5000/sites/nancy/clusters/grdix/grdix.yaml @@ -1,5 +1,5 @@ --- -model: ProLiant DL385 Gen11 +model: ProLiant DL365 Gen11 created_at: 2024-09-27 kavlan: true boot_type: uefi diff --git a/input/grid5000/sites/nancy/clusters/grdix/grdix_metrics.yaml b/input/grid5000/sites/nancy/clusters/grdix/grdix_metrics.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5cb604bf4244ed2fc079786101bbc138ed811fad --- /dev/null +++ b/input/grid5000/sites/nancy/clusters/grdix/grdix_metrics.yaml @@ -0,0 +1,176 @@ +--- +metrics: + + - name: bmc_psu_power_watt + labels: {"psu": "1"} + description: Power consumption of Psu 1 reported by BMC, in watt + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 51 #PS 1 Input + + - name: bmc_psu_power_watt + labels: {"psu": "2"} + description: Power consumption of Psu 2 reported by BMC, in watt + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 53 #PS 2 Input + + - name: bmc_node_power_watt + description: Power consumption reported by BMC, in watt + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 54 #Power Meter + + - name: bmc_other_power_watt + labels: {"id": "ps1output"} + description: Power consumption of PS 1 Output reported by BMC, in watt + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 62 #PS 1 Output + + - name: bmc_other_voltage_volt + labels: {"id": "ps_volt_out_01"} + description: Voltage of PS_Volt_Out_01 reported by BMC, in volt + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 63 #PS_Volt_Out_01 + + - name: bmc_other_voltage_volt + labels: {"id": "ps_volt_in_01"} + description: Voltage of PS_Volt_In_01 reported by BMC, in volt + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 64 #PS_Volt_In_01 + + - name: bmc_other_current_amp + labels: {"id": "ps_curr_out_01"} + description: Current of PS_Curr_Out_01 reported by BMC, in amp + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 65 #PS_Curr_Out_01 + + - name: bmc_other_current_amp + labels: {"id": "ps_curr_in_01"} + description: Current of PS_Curr_In_01 reported by BMC, in amp + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 66 #PS_Curr_In_01 + + - name: bmc_other_power_watt + labels: {"id": "ps2output"} + description: Power consumption of PS 2 Output reported by BMC, in watt + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 67 #PS 2 Output + + - name: bmc_other_voltage_volt + labels: {"id": "ps_volt_out_02"} + description: Voltage of PS_Volt_Out_02 reported by BMC, in volt + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 68 #PS_Volt_Out_02 + + - name: bmc_other_voltage_volt + labels: {"id": "ps_volt_in_02"} + description: Voltage of PS_Volt_In_02 reported by BMC, in volt + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 69 #PS_Volt_In_02 + + - name: bmc_other_current_amp + labels: {"id": "ps_curr_out_02"} + description: Current of PS_Curr_Out_02 reported by BMC, in amp + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 70 #PS_Curr_Out_02 + + - name: bmc_other_current_amp + labels: {"id": "ps_curr_in_02"} + description: Current of PS_Curr_In_02 reported by BMC, in amp + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 71 #PS_Curr_In_02 + + - name: bmc_other_temp_celsius + labels: {"id": "30.1-ocp2-i/om"} + description: Temperature of 30.1-OCP 2-I/O m reported by BMC, in celsius + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 72 #30.1-OCP 2-I/O m + + - name: bmc_other_temp_celsius + labels: {"id": "32.1-pci1-i/om"} + description: Temperature of 32.1-PCI 1-I/O m reported by BMC, in celsius + period: 0 + optional_period: 5000 + source: + protocol: ipmisensor + id: 73 #32.1-PCI 1-I/O m + + - 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 \ No newline at end of file diff --git a/input/grid5000/sites/nancy/networks/sgrdix.yaml b/input/grid5000/sites/nancy/networks/sgrdix.yaml index 9dec4a5737210270205141c09e36e87a8c5e5360..74b623cedf2f906ef4e5c313400ed69e50fa11e5 100644 --- a/input/grid5000/sites/nancy/networks/sgrdix.yaml +++ b/input/grid5000/sites/nancy/networks/sgrdix.yaml @@ -8,6 +8,55 @@ sgrdix: site: nancy ip: 172.17.79.213 snmp_community: public + metrics: + - name: network_ifacein_bytes_total + description: Input byte counter for the network device port + period: 5000 + 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: 5000 + 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: 5000 + 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: 5000 + 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: 5000 + 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: 5000 + 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: 5000 + 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: 5000 + 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% }} backplane_bps: 1440000000000 linecards: 1: