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

   1 local g = import 'grafonnet/grafana.libsonnet';
   2 local u = import 'utils.libsonnet';
   3
   4 {
   5   grafanaDashboards+:: {
   6     'pool-overview.json':
   7       local PoolOverviewSingleStatPanel(format,
   8                                         title,
   9                                         description,
  10                                         valueName,
  11                                         expr,
  12                                         targetFormat,
  13                                         x,
  14                                         y,
  15                                         w,
  16                                         h) =
  17         u.addSingleStatSchema(['#299c46', 'rgba(237, 129, 40, 0.89)', '#d44a3a'],
  18                               '$datasource',
  19                               format,
  20                               title,
  21                               description,
  22                               valueName,
  23                               false,
  24                               100,
  25                               false,
  26                               false,
  27                               '')
  28         .addTarget(u.addTargetSchema(expr, 1, targetFormat, '')) + { gridPos: { x: x, y: y, w: w, h: h } };
  29
  30       local PoolOverviewStyle(alias,
  31                               pattern,
  32                               type,
  33                               unit,
  34                               colorMode,
  35                               thresholds,
  36                               valueMaps) =
  37         u.addStyle(alias,
  38                    colorMode,
  39                    [
  40                      'rgba(245, 54, 54, 0.9)',
  41                      'rgba(237, 129, 40, 0.89)',
  42                      'rgba(50, 172, 45, 0.97)',
  43                    ],
  44                    'YYYY-MM-DD HH:mm:ss',
  45                    2,
  46                    1,
  47                    pattern,
  48                    thresholds,
  49                    type,
  50                    unit,
  51                    valueMaps);
  52
  53       local PoolOverviewGraphPanel(title,
  54                                    description,
  55                                    formatY1,
  56                                    labelY1,
  57                                    expr,
  58                                    targetFormat,
  59                                    legendFormat,
  60                                    x,
  61                                    y,
  62                                    w,
  63                                    h) =
  64         u.graphPanelSchema({},
  65                            title,
  66                            description,
  67                            'null as zero',
  68                            false,
  69                            formatY1,
  70                            'short',
  71                            labelY1,
  72                            null,
  73                            0,
  74                            1,
  75                            '$datasource')
  76         .addTargets(
  77           [u.addTargetSchema(expr,
  78                              1,
  79                              'time_series',
  80                              legendFormat)]
  81         ) + { gridPos: { x: x, y: y, w: w, h: h } };
  82
  83       u.dashboardSchema(
  84         'Ceph Pools Overview',
  85         '',
  86         'z99hzWtmk',
  87         'now-1h',
  88         '15s',
  89         22,
  90         [],
  91         '',
  92         { refresh_intervals: ['5s', '10s', '15s', '30s', '1m', '5m', '15m', '30m', '1h', '2h', '1d'], time_options: ['5m', '15m', '1h', '6h', '12h', '24h', '2d', '7d', '30d'] }
  93       )
  94       .addAnnotation(
  95         u.addAnnotationSchema(
  96           1,
  97           '-- Grafana --',
  98           true,
  99           true,
 100           'rgba(0, 211, 255, 1)',
 101           'Annotations & Alerts',
 102           'dashboard'
 103         )
 104       )
 105       .addTemplate(
 106         g.template.datasource('datasource',
 107                               'prometheus',
 108                               'Dashboard1',
 109                               label='Data Source')
 110       )
 111       .addTemplate(
 112         g.template.custom(label='TopK',
 113                           name='topk',
 114                           current='15',
 115                           query='15')
 116       )
 117       .addPanels([
 118         PoolOverviewSingleStatPanel(
 119           'none',
 120           'Pools',
 121           '',
 122           'avg',
 123           'count(ceph_pool_metadata)',
 124           'table',
 125           0,
 126           0,
 127           3,
 128           3
 129         ),
 130         PoolOverviewSingleStatPanel(
 131           'none',
 132           'Pools with Compression',
 133           'Count of the pools that have compression enabled',
 134           'current',
 135           'count(ceph_pool_metadata{compression_mode!="none"})',
 136           '',
 137           3,
 138           0,
 139           3,
 140           3
 141         ),
 142         PoolOverviewSingleStatPanel(
 143           'bytes',
 144           'Total Raw Capacity',
 145           'Total raw capacity available to the cluster',
 146           'current',
 147           'sum(ceph_osd_stat_bytes)',
 148           '',
 149           6,
 150           0,
 151           3,
 152           3
 153         ),
 154         PoolOverviewSingleStatPanel(
 155           'bytes',
 156           'Raw Capacity Consumed',
 157           'Total raw capacity consumed by user data and associated overheads (metadata + redundancy)',
 158           'current',
 159           'sum(ceph_pool_bytes_used)',
 160           '',
 161           9,
 162           0,
 163           3,
 164           3
 165         ),
 166         PoolOverviewSingleStatPanel(
 167           'bytes',
 168           'Logical Stored ',
 169           'Total of client data stored in the cluster',
 170           'current',
 171           'sum(ceph_pool_stored)',
 172           '',
 173           12,
 174           0,
 175           3,
 176           3
 177         ),
 178         PoolOverviewSingleStatPanel(
 179           'bytes',
 180           'Compression Savings',
 181           'A compression saving is determined as the data eligible to be compressed minus the capacity used to store the data after compression',
 182           'current',
 183           'sum(ceph_pool_compress_under_bytes - ceph_pool_compress_bytes_used)',
 184           '',
 185           15,
 186           0,
 187           3,
 188           3
 189         ),
 190         PoolOverviewSingleStatPanel(
 191           'percent',
 192           'Compression Eligibility',
 193           'Indicates how suitable the data is within the pools that are/have been enabled for compression - averaged across all pools holding compressed data\n',
 194           'current',
 195           '(sum(ceph_pool_compress_under_bytes > 0) / sum(ceph_pool_stored_raw and ceph_pool_compress_under_bytes > 0)) * 100',
 196           'table',
 197           18,
 198           0,
 199           3,
 200           3
 201         ),
 202         PoolOverviewSingleStatPanel(
 203           'none',
 204           'Compression Factor',
 205           'This factor describes the average ratio of data eligible to be compressed divided by the data actually stored. It does not account for data written that was ineligible for compression (too small, or compression yield too low)',
 206           'current',
 207           'sum(ceph_pool_compress_under_bytes > 0) / sum(ceph_pool_compress_bytes_used > 0)',
 208           '',
 209           21,
 210           0,
 211           3,
 212           3
 213         ),
 214         u.addTableSchema(
 215           '$datasource',
 216           '',
 217           { col: 5, desc: true },
 218           [
 219             PoolOverviewStyle('', 'Time', 'hidden', 'short', null, [], []),
 220             PoolOverviewStyle('', 'instance', 'hidden', 'short', null, [], []),
 221             PoolOverviewStyle('', 'job', 'hidden', 'short', null, [], []),
 222             PoolOverviewStyle('Pool Name', 'name', 'string', 'short', null, [], []),
 223             PoolOverviewStyle('Pool ID', 'pool_id', 'hidden', 'none', null, [], []),
 224             PoolOverviewStyle('Compression Factor', 'Value #A', 'number', 'none', null, [], []),
 225             PoolOverviewStyle('% Used', 'Value #D', 'number', 'percentunit', 'value', ['70', '85'], []),
 226             PoolOverviewStyle('Usable Free', 'Value #B', 'number', 'bytes', null, [], []),
 227             PoolOverviewStyle('Compression Eligibility', 'Value #C', 'number', 'percent', null, [], []),
 228             PoolOverviewStyle('Compression Savings', 'Value #E', 'number', 'bytes', null, [], []),
 229             PoolOverviewStyle('Growth (5d)', 'Value #F', 'number', 'bytes', 'value', ['0', '0'], []),
 230             PoolOverviewStyle('IOPS', 'Value #G', 'number', 'none', null, [], []),
 231             PoolOverviewStyle('Bandwidth', 'Value #H', 'number', 'Bps', null, [], []),
 232             PoolOverviewStyle('', '__name__', 'hidden', 'short', null, [], []),
 233             PoolOverviewStyle('', 'type', 'hidden', 'short', null, [], []),
 234             PoolOverviewStyle('', 'compression_mode', 'hidden', 'short', null, [], []),
 235             PoolOverviewStyle('Type', 'description', 'string', 'short', null, [], []),
 236             PoolOverviewStyle('Stored', 'Value #J', 'number', 'bytes', null, [], []),
 237             PoolOverviewStyle('', 'Value #I', 'hidden', 'short', null, [], []),
 238             PoolOverviewStyle('Compression', 'Value #K', 'string', 'short', null, [], [{ text: 'ON', value: '1' }]),
 239           ],
 240           'Pool Overview',
 241           'table'
 242         )
 243         .addTargets(
 244           [
 245             u.addTargetSchema(
 246               '(ceph_pool_compress_under_bytes / ceph_pool_compress_bytes_used > 0) and on(pool_id) (((ceph_pool_compress_under_bytes > 0) / ceph_pool_stored_raw) * 100 > 0.5)',
 247               1,
 248               'table',
 249               'A'
 250             ),
 251             u.addTargetSchema(
 252               'ceph_pool_max_avail * on(pool_id) group_left(name) ceph_pool_metadata',
 253               1,
 254               'table',
 255               'B'
 256             ),
 257             u.addTargetSchema(
 258               '((ceph_pool_compress_under_bytes > 0) / ceph_pool_stored_raw) * 100',
 259               1,
 260               'table',
 261               'C'
 262             ),
 263             u.addTargetSchema(
 264               '(ceph_pool_percent_used * on(pool_id) group_left(name) ceph_pool_metadata)',
 265               1,
 266               'table',
 267               'D'
 268             ),
 269             u.addTargetSchema(
 270               '(ceph_pool_compress_under_bytes - ceph_pool_compress_bytes_used > 0)',
 271               1,
 272               'table',
 273               'E'
 274             ),
 275             u.addTargetSchema(
 276               'delta(ceph_pool_stored[5d])', 1, 'table', 'F'
 277             ),
 278             u.addTargetSchema(
 279               'rate(ceph_pool_rd[30s]) + rate(ceph_pool_wr[30s])',
 280               1,
 281               'table',
 282               'G'
 283             ),
 284             u.addTargetSchema(
 285               'rate(ceph_pool_rd_bytes[30s]) + rate(ceph_pool_wr_bytes[30s])',
 286               1,
 287               'table',
 288               'H'
 289             ),
 290             u.addTargetSchema(
 291               'ceph_pool_metadata', 1, 'table', 'I'
 292             ),
 293             u.addTargetSchema(
 294               'ceph_pool_stored * on(pool_id) group_left ceph_pool_metadata',
 295               1,
 296               'table',
 297               'J'
 298             ),
 299             u.addTargetSchema(
 300               'ceph_pool_metadata{compression_mode!="none"}', 1, 'table', 'K'
 301             ),
 302             u.addTargetSchema('', '', '', 'L'),
 303           ]
 304         ) + { gridPos: { x: 0, y: 3, w: 24, h: 6 } },
 305         PoolOverviewGraphPanel(
 306           'Top $topk Client IOPS by Pool',
 307           'This chart shows the sum of read and write IOPS from all clients by pool',
 308           'short',
 309           'IOPS',
 310           'topk($topk,round((rate(ceph_pool_rd[30s]) + rate(ceph_pool_wr[30s])),1) * on(pool_id) group_left(instance,name) ceph_pool_metadata) ',
 311           'time_series',
 312           '{{name}} ',
 313           0,
 314           9,
 315           12,
 316           8
 317         )
 318         .addTarget(
 319           u.addTargetSchema(
 320             'topk($topk,rate(ceph_pool_wr[30s]) + on(pool_id) group_left(instance,name) ceph_pool_metadata) ',
 321             1,
 322             'time_series',
 323             '{{name}} - write'
 324           )
 325         ),
 326         PoolOverviewGraphPanel(
 327           'Top $topk Client Bandwidth by Pool',
 328           'The chart shows the sum of read and write bytes from all clients, by pool',
 329           'Bps',
 330           'Throughput',
 331           'topk($topk,(rate(ceph_pool_rd_bytes[30s]) + rate(ceph_pool_wr_bytes[30s])) * on(pool_id) group_left(instance,name) ceph_pool_metadata)',
 332           'time_series',
 333           '{{name}}',
 334           12,
 335           9,
 336           12,
 337           8
 338         ),
 339         PoolOverviewGraphPanel(
 340           'Pool Capacity Usage (RAW)',
 341           'Historical view of capacity usage, to help identify growth and trends in pool consumption',
 342           'bytes',
 343           'Capacity Used',
 344           'ceph_pool_bytes_used * on(pool_id) group_right ceph_pool_metadata',
 345           '',
 346           '{{name}}',
 347           0,
 348           17,
 349           24,
 350           7
 351         ),
 352       ]),
 353     'pool-detail.json':
 354       local PoolDetailSingleStatPanel(format,
 355                                       title,
 356                                       description,
 357                                       valueName,
 358                                       colorValue,
 359                                       gaugeMaxValue,
 360                                       gaugeShow,
 361                                       sparkLineShow,
 362                                       thresholds,
 363                                       expr,
 364                                       targetFormat,
 365                                       x,
 366                                       y,
 367                                       w,
 368                                       h) =
 369         u.addSingleStatSchema(['#299c46', 'rgba(237, 129, 40, 0.89)', '#d44a3a'],
 370                               '$datasource',
 371                               format,
 372                               title,
 373                               description,
 374                               valueName,
 375                               colorValue,
 376                               gaugeMaxValue,
 377                               gaugeShow,
 378                               sparkLineShow,
 379                               thresholds)
 380         .addTarget(u.addTargetSchema(expr, 1, targetFormat, '')) + { gridPos: { x: x, y: y, w: w, h: h } };
 381
 382       local PoolDetailGraphPanel(alias,
 383                                  title,
 384                                  description,
 385                                  formatY1,
 386                                  labelY1,
 387                                  expr,
 388                                  targetFormat,
 389                                  legendFormat,
 390                                  x,
 391                                  y,
 392                                  w,
 393                                  h) =
 394         u.graphPanelSchema(alias,
 395                            title,
 396                            description,
 397                            'null as zero',
 398                            false,
 399                            formatY1,
 400                            'short',
 401                            labelY1,
 402                            null,
 403                            null,
 404                            1,
 405                            '$datasource')
 406         .addTargets(
 407           [u.addTargetSchema(expr, 1, 'time_series', legendFormat)]
 408         ) + { gridPos: { x: x, y: y, w: w, h: h } };
 409
 410       u.dashboardSchema(
 411         'Ceph Pool Details',
 412         '',
 413         '-xyV8KCiz',
 414         'now-1h',
 415         '15s',
 416         22,
 417         [],
 418         '',
 419         {
 420           refresh_intervals: ['5s', '10s', '15s', '30s', '1m', '5m', '15m', '30m', '1h', '2h', '1d'],
 421           time_options: ['5m', '15m', '1h', '6h', '12h', '24h', '2d', '7d', '30d'],
 422         }
 423       )
 424       .addRequired(
 425         type='grafana', id='grafana', name='Grafana', version='5.3.2'
 426       )
 427       .addRequired(
 428         type='panel', id='graph', name='Graph', version='5.0.0'
 429       )
 430       .addRequired(
 431         type='panel', id='singlestat', name='Singlestat', version='5.0.0'
 432       )
 433       .addAnnotation(
 434         u.addAnnotationSchema(
 435           1,
 436           '-- Grafana --',
 437           true,
 438           true,
 439           'rgba(0, 211, 255, 1)',
 440           'Annotations & Alerts',
 441           'dashboard'
 442         )
 443       )
 444       .addTemplate(
 445         g.template.datasource('datasource',
 446                               'prometheus',
 447                               'Prometheus admin.virt1.home.fajerski.name:9090',
 448                               label='Data Source')
 449       )
 450       .addTemplate(
 451         u.addTemplateSchema('pool_name',
 452                             '$datasource',
 453                             'label_values(ceph_pool_metadata,name)',
 454                             1,
 455                             false,
 456                             1,
 457                             'Pool Name',
 458                             '')
 459       )
 460       .addPanels([
 461         PoolDetailSingleStatPanel(
 462           'percentunit',
 463           'Capacity used',
 464           '',
 465           'current',
 466           true,
 467           1,
 468           true,
 469           true,
 470           '.7,.8',
 471           '(ceph_pool_stored / (ceph_pool_stored + ceph_pool_max_avail)) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}',
 472           'time_series',
 473           0,
 474           0,
 475           7,
 476           7
 477         ),
 478         PoolDetailSingleStatPanel(
 479           's',
 480           'Time till full',
 481           'Time till pool is full assuming the average fill rate of the last 6 hours',
 482           false,
 483           100,
 484           false,
 485           false,
 486           '',
 487           'current',
 488           '(ceph_pool_max_avail / deriv(ceph_pool_stored[6h])) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"} > 0',
 489           'time_series',
 490           7,
 491           0,
 492           5,
 493           7
 494         ),
 495         PoolDetailGraphPanel(
 496           {
 497             read_op_per_sec:
 498               '#3F6833',
 499             write_op_per_sec: '#E5AC0E',
 500           },
 501           '$pool_name Object Ingress/Egress',
 502           '',
 503           'ops',
 504           'Objects out(-) / in(+) ',
 505           'deriv(ceph_pool_objects[1m]) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}',
 506           'time_series',
 507           'Objects per second',
 508           12,
 509           0,
 510           12,
 511           7
 512         ),
 513         PoolDetailGraphPanel(
 514           {
 515             read_op_per_sec: '#3F6833',
 516             write_op_per_sec: '#E5AC0E',
 517           }, '$pool_name Client IOPS', '', 'iops', 'Read (-) / Write (+)', 'irate(ceph_pool_rd[1m]) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}', 'time_series', 'reads', 0, 7, 12, 7
 518         )
 519         .addSeriesOverride({ alias: 'reads', transform: 'negative-Y' })
 520         .addTarget(
 521           u.addTargetSchema(
 522             'irate(ceph_pool_wr[1m]) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}', 1, 'time_series', 'writes'
 523           )
 524         ),
 525         PoolDetailGraphPanel(
 526           {
 527             read_op_per_sec: '#3F6833',
 528             write_op_per_sec: '#E5AC0E',
 529           },
 530           '$pool_name Client Throughput',
 531           '',
 532           'Bps',
 533           'Read (-) / Write (+)',
 534           'irate(ceph_pool_rd_bytes[1m]) + on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}',
 535           'time_series',
 536           'reads',
 537           12,
 538           7,
 539           12,
 540           7
 541         )
 542         .addSeriesOverride({ alias: 'reads', transform: 'negative-Y' })
 543         .addTarget(
 544           u.addTargetSchema(
 545             'irate(ceph_pool_wr_bytes[1m]) + on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}',
 546             1,
 547             'time_series',
 548             'writes'
 549           )
 550         ),
 551         PoolDetailGraphPanel(
 552           {
 553             read_op_per_sec: '#3F6833',
 554             write_op_per_sec: '#E5AC0E',
 555           },
 556           '$pool_name Objects',
 557           '',
 558           'short',
 559           'Objects',
 560           'ceph_pool_objects * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}',
 561           'time_series',
 562           'Number of Objects',
 563           0,
 564           14,
 565           12,
 566           7
 567         ),
 568       ]),
 569   },
 570 }