ceph/monitoring/ceph-mixin/dashboards/rbd.libsonnet

   1 local g = import 'grafonnet/grafana.libsonnet';
   2 local u = import 'utils.libsonnet';
   3
   4 {
   5   grafanaDashboards+:: {
   6     'rbd-details.json':
   7       local RbdDetailsPanel(title, formatY1, expr1, expr2, x, y, w, h) =
   8         u.graphPanelSchema({},
   9                            title,
  10                            '',
  11                            'null as zero',
  12                            false,
  13                            formatY1,
  14                            formatY1,
  15                            null,
  16                            null,
  17                            0,
  18                            1,
  19                            '$Datasource')
  20         .addTargets(
  21           [
  22             u.addTargetSchema(expr1,
  23                               1,
  24                               'time_series',
  25                               '{{pool}} Write'),
  26             u.addTargetSchema(expr2, 1, 'time_series', '{{pool}} Read'),
  27           ]
  28         ) + { gridPos: { x: x, y: y, w: w, h: h } };
  29
  30       u.dashboardSchema(
  31         'RBD Details',
  32         'Detailed Performance of RBD Images (IOPS/Throughput/Latency)',
  33         'YhCYGcuZz',
  34         'now-1h',
  35         false,
  36         16,
  37         [],
  38         '',
  39         {
  40           refresh_intervals: ['5s', '10s', '30s', '1m', '5m', '15m', '30m', '1h', '2h', '1d'],
  41           time_options: ['5m', '15m', '1h', '6h', '12h', '24h', '2d', '7d', '30d'],
  42         }
  43       )
  44       .addAnnotation(
  45         u.addAnnotationSchema(
  46           1,
  47           '-- Grafana --',
  48           true,
  49           true,
  50           'rgba(0, 211, 255, 1)',
  51           'Annotations & Alerts',
  52           'dashboard'
  53         )
  54       )
  55       .addRequired(
  56         type='grafana', id='grafana', name='Grafana', version='5.3.3'
  57       )
  58       .addRequired(
  59         type='panel', id='graph', name='Graph', version='5.0.0'
  60       )
  61       .addTemplate(
  62         g.template.datasource('Datasource', 'prometheus', 'default', label=null)
  63       )
  64       .addTemplate(
  65         u.addTemplateSchema('Pool',
  66                             '$Datasource',
  67                             'label_values(pool)',
  68                             1,
  69                             false,
  70                             0,
  71                             '',
  72                             '')
  73       )
  74       .addTemplate(
  75         u.addTemplateSchema('Image',
  76                             '$Datasource',
  77                             'label_values(image)',
  78                             1,
  79                             false,
  80                             0,
  81                             '',
  82                             '')
  83       )
  84       .addPanels([
  85         RbdDetailsPanel(
  86           'IOPS',
  87           'iops',
  88           'irate(ceph_rbd_write_ops{pool="$Pool", image="$Image"}[30s])',
  89           'irate(ceph_rbd_read_ops{pool="$Pool", image="$Image"}[30s])',
  90           0,
  91           0,
  92           8,
  93           9
  94         ),
  95         RbdDetailsPanel(
  96           'Throughput',
  97           'Bps',
  98           'irate(ceph_rbd_write_bytes{pool="$Pool", image="$Image"}[30s])',
  99           'irate(ceph_rbd_read_bytes{pool="$Pool", image="$Image"}[30s])',
 100           8,
 101           0,
 102           8,
 103           9
 104         ),
 105         RbdDetailsPanel(
 106           'Average Latency',
 107           'ns',
 108           'irate(ceph_rbd_write_latency_sum{pool="$Pool", image="$Image"}[30s]) / irate(ceph_rbd_write_latency_count{pool="$Pool", image="$Image"}[30s])',
 109           'irate(ceph_rbd_read_latency_sum{pool="$Pool", image="$Image"}[30s]) / irate(ceph_rbd_read_latency_count{pool="$Pool", image="$Image"}[30s])',
 110           16,
 111           0,
 112           8,
 113           9
 114         ),
 115       ]),
 116     'rbd-overview.json':
 117       local RgwOverviewStyle(alias, pattern, type, unit) =
 118         u.addStyle(alias,
 119                    null,
 120                    ['rgba(245, 54, 54, 0.9)', 'rgba(237, 129, 40, 0.89)', 'rgba(50, 172, 45, 0.97)'],
 121                    'YYYY-MM-DD HH:mm:ss',
 122                    2,
 123                    1,
 124                    pattern,
 125                    [],
 126                    type,
 127                    unit,
 128                    []);
 129       local RbdOverviewPanel(title,
 130                              formatY1,
 131                              expr1,
 132                              expr2,
 133                              legendFormat1,
 134                              legendFormat2,
 135                              x,
 136                              y,
 137                              w,
 138                              h) =
 139         u.graphPanelSchema({},
 140                            title,
 141                            '',
 142                            'null',
 143                            false,
 144                            formatY1,
 145                            'short',
 146                            null,
 147                            null,
 148                            0,
 149                            1,
 150                            '$datasource')
 151         .addTargets(
 152           [
 153             u.addTargetSchema(expr1,
 154                               1,
 155                               'time_series',
 156                               legendFormat1),
 157             u.addTargetSchema(expr2,
 158                               1,
 159                               'time_series',
 160                               legendFormat2),
 161           ]
 162         ) + { gridPos: { x: x, y: y, w: w, h: h } };
 163
 164       u.dashboardSchema(
 165         'RBD Overview',
 166         '',
 167         '41FrpeUiz',
 168         'now-1h',
 169         '30s',
 170         16,
 171         ['overview'],
 172         '',
 173         {
 174           refresh_intervals: ['5s', '10s', '15s', '30s', '1m', '5m', '15m', '30m', '1h', '2h', '1d'],
 175           time_options: ['5m', '15m', '1h', '6h', '12h', '24h', '2d', '7d', '30d'],
 176         }
 177       )
 178       .addAnnotation(
 179         u.addAnnotationSchema(
 180           1,
 181           '-- Grafana --',
 182           true,
 183           true,
 184           'rgba(0, 211, 255, 1)',
 185           'Annotations & Alerts',
 186           'dashboard'
 187         )
 188       )
 189       .addRequired(
 190         type='grafana', id='grafana', name='Grafana', version='5.4.2'
 191       )
 192       .addRequired(
 193         type='panel', id='graph', name='Graph', version='5.0.0'
 194       )
 195       .addRequired(
 196         type='datasource', id='prometheus', name='Prometheus', version='5.0.0'
 197       )
 198       .addRequired(
 199         type='panel', id='table', name='Table', version='5.0.0'
 200       )
 201       .addTemplate(
 202         g.template.datasource('datasource',
 203                               'prometheus',
 204                               'default',
 205                               label='Data Source')
 206       )
 207       .addPanels([
 208         RbdOverviewPanel(
 209           'IOPS',
 210           'short',
 211           'round(sum(irate(ceph_rbd_write_ops[30s])))',
 212           'round(sum(irate(ceph_rbd_read_ops[30s])))',
 213           'Writes',
 214           'Reads',
 215           0,
 216           0,
 217           8,
 218           7
 219         ),
 220         RbdOverviewPanel(
 221           'Throughput',
 222           'Bps',
 223           'round(sum(irate(ceph_rbd_write_bytes[30s])))',
 224           'round(sum(irate(ceph_rbd_read_bytes[30s])))',
 225           'Write',
 226           'Read',
 227           8,
 228           0,
 229           8,
 230           7
 231         ),
 232         RbdOverviewPanel(
 233           'Average Latency',
 234           'ns',
 235           'round(sum(irate(ceph_rbd_write_latency_sum[30s])) / sum(irate(ceph_rbd_write_latency_count[30s])))',
 236           'round(sum(irate(ceph_rbd_read_latency_sum[30s])) / sum(irate(ceph_rbd_read_latency_count[30s])))',
 237           'Write',
 238           'Read',
 239           16,
 240           0,
 241           8,
 242           7
 243         ),
 244         u.addTableSchema(
 245           '$datasource',
 246           '',
 247           { col: 3, desc: true },
 248           [
 249             RgwOverviewStyle('Pool', 'pool', 'string', 'short'),
 250             RgwOverviewStyle('Image', 'image', 'string', 'short'),
 251             RgwOverviewStyle('IOPS', 'Value', 'number', 'iops'),
 252             RgwOverviewStyle('', '/.*/', 'hidden', 'short'),
 253           ],
 254           'Highest IOPS',
 255           'table'
 256         )
 257         .addTarget(
 258           u.addTargetSchema(
 259             'topk(10, (sort((irate(ceph_rbd_write_ops[30s]) + on (image, pool, namespace) irate(ceph_rbd_read_ops[30s])))))',
 260             1,
 261             'table',
 262             ''
 263           )
 264         ) + { gridPos: { x: 0, y: 7, w: 8, h: 7 } },
 265         u.addTableSchema(
 266           '$datasource',
 267           '',
 268           { col: 3, desc: true },
 269           [
 270             RgwOverviewStyle('Pool', 'pool', 'string', 'short'),
 271             RgwOverviewStyle('Image', 'image', 'string', 'short'),
 272             RgwOverviewStyle('Throughput', 'Value', 'number', 'Bps'),
 273             RgwOverviewStyle('', '/.*/', 'hidden', 'short'),
 274           ],
 275           'Highest Throughput',
 276           'table'
 277         )
 278         .addTarget(
 279           u.addTargetSchema(
 280             'topk(10, sort(sum(irate(ceph_rbd_read_bytes[30s]) + irate(ceph_rbd_write_bytes[30s])) by (pool, image, namespace)))',
 281             1,
 282             'table',
 283             ''
 284           )
 285         ) + { gridPos: { x: 8, y: 7, w: 8, h: 7 } },
 286         u.addTableSchema(
 287           '$datasource',
 288           '',
 289           { col: 3, desc: true },
 290           [
 291             RgwOverviewStyle('Pool', 'pool', 'string', 'short'),
 292             RgwOverviewStyle('Image', 'image', 'string', 'short'),
 293             RgwOverviewStyle('Latency', 'Value', 'number', 'ns'),
 294             RgwOverviewStyle('', '/.*/', 'hidden', 'short'),
 295           ],
 296           'Highest Latency',
 297           'table'
 298         )
 299         .addTarget(
 300           u.addTargetSchema(
 301             'topk(10,\n  sum(\n    irate(ceph_rbd_write_latency_sum[30s]) / clamp_min(irate(ceph_rbd_write_latency_count[30s]), 1) +\n    irate(ceph_rbd_read_latency_sum[30s]) / clamp_min(irate(ceph_rbd_read_latency_count[30s]), 1)\n  ) by (pool, image, namespace)\n)',
 302             1,
 303             'table',
 304             ''
 305           )
 306         ) + { gridPos: { x: 16, y: 7, w: 8, h: 7 } },
 307       ]),
 308   },
 309 }