]> git-server-git.apps.pok.os.sepia.ceph.com Git - cephmetrics.git/commitdiff
Add dashboard for monitoring the db host itself
authorZack Cerza <zack@redhat.com>
Tue, 1 May 2018 17:59:31 +0000 (11:59 -0600)
committerZack Cerza <zack@redhat.com>
Tue, 1 May 2018 18:34:19 +0000 (12:34 -0600)
Signed-off-by: Zack Cerza <zack@redhat.com>
dashboards/mgr-prometheus/cephmetrics-host.json [new file with mode: 0644]

diff --git a/dashboards/mgr-prometheus/cephmetrics-host.json b/dashboards/mgr-prometheus/cephmetrics-host.json
new file mode 100644 (file)
index 0000000..77dd717
--- /dev/null
@@ -0,0 +1,1001 @@
+{
+  "__inputs": [
+    {
+      "name": "DS_LOCAL_PROMETHEUS",
+      "label": "local_prometheus",
+      "description": "",
+      "type": "datasource",
+      "pluginId": "prometheus",
+      "pluginName": "Prometheus"
+    }
+  ],
+  "__requires": [
+    {
+      "type": "grafana",
+      "id": "grafana",
+      "name": "Grafana",
+      "version": "5.0.4"
+    },
+    {
+      "type": "panel",
+      "id": "graph",
+      "name": "Graph",
+      "version": "5.0.0"
+    },
+    {
+      "type": "datasource",
+      "id": "prometheus",
+      "name": "Prometheus",
+      "version": "5.0.0"
+    }
+  ],
+  "annotations": {
+    "list": [
+      {
+        "builtIn": 1,
+        "datasource": "-- Grafana --",
+        "enable": true,
+        "hide": true,
+        "iconColor": "rgba(0, 211, 255, 1)",
+        "name": "Annotations & Alerts",
+        "type": "dashboard"
+      }
+    ]
+  },
+  "editable": true,
+  "gnetId": null,
+  "graphTooltip": 0,
+  "id": null,
+  "links": [],
+  "panels": [
+    {
+      "collapsed": false,
+      "gridPos": {
+        "h": 1,
+        "w": 24,
+        "x": 0,
+        "y": 0
+      },
+      "id": 9,
+      "panels": [],
+      "repeat": null,
+      "title": "CPU & RAM",
+      "type": "row"
+    },
+    {
+      "aliasColors": {},
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 8,
+        "x": 0,
+        "y": 1
+      },
+      "id": 1,
+      "legend": {
+        "avg": false,
+        "current": false,
+        "max": false,
+        "min": false,
+        "show": false,
+        "total": false,
+        "values": false
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "minSpan": 8,
+      "nullPointMode": "null",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [],
+      "spaceLength": 10,
+      "stack": false,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "100 - (avg by (instance) (irate(node_cpu{job=\"cephmetrics\",mode=\"idle\"}[5m])) * 100)",
+          "format": "time_series",
+          "instant": false,
+          "intervalFactor": 1,
+          "legendFormat": "{{Busy %}}",
+          "refId": "A"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": null,
+      "timeShift": null,
+      "title": "Cephmetrics Host CPU Busy %",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "decimals": null,
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": "100",
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": false
+        }
+      ]
+    },
+    {
+      "aliasColors": {},
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 8,
+        "x": 8,
+        "y": 1
+      },
+      "id": 2,
+      "legend": {
+        "avg": false,
+        "current": false,
+        "max": false,
+        "min": false,
+        "show": true,
+        "total": false,
+        "values": false
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "minSpan": 8,
+      "nullPointMode": "null",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [],
+      "spaceLength": 10,
+      "stack": true,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "(avg by(mode) (irate(node_cpu{job=\"cephmetrics\",mode!=\"idle\"}[30s]))*100)",
+          "format": "time_series",
+          "instant": false,
+          "interval": "",
+          "intervalFactor": 1,
+          "legendFormat": "{{mode}}",
+          "refId": "A"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": null,
+      "timeShift": null,
+      "title": "Cephmetrics Host  CPU Usage Breakdown",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "decimals": null,
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": "100",
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": false
+        }
+      ]
+    },
+    {
+      "aliasColors": {
+        "Installed": "#890f02"
+      },
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 8,
+        "x": 16,
+        "y": 1
+      },
+      "id": 7,
+      "legend": {
+        "avg": false,
+        "current": true,
+        "max": false,
+        "min": false,
+        "show": true,
+        "total": false,
+        "values": true
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "minSpan": 8,
+      "nullPointMode": "null",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [
+        {
+          "alias": "Installed",
+          "fill": 0,
+          "linewidth": 2
+        }
+      ],
+      "spaceLength": 10,
+      "stack": false,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "node_memory_MemTotal{job=\"cephmetrics\"} - node_memory_MemAvailable{job=\"cephmetrics\"}",
+          "format": "time_series",
+          "interval": "",
+          "intervalFactor": 1,
+          "legendFormat": "Used",
+          "refId": "C"
+        },
+        {
+          "expr": "node_memory_MemTotal{job=\"cephmetrics\"} ",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "Installed",
+          "refId": "B"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": "7d",
+      "timeShift": null,
+      "title": "RAM Usage",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "format": "bytes",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": true
+        }
+      ]
+    },
+    {
+      "collapsed": false,
+      "gridPos": {
+        "h": 1,
+        "w": 24,
+        "x": 0,
+        "y": 8
+      },
+      "id": 10,
+      "panels": [],
+      "repeat": null,
+      "title": "Disk",
+      "type": "row"
+    },
+    {
+      "aliasColors": {},
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 8,
+        "x": 0,
+        "y": 9
+      },
+      "id": 3,
+      "legend": {
+        "avg": false,
+        "current": false,
+        "max": false,
+        "min": false,
+        "show": true,
+        "total": false,
+        "values": false
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "minSpan": 8,
+      "nullPointMode": "null",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [],
+      "spaceLength": 10,
+      "stack": false,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "irate(node_disk_io_time_ms{job='cephmetrics',device=~\"[h,s,v]d[a-z]\"}[1m]) / 10",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "{{device}}",
+          "refId": "A"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": null,
+      "timeShift": null,
+      "title": "Disk Utilization",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "format": "percent",
+          "label": "%UTIL",
+          "logBase": 1,
+          "max": null,
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": false
+        }
+      ]
+    },
+    {
+      "aliasColors": {},
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 8,
+        "x": 8,
+        "y": 9
+      },
+      "id": 4,
+      "legend": {
+        "avg": false,
+        "current": false,
+        "max": false,
+        "min": false,
+        "show": true,
+        "total": false,
+        "values": false
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "minSpan": 8,
+      "nullPointMode": "null as zero",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [],
+      "spaceLength": 10,
+      "stack": false,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "irate(node_disk_reads_completed{job=\"cephmetrics\",device=~\"[h,s,v]d[a-z]\"}[30s]) + irate(node_disk_writes_completed{job=~\"cephmetrics\",device=~\"[h,s,v]d[a-z]\"}[30s])",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "{{device}}",
+          "refId": "A"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": null,
+      "timeShift": null,
+      "title": "Disk IOPS",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "format": "none",
+          "label": "IOPS",
+          "logBase": 1,
+          "max": null,
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": false
+        }
+      ]
+    },
+    {
+      "aliasColors": {},
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 8,
+        "x": 16,
+        "y": 9
+      },
+      "id": 5,
+      "legend": {
+        "avg": false,
+        "current": false,
+        "max": false,
+        "min": false,
+        "show": true,
+        "total": false,
+        "values": false
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "minSpan": 8,
+      "nullPointMode": "null as zero",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [],
+      "spaceLength": 10,
+      "stack": false,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "irate(node_disk_bytes_read{job='cephmetrics',device=~\"[h,s,v]d[a-z]\"}[1m]) + irate(node_disk_bytes_written{job='cephmetrics',device=~\"[h,s,v]d[a-z]\"}[1m])  ",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "{{device}}",
+          "refId": "A"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": null,
+      "timeShift": null,
+      "title": "Disk Throughput",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "format": "bytes",
+          "label": "Throughput",
+          "logBase": 1,
+          "max": null,
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": false
+        }
+      ]
+    },
+    {
+      "aliasColors": {},
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 8,
+        "x": 0,
+        "y": 16
+      },
+      "id": 6,
+      "legend": {
+        "avg": false,
+        "current": false,
+        "max": false,
+        "min": false,
+        "show": true,
+        "total": false,
+        "values": false
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "minSpan": 8,
+      "nullPointMode": "null as zero",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [],
+      "spaceLength": 10,
+      "stack": false,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "(irate(node_disk_write_time_ms{job='cephmetrics',device=~\"[h,s,v]d[a-z]\"}[1m]) + irate(node_disk_read_time_ms{job='cephmetrics',device=~\"[h,s,v]d[a-z]\"}[1m])) / \n(irate(node_disk_reads_completed{job='cephmetrics',device=~\"[h,s,v]d[a-z]\"}[1m]) + irate(node_disk_writes_completed{job='cephmetrics',device=~\"[h,s,v]d[a-z]\"}[1m]))",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "{{device}}",
+          "refId": "A"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": null,
+      "timeShift": null,
+      "title": "Disk Latency",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "format": "ms",
+          "label": "",
+          "logBase": 1,
+          "max": null,
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": true
+        }
+      ]
+    },
+    {
+      "aliasColors": {
+        "Size": "#890f02"
+      },
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 16,
+        "x": 8,
+        "y": 16
+      },
+      "id": 13,
+      "legend": {
+        "avg": false,
+        "current": false,
+        "max": false,
+        "min": false,
+        "show": true,
+        "total": false,
+        "values": false
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "nullPointMode": "null",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [
+        {
+          "alias": "Size",
+          "fill": 0,
+          "linewidth": 2
+        }
+      ],
+      "spaceLength": 10,
+      "stack": false,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "node_filesystem_size{mountpoint='/var/lib/cephmetrics/data',fstype='xfs'} - node_filesystem_free{mountpoint='/var/lib/cephmetrics/data',fstype='xfs'}",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "Used",
+          "refId": "A"
+        },
+        {
+          "expr": "node_filesystem_size{mountpoint='/var/lib/cephmetrics/data',fstype='xfs'}",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "Size",
+          "refId": "B"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": null,
+      "timeShift": null,
+      "title": "Filesystem Usage - Prometheus Growth",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "format": "bytes",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": true
+        }
+      ]
+    },
+    {
+      "collapsed": false,
+      "gridPos": {
+        "h": 1,
+        "w": 24,
+        "x": 0,
+        "y": 23
+      },
+      "id": 11,
+      "panels": [],
+      "repeat": null,
+      "title": "Network",
+      "type": "row"
+    },
+    {
+      "aliasColors": {},
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 24,
+        "x": 0,
+        "y": 24
+      },
+      "id": 8,
+      "legend": {
+        "avg": false,
+        "current": false,
+        "max": false,
+        "min": false,
+        "show": true,
+        "total": false,
+        "values": false
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "nullPointMode": "null",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [],
+      "spaceLength": 10,
+      "stack": true,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "sum(\n  irate(node_network_receive_bytes{job='cephmetrics'}[1m]) \n  ) by(instance)",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "Network RX",
+          "refId": "A"
+        },
+        {
+          "expr": "sum(\n  irate(node_network_transmit_bytes{job='cephmetrics'}[1m]) \n  ) by(instance)",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "Network TX",
+          "refId": "B"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": null,
+      "timeShift": null,
+      "title": "Network Load",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "format": "Bps",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": true
+        }
+      ]
+    },
+    {
+      "aliasColors": {
+        "Size": "#890f02"
+      },
+      "bars": false,
+      "dashLength": 10,
+      "dashes": false,
+      "datasource": null,
+      "description": "Shows data growth on '/' which includes the prometheus data. Prometheus installed 04/14, migrated to separate logical volume 04/22",
+      "fill": 1,
+      "gridPos": {
+        "h": 7,
+        "w": 16,
+        "x": 0,
+        "y": 31
+      },
+      "id": 14,
+      "legend": {
+        "avg": false,
+        "current": false,
+        "max": false,
+        "min": false,
+        "show": true,
+        "total": false,
+        "values": false
+      },
+      "lines": true,
+      "linewidth": 1,
+      "links": [],
+      "nullPointMode": "null",
+      "percentage": false,
+      "pointradius": 5,
+      "points": false,
+      "renderer": "flot",
+      "seriesOverrides": [
+        {
+          "alias": "Size",
+          "fill": 0,
+          "linewidth": 2
+        }
+      ],
+      "spaceLength": 10,
+      "stack": false,
+      "steppedLine": false,
+      "targets": [
+        {
+          "expr": "node_filesystem_size{mountpoint='/',fstype='xfs'} - node_filesystem_free{mountpoint='/',fstype='xfs'}",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "Used",
+          "refId": "A"
+        },
+        {
+          "expr": "node_filesystem_size{mountpoint='/',fstype='xfs'}",
+          "format": "time_series",
+          "intervalFactor": 1,
+          "legendFormat": "Size",
+          "refId": "B"
+        }
+      ],
+      "thresholds": [],
+      "timeFrom": "14d",
+      "timeShift": null,
+      "title": "Filesystem Usage - '/'",
+      "tooltip": {
+        "shared": true,
+        "sort": 0,
+        "value_type": "individual"
+      },
+      "type": "graph",
+      "xaxis": {
+        "buckets": null,
+        "mode": "time",
+        "name": null,
+        "show": true,
+        "values": []
+      },
+      "yaxes": [
+        {
+          "format": "bytes",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": "0",
+          "show": true
+        },
+        {
+          "format": "short",
+          "label": null,
+          "logBase": 1,
+          "max": null,
+          "min": null,
+          "show": true
+        }
+      ]
+    }
+  ],
+  "refresh": "15s",
+  "schemaVersion": 16,
+  "style": "dark",
+  "tags": [],
+  "templating": {
+    "list": []
+  },
+  "time": {
+    "from": "now-1h",
+    "to": "now"
+  },
+  "timepicker": {
+    "refresh_intervals": [
+      "5s",
+      "10s",
+      "15s",
+      "30s",
+      "1m",
+      "5m",
+      "15m",
+      "30m",
+      "1h",
+      "2h",
+      "1d"
+    ],
+    "time_options": [
+      "5m",
+      "15m",
+      "1h",
+      "6h",
+      "12h",
+      "24h",
+      "2d",
+      "7d",
+      "30d"
+    ]
+  },
+  "timezone": "",
+  "title": "Cephmetrics Host Metrics",
+  "uid": "000000001",
+  "version": 12
+}