mirror of
https://github.com/ceph/ceph
synced 2026-08-02 07:03:18 +00:00
monitoring: add panel descriptions and fix tooltip labels
Signed-off-by: Afreen Misbah <afreen@ibm.com>
This commit is contained in:
Afreen Misbah
parent
4be833e11e
commit
b231e6f318
@ -82,9 +82,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
)
|
||||
.addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_health)',
|
||||
legendFormat='__auto'
|
||||
legendFormat='Health'
|
||||
))
|
||||
+ {
|
||||
description: 'Overall hardware health across all hosts',
|
||||
fieldConfig: {
|
||||
defaults: {
|
||||
mappings: [
|
||||
@ -116,8 +117,9 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
])
|
||||
.addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_temperature_celsius{sensor_name=~".*CPU_TEMP"})',
|
||||
legendFormat='__auto'
|
||||
)),
|
||||
legendFormat='CPU'
|
||||
))
|
||||
+ { description: 'Highest CPU temperature across all hosts' },
|
||||
|
||||
// BMC Versions
|
||||
$.pieChartPanel(
|
||||
@ -220,8 +222,9 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 4, w: 2, x: 0, y: 5 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'count(count by(hostname) (ceph_hardware_health))',
|
||||
legendFormat='__auto'
|
||||
)),
|
||||
legendFormat='Hosts'
|
||||
))
|
||||
+ { description: 'Total hosts with hardware monitoring' },
|
||||
|
||||
// Drives
|
||||
$.addStatPanel(
|
||||
@ -231,8 +234,9 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 4, w: 2, x: 2, y: 5 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'count(ceph_hardware_storage_capacity_bytes)',
|
||||
legendFormat='__auto'
|
||||
)),
|
||||
legendFormat='Drives'
|
||||
))
|
||||
+ { description: 'Total storage drives detected' },
|
||||
|
||||
// NVMe Temp
|
||||
$.addGaugePanel(
|
||||
@ -250,8 +254,9 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
])
|
||||
.addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_temperature_celsius{sensor_name=~"NVME.*_TEMP"})',
|
||||
legendFormat='__auto'
|
||||
)),
|
||||
legendFormat='NVMe'
|
||||
))
|
||||
+ { description: 'Highest NVMe drive temperature across all hosts' },
|
||||
] },
|
||||
|
||||
// Row 2: Host Overview
|
||||
@ -264,9 +269,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 3, w: 2, x: 0, y: 2 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_health{hostname=~"$hostname",category="power"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='Power'
|
||||
))
|
||||
+ {
|
||||
description: 'Power supply health status',
|
||||
fieldConfig: {
|
||||
defaults: {
|
||||
mappings: [
|
||||
@ -287,9 +293,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 4, w: 3, x: 2, y: 2 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_temperature_celsius{hostname=~"$hostname", sensor_name=~".*MB_TEMP.*"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='Motherboard'
|
||||
))
|
||||
+ {
|
||||
description: 'Highest motherboard temperature',
|
||||
fieldConfig+: { defaults+: { thresholds: { mode: 'absolute', steps: [{ color: 'green' }, { color: '#EAB839', value: 70 }, { color: 'red', value: 80 }] } } },
|
||||
options+: { colorMode: 'none', graphMode: 'area' },
|
||||
},
|
||||
@ -302,9 +309,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 4, w: 3, x: 5, y: 2 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'avg(ceph_hardware_temperature_celsius{hostname=~"$hostname", sensor_name=~".*CPU_TEMP"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='CPU'
|
||||
))
|
||||
+ {
|
||||
description: 'Average CPU temperature',
|
||||
fieldConfig+: { defaults+: { max: 85, min: 0, thresholds: { mode: 'absolute', steps: [{ color: 'green' }, { color: '#EAB839', value: 80 }, { color: 'semi-dark-red', value: 90 }] } } },
|
||||
options+: { colorMode: 'none', graphMode: 'area' },
|
||||
},
|
||||
@ -317,9 +325,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 4, w: 3, x: 8, y: 2 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_temperature_celsius{hostname=~"$hostname", sensor_name=~".*DIMM.*_TEMP"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='DIMM'
|
||||
))
|
||||
+ {
|
||||
description: 'Highest DIMM temperature',
|
||||
fieldConfig+: { defaults+: { max: 88, min: 0, thresholds: { mode: 'absolute', steps: [{ color: 'green' }, { color: 'yellow', value: 70 }, { color: 'semi-dark-red', value: 80 }] } } },
|
||||
options+: { colorMode: 'none', graphMode: 'area' },
|
||||
},
|
||||
@ -332,9 +341,12 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 4, w: 3, x: 11, y: 2 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'count(ceph_hardware_fan_rpm{hostname=~"$hostname", fan_name=~"PSU.*"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='PSU Fans'
|
||||
))
|
||||
+ { options+: { colorMode: 'none', graphMode: 'none' } },
|
||||
+ {
|
||||
description: 'Number of PSU fans detected',
|
||||
options+: { colorMode: 'none', graphMode: 'none' },
|
||||
},
|
||||
|
||||
// AVG PSU Temperature
|
||||
$.addStatPanel(
|
||||
@ -344,9 +356,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 4, w: 3, x: 14, y: 2 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'avg(ceph_hardware_temperature_celsius{hostname=~"$hostname", sensor_name=~"PSU.*_TEMP.*"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='PSU'
|
||||
))
|
||||
+ {
|
||||
description: 'Average power supply temperature',
|
||||
fieldConfig+: { defaults+: { color: { fixedColor: 'blue', mode: 'fixed' }, max: 100, min: 0, thresholds: { mode: 'absolute', steps: [{ color: 'green' }, { color: 'yellow', value: 80 }, { color: 'semi-dark-red', value: 90 }] } } },
|
||||
options+: { colorMode: 'none', graphMode: 'area' },
|
||||
},
|
||||
@ -359,9 +372,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 4, w: 3, x: 17, y: 2 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'count(ceph_hardware_storage_capacity_bytes{hostname=~"$hostname", protocol="NVMe"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='NVMe'
|
||||
))
|
||||
+ {
|
||||
description: 'Number of NVMe drives detected',
|
||||
fieldConfig+: { defaults+: { color: { fixedColor: 'blue', mode: 'thresholds' }, thresholds: { mode: 'absolute', steps: [{ color: 'green' }, { color: '#EAB839', value: 70 }, { color: 'semi-dark-red', value: 78 }] } } },
|
||||
options+: { colorMode: 'none', graphMode: 'none' },
|
||||
},
|
||||
@ -374,9 +388,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 4, w: 3, x: 20, y: 2 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_temperature_celsius{hostname=~"$hostname", sensor_name=~"NVME.*"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='NVMe'
|
||||
))
|
||||
+ {
|
||||
description: 'Highest NVMe drive temperature',
|
||||
fieldConfig+: { defaults+: { color: { fixedColor: 'blue', mode: 'thresholds' }, max: 85, min: 0, thresholds: { mode: 'absolute', steps: [{ color: 'green' }, { color: '#EAB839', value: 70 }, { color: 'semi-dark-red', value: 78 }] } } },
|
||||
options+: { colorMode: 'none', graphMode: 'area' },
|
||||
},
|
||||
@ -389,9 +404,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 3, w: 2, x: 0, y: 5 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_health{hostname=~"$hostname",category="network"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='Network'
|
||||
))
|
||||
+ {
|
||||
description: 'Network adapter health status',
|
||||
fieldConfig: {
|
||||
defaults: {
|
||||
mappings: [
|
||||
@ -412,9 +428,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 3, w: 2, x: 0, y: 8 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_health{hostname=~"$hostname",category="fans"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='Cooling'
|
||||
))
|
||||
+ {
|
||||
description: 'Cooling fans health status',
|
||||
fieldConfig: {
|
||||
defaults: {
|
||||
mappings: [
|
||||
@ -435,9 +452,10 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
gridPosition={ h: 3, w: 2, x: 0, y: 11 }
|
||||
).addTarget($.addTargetSchema(
|
||||
'max(ceph_hardware_health{hostname=~"$hostname",category="storage"})',
|
||||
legendFormat='__auto'
|
||||
legendFormat='Drives'
|
||||
))
|
||||
+ {
|
||||
description: 'Storage drives health status',
|
||||
fieldConfig: {
|
||||
defaults: {
|
||||
mappings: [
|
||||
@ -467,7 +485,7 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
)
|
||||
.addTarget($.addTargetSchema(
|
||||
'ceph_hardware_storage_capacity_bytes{hostname=~"$hostname"}',
|
||||
legendFormat='__auto',
|
||||
legendFormat='',
|
||||
instant=true
|
||||
) + { format: 'table' })
|
||||
.addTransformations([
|
||||
@ -509,7 +527,7 @@ local g = import 'grafonnet/grafana.libsonnet';
|
||||
)
|
||||
.addTarget($.addTargetSchema(
|
||||
'ceph_hardware_firmware_info{hostname=~"$hostname"}',
|
||||
legendFormat='__auto',
|
||||
legendFormat='',
|
||||
instant=true
|
||||
) + { format: 'table' })
|
||||
.addTransformations([
|
||||
|
||||
@ -51,7 +51,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Overall hardware health across all hosts",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"mappings": [
|
||||
@ -120,7 +120,7 @@
|
||||
"expr": "max(ceph_hardware_health)",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "Health",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -130,7 +130,7 @@
|
||||
},
|
||||
{
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Highest CPU temperature across all hosts",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"links": [ ],
|
||||
@ -183,7 +183,7 @@
|
||||
"expr": "max(ceph_hardware_temperature_celsius{sensor_name=~\".*CPU_TEMP\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "CPU",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -409,7 +409,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Total hosts with hardware monitoring",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"decimals": 0,
|
||||
@ -450,7 +450,7 @@
|
||||
"expr": "count(count by(hostname) (ceph_hardware_health))",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "Hosts",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -461,7 +461,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Total storage drives detected",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"decimals": 0,
|
||||
@ -502,7 +502,7 @@
|
||||
"expr": "count(ceph_hardware_storage_capacity_bytes)",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "Drives",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -512,7 +512,7 @@
|
||||
},
|
||||
{
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Highest NVMe drive temperature across all hosts",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"links": [ ],
|
||||
@ -565,7 +565,7 @@
|
||||
"expr": "max(ceph_hardware_temperature_celsius{sensor_name=~\"NVME.*_TEMP\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "NVMe",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -596,7 +596,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Power supply health status",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"mappings": [
|
||||
@ -665,7 +665,7 @@
|
||||
"expr": "max(ceph_hardware_health{hostname=~\"$hostname\",category=\"power\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "Power",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -676,7 +676,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Highest motherboard temperature",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"decimals": 0,
|
||||
@ -729,7 +729,7 @@
|
||||
"expr": "max(ceph_hardware_temperature_celsius{hostname=~\"$hostname\", sensor_name=~\".*MB_TEMP.*\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "Motherboard",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -740,7 +740,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Average CPU temperature",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"decimals": 0,
|
||||
@ -795,7 +795,7 @@
|
||||
"expr": "avg(ceph_hardware_temperature_celsius{hostname=~\"$hostname\", sensor_name=~\".*CPU_TEMP\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "CPU",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -806,7 +806,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Highest DIMM temperature",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"decimals": 0,
|
||||
@ -861,7 +861,7 @@
|
||||
"expr": "max(ceph_hardware_temperature_celsius{hostname=~\"$hostname\", sensor_name=~\".*DIMM.*_TEMP\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "DIMM",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -872,7 +872,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Number of PSU fans detected",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"decimals": 0,
|
||||
@ -913,7 +913,7 @@
|
||||
"expr": "count(ceph_hardware_fan_rpm{hostname=~\"$hostname\", fan_name=~\"PSU.*\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "PSU Fans",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -924,7 +924,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Average power supply temperature",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"color": {
|
||||
@ -983,7 +983,7 @@
|
||||
"expr": "avg(ceph_hardware_temperature_celsius{hostname=~\"$hostname\", sensor_name=~\"PSU.*_TEMP.*\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "PSU",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -994,7 +994,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Number of NVMe drives detected",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"color": {
|
||||
@ -1051,7 +1051,7 @@
|
||||
"expr": "count(ceph_hardware_storage_capacity_bytes{hostname=~\"$hostname\", protocol=\"NVMe\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "NVMe",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -1062,7 +1062,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Highest NVMe drive temperature",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"color": {
|
||||
@ -1121,7 +1121,7 @@
|
||||
"expr": "max(ceph_hardware_temperature_celsius{hostname=~\"$hostname\", sensor_name=~\"NVME.*\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "NVMe",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -1132,7 +1132,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Network adapter health status",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"mappings": [
|
||||
@ -1201,7 +1201,7 @@
|
||||
"expr": "max(ceph_hardware_health{hostname=~\"$hostname\",category=\"network\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "Network",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -1212,7 +1212,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Cooling fans health status",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"mappings": [
|
||||
@ -1281,7 +1281,7 @@
|
||||
"expr": "max(ceph_hardware_health{hostname=~\"$hostname\",category=\"fans\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "Cooling",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -1292,7 +1292,7 @@
|
||||
{
|
||||
"colors": null,
|
||||
"datasource": "$datasource",
|
||||
"description": "",
|
||||
"description": "Storage drives health status",
|
||||
"fieldConfig": {
|
||||
"defaults": {
|
||||
"mappings": [
|
||||
@ -1361,7 +1361,7 @@
|
||||
"expr": "max(ceph_hardware_health{hostname=~\"$hostname\",category=\"storage\"})",
|
||||
"format": "time_series",
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "Drives",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -1416,7 +1416,7 @@
|
||||
"format": "table",
|
||||
"instant": true,
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
@ -1493,7 +1493,7 @@
|
||||
"format": "table",
|
||||
"instant": true,
|
||||
"intervalFactor": 1,
|
||||
"legendFormat": "__auto",
|
||||
"legendFormat": "",
|
||||
"refId": "A"
|
||||
}
|
||||
],
|
||||
|
||||
Loading…
Reference in New Issue
Block a user
