]>
Commit | Line | Data |
---|---|---|
20effc67 TL |
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, | |
1d09f67e | 12 | instant, |
20effc67 TL |
13 | targetFormat, |
14 | x, | |
15 | y, | |
16 | w, | |
17 | h) = | |
18 | u.addSingleStatSchema(['#299c46', 'rgba(237, 129, 40, 0.89)', '#d44a3a'], | |
19 | '$datasource', | |
20 | format, | |
21 | title, | |
22 | description, | |
23 | valueName, | |
24 | false, | |
25 | 100, | |
26 | false, | |
27 | false, | |
28 | '') | |
1d09f67e | 29 | .addTarget(u.addTargetSchema(expr, '', targetFormat, 1, instant)) + { gridPos: { x: x, y: y, w: w, h: h } }; |
20effc67 TL |
30 | |
31 | local PoolOverviewStyle(alias, | |
32 | pattern, | |
33 | type, | |
34 | unit, | |
35 | colorMode, | |
36 | thresholds, | |
37 | valueMaps) = | |
38 | u.addStyle(alias, | |
39 | colorMode, | |
40 | [ | |
41 | 'rgba(245, 54, 54, 0.9)', | |
42 | 'rgba(237, 129, 40, 0.89)', | |
43 | 'rgba(50, 172, 45, 0.97)', | |
44 | ], | |
45 | 'YYYY-MM-DD HH:mm:ss', | |
46 | 2, | |
47 | 1, | |
48 | pattern, | |
49 | thresholds, | |
50 | type, | |
51 | unit, | |
52 | valueMaps); | |
53 | ||
54 | local PoolOverviewGraphPanel(title, | |
55 | description, | |
56 | formatY1, | |
57 | labelY1, | |
58 | expr, | |
20effc67 TL |
59 | legendFormat, |
60 | x, | |
61 | y, | |
62 | w, | |
63 | h) = | |
64 | u.graphPanelSchema({}, | |
65 | title, | |
66 | description, | |
67 | 'null as zero', | |
68 | false, | |
69 | formatY1, | |
70 | 'short', | |
71 | labelY1, | |
72 | null, | |
73 | 0, | |
74 | 1, | |
75 | '$datasource') | |
76 | .addTargets( | |
77 | [u.addTargetSchema(expr, | |
20effc67 TL |
78 | legendFormat)] |
79 | ) + { gridPos: { x: x, y: y, w: w, h: h } }; | |
80 | ||
81 | u.dashboardSchema( | |
82 | 'Ceph Pools Overview', | |
83 | '', | |
84 | 'z99hzWtmk', | |
85 | 'now-1h', | |
86 | '15s', | |
87 | 22, | |
88 | [], | |
89 | '', | |
90 | { refresh_intervals: ['5s', '10s', '15s', '30s', '1m', '5m', '15m', '30m', '1h', '2h', '1d'], time_options: ['5m', '15m', '1h', '6h', '12h', '24h', '2d', '7d', '30d'] } | |
91 | ) | |
92 | .addAnnotation( | |
93 | u.addAnnotationSchema( | |
94 | 1, | |
95 | '-- Grafana --', | |
96 | true, | |
97 | true, | |
98 | 'rgba(0, 211, 255, 1)', | |
99 | 'Annotations & Alerts', | |
100 | 'dashboard' | |
101 | ) | |
102 | ) | |
103 | .addTemplate( | |
104 | g.template.datasource('datasource', | |
105 | 'prometheus', | |
106 | 'Dashboard1', | |
107 | label='Data Source') | |
108 | ) | |
109 | .addTemplate( | |
110 | g.template.custom(label='TopK', | |
111 | name='topk', | |
112 | current='15', | |
113 | query='15') | |
114 | ) | |
115 | .addPanels([ | |
116 | PoolOverviewSingleStatPanel( | |
117 | 'none', | |
118 | 'Pools', | |
119 | '', | |
120 | 'avg', | |
121 | 'count(ceph_pool_metadata)', | |
1d09f67e | 122 | true, |
20effc67 TL |
123 | 'table', |
124 | 0, | |
125 | 0, | |
126 | 3, | |
127 | 3 | |
128 | ), | |
129 | PoolOverviewSingleStatPanel( | |
130 | 'none', | |
131 | 'Pools with Compression', | |
132 | 'Count of the pools that have compression enabled', | |
133 | 'current', | |
134 | 'count(ceph_pool_metadata{compression_mode!="none"})', | |
1d09f67e | 135 | null, |
20effc67 TL |
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)', | |
1d09f67e | 148 | null, |
20effc67 TL |
149 | '', |
150 | 6, | |
151 | 0, | |
152 | 3, | |
153 | 3 | |
154 | ), | |
155 | PoolOverviewSingleStatPanel( | |
156 | 'bytes', | |
157 | 'Raw Capacity Consumed', | |
158 | 'Total raw capacity consumed by user data and associated overheads (metadata + redundancy)', | |
159 | 'current', | |
160 | 'sum(ceph_pool_bytes_used)', | |
1d09f67e | 161 | true, |
20effc67 TL |
162 | '', |
163 | 9, | |
164 | 0, | |
165 | 3, | |
166 | 3 | |
167 | ), | |
168 | PoolOverviewSingleStatPanel( | |
169 | 'bytes', | |
170 | 'Logical Stored ', | |
171 | 'Total of client data stored in the cluster', | |
172 | 'current', | |
173 | 'sum(ceph_pool_stored)', | |
1d09f67e | 174 | true, |
20effc67 TL |
175 | '', |
176 | 12, | |
177 | 0, | |
178 | 3, | |
179 | 3 | |
180 | ), | |
181 | PoolOverviewSingleStatPanel( | |
182 | 'bytes', | |
183 | 'Compression Savings', | |
184 | 'A compression saving is determined as the data eligible to be compressed minus the capacity used to store the data after compression', | |
185 | 'current', | |
186 | 'sum(ceph_pool_compress_under_bytes - ceph_pool_compress_bytes_used)', | |
1d09f67e | 187 | null, |
20effc67 TL |
188 | '', |
189 | 15, | |
190 | 0, | |
191 | 3, | |
192 | 3 | |
193 | ), | |
194 | PoolOverviewSingleStatPanel( | |
195 | 'percent', | |
196 | 'Compression Eligibility', | |
197 | 'Indicates how suitable the data is within the pools that are/have been enabled for compression - averaged across all pools holding compressed data\n', | |
198 | 'current', | |
199 | '(sum(ceph_pool_compress_under_bytes > 0) / sum(ceph_pool_stored_raw and ceph_pool_compress_under_bytes > 0)) * 100', | |
1d09f67e | 200 | null, |
20effc67 TL |
201 | 'table', |
202 | 18, | |
203 | 0, | |
204 | 3, | |
205 | 3 | |
206 | ), | |
207 | PoolOverviewSingleStatPanel( | |
208 | 'none', | |
209 | 'Compression Factor', | |
210 | '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)', | |
211 | 'current', | |
212 | 'sum(ceph_pool_compress_under_bytes > 0) / sum(ceph_pool_compress_bytes_used > 0)', | |
1d09f67e | 213 | null, |
20effc67 TL |
214 | '', |
215 | 21, | |
216 | 0, | |
217 | 3, | |
218 | 3 | |
219 | ), | |
220 | u.addTableSchema( | |
221 | '$datasource', | |
222 | '', | |
223 | { col: 5, desc: true }, | |
224 | [ | |
225 | PoolOverviewStyle('', 'Time', 'hidden', 'short', null, [], []), | |
226 | PoolOverviewStyle('', 'instance', 'hidden', 'short', null, [], []), | |
227 | PoolOverviewStyle('', 'job', 'hidden', 'short', null, [], []), | |
228 | PoolOverviewStyle('Pool Name', 'name', 'string', 'short', null, [], []), | |
229 | PoolOverviewStyle('Pool ID', 'pool_id', 'hidden', 'none', null, [], []), | |
230 | PoolOverviewStyle('Compression Factor', 'Value #A', 'number', 'none', null, [], []), | |
231 | PoolOverviewStyle('% Used', 'Value #D', 'number', 'percentunit', 'value', ['70', '85'], []), | |
232 | PoolOverviewStyle('Usable Free', 'Value #B', 'number', 'bytes', null, [], []), | |
233 | PoolOverviewStyle('Compression Eligibility', 'Value #C', 'number', 'percent', null, [], []), | |
234 | PoolOverviewStyle('Compression Savings', 'Value #E', 'number', 'bytes', null, [], []), | |
235 | PoolOverviewStyle('Growth (5d)', 'Value #F', 'number', 'bytes', 'value', ['0', '0'], []), | |
236 | PoolOverviewStyle('IOPS', 'Value #G', 'number', 'none', null, [], []), | |
237 | PoolOverviewStyle('Bandwidth', 'Value #H', 'number', 'Bps', null, [], []), | |
238 | PoolOverviewStyle('', '__name__', 'hidden', 'short', null, [], []), | |
239 | PoolOverviewStyle('', 'type', 'hidden', 'short', null, [], []), | |
240 | PoolOverviewStyle('', 'compression_mode', 'hidden', 'short', null, [], []), | |
241 | PoolOverviewStyle('Type', 'description', 'string', 'short', null, [], []), | |
242 | PoolOverviewStyle('Stored', 'Value #J', 'number', 'bytes', null, [], []), | |
243 | PoolOverviewStyle('', 'Value #I', 'hidden', 'short', null, [], []), | |
244 | PoolOverviewStyle('Compression', 'Value #K', 'string', 'short', null, [], [{ text: 'ON', value: '1' }]), | |
245 | ], | |
246 | 'Pool Overview', | |
247 | 'table' | |
248 | ) | |
249 | .addTargets( | |
250 | [ | |
251 | u.addTargetSchema( | |
252 | '(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)', | |
1d09f67e | 253 | 'A', |
20effc67 | 254 | 'table', |
1d09f67e TL |
255 | 1, |
256 | true | |
20effc67 TL |
257 | ), |
258 | u.addTargetSchema( | |
259 | 'ceph_pool_max_avail * on(pool_id) group_left(name) ceph_pool_metadata', | |
1d09f67e | 260 | 'B', |
20effc67 | 261 | 'table', |
1d09f67e TL |
262 | 1, |
263 | true | |
20effc67 TL |
264 | ), |
265 | u.addTargetSchema( | |
266 | '((ceph_pool_compress_under_bytes > 0) / ceph_pool_stored_raw) * 100', | |
1d09f67e | 267 | 'C', |
20effc67 | 268 | 'table', |
1d09f67e TL |
269 | 1, |
270 | true | |
20effc67 TL |
271 | ), |
272 | u.addTargetSchema( | |
273 | '(ceph_pool_percent_used * on(pool_id) group_left(name) ceph_pool_metadata)', | |
1d09f67e | 274 | 'D', |
20effc67 | 275 | 'table', |
1d09f67e TL |
276 | 1, |
277 | true | |
20effc67 TL |
278 | ), |
279 | u.addTargetSchema( | |
280 | '(ceph_pool_compress_under_bytes - ceph_pool_compress_bytes_used > 0)', | |
1d09f67e | 281 | 'E', |
20effc67 | 282 | 'table', |
1d09f67e TL |
283 | 1, |
284 | true | |
20effc67 TL |
285 | ), |
286 | u.addTargetSchema( | |
1d09f67e | 287 | 'delta(ceph_pool_stored[5d])', 'F', 'table', 1, true |
20effc67 TL |
288 | ), |
289 | u.addTargetSchema( | |
290 | 'rate(ceph_pool_rd[30s]) + rate(ceph_pool_wr[30s])', | |
1d09f67e | 291 | 'G', |
20effc67 | 292 | 'table', |
1d09f67e TL |
293 | 1, |
294 | true | |
20effc67 TL |
295 | ), |
296 | u.addTargetSchema( | |
297 | 'rate(ceph_pool_rd_bytes[30s]) + rate(ceph_pool_wr_bytes[30s])', | |
1d09f67e | 298 | 'H', |
20effc67 | 299 | 'table', |
1d09f67e TL |
300 | 1, |
301 | true | |
20effc67 TL |
302 | ), |
303 | u.addTargetSchema( | |
1d09f67e | 304 | 'ceph_pool_metadata', 'I', 'table', 1, true |
20effc67 TL |
305 | ), |
306 | u.addTargetSchema( | |
307 | 'ceph_pool_stored * on(pool_id) group_left ceph_pool_metadata', | |
1d09f67e | 308 | 'J', |
20effc67 | 309 | 'table', |
1d09f67e TL |
310 | 1, |
311 | true | |
20effc67 TL |
312 | ), |
313 | u.addTargetSchema( | |
1d09f67e | 314 | 'ceph_pool_metadata{compression_mode!="none"}', 'K', 'table', 1, true |
20effc67 | 315 | ), |
1d09f67e | 316 | u.addTargetSchema('', 'L', '', '', null), |
20effc67 TL |
317 | ] |
318 | ) + { gridPos: { x: 0, y: 3, w: 24, h: 6 } }, | |
319 | PoolOverviewGraphPanel( | |
320 | 'Top $topk Client IOPS by Pool', | |
321 | 'This chart shows the sum of read and write IOPS from all clients by pool', | |
322 | 'short', | |
323 | 'IOPS', | |
324 | 'topk($topk,round((rate(ceph_pool_rd[30s]) + rate(ceph_pool_wr[30s])),1) * on(pool_id) group_left(instance,name) ceph_pool_metadata) ', | |
20effc67 TL |
325 | '{{name}} ', |
326 | 0, | |
327 | 9, | |
328 | 12, | |
329 | 8 | |
330 | ) | |
331 | .addTarget( | |
332 | u.addTargetSchema( | |
333 | 'topk($topk,rate(ceph_pool_wr[30s]) + on(pool_id) group_left(instance,name) ceph_pool_metadata) ', | |
20effc67 TL |
334 | '{{name}} - write' |
335 | ) | |
336 | ), | |
337 | PoolOverviewGraphPanel( | |
338 | 'Top $topk Client Bandwidth by Pool', | |
339 | 'The chart shows the sum of read and write bytes from all clients, by pool', | |
340 | 'Bps', | |
341 | 'Throughput', | |
342 | 'topk($topk,(rate(ceph_pool_rd_bytes[30s]) + rate(ceph_pool_wr_bytes[30s])) * on(pool_id) group_left(instance,name) ceph_pool_metadata)', | |
20effc67 TL |
343 | '{{name}}', |
344 | 12, | |
345 | 9, | |
346 | 12, | |
347 | 8 | |
348 | ), | |
349 | PoolOverviewGraphPanel( | |
350 | 'Pool Capacity Usage (RAW)', | |
351 | 'Historical view of capacity usage, to help identify growth and trends in pool consumption', | |
352 | 'bytes', | |
353 | 'Capacity Used', | |
354 | 'ceph_pool_bytes_used * on(pool_id) group_right ceph_pool_metadata', | |
20effc67 TL |
355 | '{{name}}', |
356 | 0, | |
357 | 17, | |
358 | 24, | |
359 | 7 | |
360 | ), | |
361 | ]), | |
362 | 'pool-detail.json': | |
363 | local PoolDetailSingleStatPanel(format, | |
364 | title, | |
365 | description, | |
366 | valueName, | |
367 | colorValue, | |
368 | gaugeMaxValue, | |
369 | gaugeShow, | |
370 | sparkLineShow, | |
371 | thresholds, | |
372 | expr, | |
373 | targetFormat, | |
374 | x, | |
375 | y, | |
376 | w, | |
377 | h) = | |
378 | u.addSingleStatSchema(['#299c46', 'rgba(237, 129, 40, 0.89)', '#d44a3a'], | |
379 | '$datasource', | |
380 | format, | |
381 | title, | |
382 | description, | |
383 | valueName, | |
384 | colorValue, | |
385 | gaugeMaxValue, | |
386 | gaugeShow, | |
387 | sparkLineShow, | |
388 | thresholds) | |
1d09f67e | 389 | .addTarget(u.addTargetSchema(expr, '', targetFormat)) + { gridPos: { x: x, y: y, w: w, h: h } }; |
20effc67 TL |
390 | |
391 | local PoolDetailGraphPanel(alias, | |
392 | title, | |
393 | description, | |
394 | formatY1, | |
395 | labelY1, | |
396 | expr, | |
20effc67 TL |
397 | legendFormat, |
398 | x, | |
399 | y, | |
400 | w, | |
401 | h) = | |
402 | u.graphPanelSchema(alias, | |
403 | title, | |
404 | description, | |
405 | 'null as zero', | |
406 | false, | |
407 | formatY1, | |
408 | 'short', | |
409 | labelY1, | |
410 | null, | |
411 | null, | |
412 | 1, | |
413 | '$datasource') | |
414 | .addTargets( | |
1d09f67e | 415 | [u.addTargetSchema(expr, legendFormat)] |
20effc67 TL |
416 | ) + { gridPos: { x: x, y: y, w: w, h: h } }; |
417 | ||
418 | u.dashboardSchema( | |
419 | 'Ceph Pool Details', | |
420 | '', | |
421 | '-xyV8KCiz', | |
422 | 'now-1h', | |
423 | '15s', | |
424 | 22, | |
425 | [], | |
426 | '', | |
427 | { | |
428 | refresh_intervals: ['5s', '10s', '15s', '30s', '1m', '5m', '15m', '30m', '1h', '2h', '1d'], | |
429 | time_options: ['5m', '15m', '1h', '6h', '12h', '24h', '2d', '7d', '30d'], | |
430 | } | |
431 | ) | |
432 | .addRequired( | |
433 | type='grafana', id='grafana', name='Grafana', version='5.3.2' | |
434 | ) | |
435 | .addRequired( | |
436 | type='panel', id='graph', name='Graph', version='5.0.0' | |
437 | ) | |
438 | .addRequired( | |
439 | type='panel', id='singlestat', name='Singlestat', version='5.0.0' | |
440 | ) | |
441 | .addAnnotation( | |
442 | u.addAnnotationSchema( | |
443 | 1, | |
444 | '-- Grafana --', | |
445 | true, | |
446 | true, | |
447 | 'rgba(0, 211, 255, 1)', | |
448 | 'Annotations & Alerts', | |
449 | 'dashboard' | |
450 | ) | |
451 | ) | |
452 | .addTemplate( | |
453 | g.template.datasource('datasource', | |
454 | 'prometheus', | |
455 | 'Prometheus admin.virt1.home.fajerski.name:9090', | |
456 | label='Data Source') | |
457 | ) | |
458 | .addTemplate( | |
459 | u.addTemplateSchema('pool_name', | |
460 | '$datasource', | |
461 | 'label_values(ceph_pool_metadata,name)', | |
462 | 1, | |
463 | false, | |
464 | 1, | |
465 | 'Pool Name', | |
466 | '') | |
467 | ) | |
468 | .addPanels([ | |
469 | PoolDetailSingleStatPanel( | |
470 | 'percentunit', | |
471 | 'Capacity used', | |
472 | '', | |
473 | 'current', | |
474 | true, | |
475 | 1, | |
476 | true, | |
477 | true, | |
478 | '.7,.8', | |
479 | '(ceph_pool_stored / (ceph_pool_stored + ceph_pool_max_avail)) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}', | |
480 | 'time_series', | |
481 | 0, | |
482 | 0, | |
483 | 7, | |
484 | 7 | |
485 | ), | |
486 | PoolDetailSingleStatPanel( | |
487 | 's', | |
488 | 'Time till full', | |
489 | 'Time till pool is full assuming the average fill rate of the last 6 hours', | |
490 | false, | |
491 | 100, | |
492 | false, | |
493 | false, | |
494 | '', | |
495 | 'current', | |
496 | '(ceph_pool_max_avail / deriv(ceph_pool_stored[6h])) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"} > 0', | |
497 | 'time_series', | |
498 | 7, | |
499 | 0, | |
500 | 5, | |
501 | 7 | |
502 | ), | |
503 | PoolDetailGraphPanel( | |
504 | { | |
505 | read_op_per_sec: | |
506 | '#3F6833', | |
507 | write_op_per_sec: '#E5AC0E', | |
508 | }, | |
509 | '$pool_name Object Ingress/Egress', | |
510 | '', | |
511 | 'ops', | |
512 | 'Objects out(-) / in(+) ', | |
513 | 'deriv(ceph_pool_objects[1m]) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}', | |
20effc67 TL |
514 | 'Objects per second', |
515 | 12, | |
516 | 0, | |
517 | 12, | |
518 | 7 | |
519 | ), | |
520 | PoolDetailGraphPanel( | |
521 | { | |
522 | read_op_per_sec: '#3F6833', | |
523 | write_op_per_sec: '#E5AC0E', | |
1d09f67e | 524 | }, '$pool_name Client IOPS', '', 'iops', 'Read (-) / Write (+)', 'irate(ceph_pool_rd[1m]) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}', 'reads', 0, 7, 12, 7 |
20effc67 TL |
525 | ) |
526 | .addSeriesOverride({ alias: 'reads', transform: 'negative-Y' }) | |
527 | .addTarget( | |
528 | u.addTargetSchema( | |
1d09f67e | 529 | 'irate(ceph_pool_wr[1m]) * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}', 'writes' |
20effc67 TL |
530 | ) |
531 | ), | |
532 | PoolDetailGraphPanel( | |
533 | { | |
534 | read_op_per_sec: '#3F6833', | |
535 | write_op_per_sec: '#E5AC0E', | |
536 | }, | |
537 | '$pool_name Client Throughput', | |
538 | '', | |
539 | 'Bps', | |
540 | 'Read (-) / Write (+)', | |
541 | 'irate(ceph_pool_rd_bytes[1m]) + on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}', | |
20effc67 TL |
542 | 'reads', |
543 | 12, | |
544 | 7, | |
545 | 12, | |
546 | 7 | |
547 | ) | |
548 | .addSeriesOverride({ alias: 'reads', transform: 'negative-Y' }) | |
549 | .addTarget( | |
550 | u.addTargetSchema( | |
551 | 'irate(ceph_pool_wr_bytes[1m]) + on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}', | |
20effc67 TL |
552 | 'writes' |
553 | ) | |
554 | ), | |
555 | PoolDetailGraphPanel( | |
556 | { | |
557 | read_op_per_sec: '#3F6833', | |
558 | write_op_per_sec: '#E5AC0E', | |
559 | }, | |
560 | '$pool_name Objects', | |
561 | '', | |
562 | 'short', | |
563 | 'Objects', | |
564 | 'ceph_pool_objects * on(pool_id) group_left(instance,name) ceph_pool_metadata{name=~"$pool_name"}', | |
20effc67 TL |
565 | 'Number of Objects', |
566 | 0, | |
567 | 14, | |
568 | 12, | |
569 | 7 | |
570 | ), | |
571 | ]), | |
572 | }, | |
573 | } |