]> git.proxmox.com Git - ceph.git/blame - ceph/monitoring/grafana/dashboards/osds-overview.json
import new upstream nautilus stable release 14.2.8
[ceph.git] / ceph / monitoring / grafana / dashboards / osds-overview.json
CommitLineData
11fdf7f2
TL
1{
2
3 "__requires": [
4 {
5 "type": "grafana",
6 "id": "grafana",
7 "name": "Grafana",
8 "version": "5.0.0"
9 },
10 {
11 "type": "panel",
12 "id": "grafana-piechart-panel",
13 "name": "Pie Chart",
14 "version": "1.3.3"
15 },
16 {
17 "type": "panel",
18 "id": "graph",
19 "name": "Graph",
20 "version": "5.0.0"
21 },
22 {
23 "type": "panel",
24 "id": "table",
25 "name": "Table",
26 "version": "5.0.0"
27 }
28 ],
29 "annotations": {
30 "list": [
31 {
32 "builtIn": 1,
33 "datasource": "-- Grafana --",
34 "enable": true,
35 "hide": true,
36 "iconColor": "rgba(0, 211, 255, 1)",
37 "name": "Annotations & Alerts",
38 "type": "dashboard"
39 }
40 ]
41 },
42 "editable": false,
43 "gnetId": null,
44 "graphTooltip": 0,
45 "id": null,
46 "iteration": 1538083987689,
47 "links": [],
48 "panels": [
49 {
50 "aliasColors": {
51 "@95%ile": "#e0752d"
52 },
53 "bars": false,
54 "dashLength": 10,
55 "dashes": false,
56 "datasource": "$datasource",
57 "fill": 1,
58 "gridPos": {
59 "h": 8,
60 "w": 8,
61 "x": 0,
62 "y": 0
63 },
64 "id": 12,
65 "legend": {
66 "avg": false,
67 "current": true,
68 "max": false,
69 "min": false,
70 "show": true,
71 "total": false,
72 "values": true
73 },
74 "lines": true,
75 "linewidth": 1,
76 "links": [],
77 "nullPointMode": "null",
78 "percentage": false,
79 "pointradius": 5,
80 "points": false,
81 "renderer": "flot",
82 "seriesOverrides": [],
83 "spaceLength": 10,
84 "stack": false,
85 "steppedLine": false,
86 "targets": [
87 {
88 "expr": "avg (irate(ceph_osd_op_r_latency_sum[1m]) / on (ceph_daemon) irate(ceph_osd_op_r_latency_count[1m]) * 1000)",
89 "format": "time_series",
90 "intervalFactor": 1,
91 "legendFormat": "AVG read",
92 "refId": "A"
93 },
94 {
95 "expr": "max (irate(ceph_osd_op_r_latency_sum[1m]) / on (ceph_daemon) irate(ceph_osd_op_r_latency_count[1m]) * 1000)",
96 "format": "time_series",
97 "intervalFactor": 1,
98 "legendFormat": "MAX read",
99 "refId": "B"
100 },
101 {
102 "expr": "quantile(0.95,\n (irate(ceph_osd_op_r_latency_sum[1m]) / on (ceph_daemon) irate(ceph_osd_op_r_latency_count[1m]) * 1000)\n)",
103 "format": "time_series",
104 "intervalFactor": 1,
105 "legendFormat": "@95%ile",
106 "refId": "C"
107 }
108 ],
109 "thresholds": [],
110 "timeFrom": null,
111 "timeShift": null,
112 "title": "OSD Read Latencies",
113 "tooltip": {
114 "shared": true,
eafe8130 115 "sort": 2,
11fdf7f2
TL
116 "value_type": "individual"
117 },
118 "type": "graph",
119 "xaxis": {
120 "buckets": null,
121 "mode": "time",
122 "name": null,
123 "show": true,
124 "values": []
125 },
126 "yaxes": [
127 {
128 "format": "ms",
129 "label": null,
130 "logBase": 1,
131 "max": null,
132 "min": "0",
133 "show": true
134 },
135 {
136 "format": "short",
137 "label": null,
138 "logBase": 1,
139 "max": null,
140 "min": null,
141 "show": false
142 }
143 ]
144 },
145 {
146 "columns": [],
147 "datasource": "$datasource",
148 "description": "This table shows the osd's that are delivering the 10 highest read latencies within the cluster",
149 "fontSize": "100%",
150 "gridPos": {
151 "h": 8,
152 "w": 4,
153 "x": 8,
154 "y": 0
155 },
156 "id": 15,
157 "links": [],
158 "pageSize": null,
159 "scroll": true,
160 "showHeader": true,
161 "sort": {
162 "col": 2,
163 "desc": true
164 },
165 "styles": [
166 {
167 "alias": "OSD ID",
168 "colorMode": null,
169 "colors": [
170 "rgba(245, 54, 54, 0.9)",
171 "rgba(237, 129, 40, 0.89)",
172 "rgba(50, 172, 45, 0.97)"
173 ],
174 "dateFormat": "YYYY-MM-DD HH:mm:ss",
175 "decimals": 2,
176 "pattern": "ceph_daemon",
177 "thresholds": [],
178 "type": "string",
179 "unit": "short"
180 },
181 {
182 "alias": "Latency (ms)",
183 "colorMode": null,
184 "colors": [
185 "rgba(245, 54, 54, 0.9)",
186 "rgba(237, 129, 40, 0.89)",
187 "rgba(50, 172, 45, 0.97)"
188 ],
189 "dateFormat": "YYYY-MM-DD HH:mm:ss",
190 "decimals": 0,
191 "pattern": "Value",
192 "thresholds": [],
193 "type": "number",
194 "unit": "none"
195 },
196 {
197 "alias": "",
198 "colorMode": null,
199 "colors": [
200 "rgba(245, 54, 54, 0.9)",
201 "rgba(237, 129, 40, 0.89)",
202 "rgba(50, 172, 45, 0.97)"
203 ],
204 "dateFormat": "YYYY-MM-DD HH:mm:ss",
205 "decimals": 2,
206 "pattern": "/.*/",
207 "thresholds": [],
208 "type": "hidden",
209 "unit": "short"
210 }
211 ],
212 "targets": [
213 {
214 "expr": "topk(10,\n (sort(\n (irate(ceph_osd_op_r_latency_sum[1m]) / on (ceph_daemon) irate(ceph_osd_op_r_latency_count[1m]) * 1000)\n ))\n)\n\n",
215 "format": "table",
216 "instant": true,
217 "intervalFactor": 1,
218 "legendFormat": "",
219 "refId": "A"
220 }
221 ],
222 "title": "Highest READ Latencies",
223 "transform": "table",
224 "type": "table"
225 },
226 {
227 "aliasColors": {
228 "@95%ile write": "#e0752d"
229 },
230 "bars": false,
231 "dashLength": 10,
232 "dashes": false,
233 "datasource": "$datasource",
234 "fill": 1,
235 "gridPos": {
236 "h": 8,
237 "w": 8,
238 "x": 12,
239 "y": 0
240 },
241 "id": 13,
242 "legend": {
243 "avg": false,
244 "current": true,
245 "max": false,
246 "min": false,
247 "show": true,
248 "total": false,
249 "values": true
250 },
251 "lines": true,
252 "linewidth": 1,
253 "links": [],
254 "nullPointMode": "null",
255 "percentage": false,
256 "pointradius": 5,
257 "points": false,
258 "renderer": "flot",
259 "seriesOverrides": [],
260 "spaceLength": 10,
261 "stack": false,
262 "steppedLine": false,
263 "targets": [
264 {
265 "expr": "avg (irate(ceph_osd_op_w_latency_sum[1m]) / on (ceph_daemon) irate(ceph_osd_op_w_latency_count[1m]) * 1000)",
266 "format": "time_series",
267 "intervalFactor": 1,
268 "legendFormat": "AVG write",
269 "refId": "A"
270 },
271 {
272 "expr": "max (irate(ceph_osd_op_w_latency_sum[1m]) / on (ceph_daemon) irate(ceph_osd_op_w_latency_count[1m]) * 1000)",
273 "format": "time_series",
274 "intervalFactor": 1,
275 "legendFormat": "MAX write",
276 "refId": "B"
277 },
278 {
279 "expr": "quantile(0.95,\n (irate(ceph_osd_op_w_latency_sum[1m]) / on (ceph_daemon) irate(ceph_osd_op_w_latency_count[1m]) * 1000)\n)",
280 "format": "time_series",
281 "intervalFactor": 1,
282 "legendFormat": "@95%ile write",
283 "refId": "C"
284 }
285 ],
286 "thresholds": [],
287 "timeFrom": null,
288 "timeShift": null,
289 "title": "OSD Write Latencies",
290 "tooltip": {
291 "shared": true,
eafe8130 292 "sort": 2,
11fdf7f2
TL
293 "value_type": "individual"
294 },
295 "type": "graph",
296 "xaxis": {
297 "buckets": null,
298 "mode": "time",
299 "name": null,
300 "show": true,
301 "values": []
302 },
303 "yaxes": [
304 {
305 "format": "ms",
306 "label": null,
307 "logBase": 1,
308 "max": null,
309 "min": "0",
310 "show": true
311 },
312 {
313 "format": "short",
314 "label": null,
315 "logBase": 1,
316 "max": null,
317 "min": null,
318 "show": false
319 }
320 ]
321 },
322 {
323 "columns": [],
324 "datasource": "$datasource",
325 "description": "This table shows the osd's that are delivering the 10 highest write latencies within the cluster",
326 "fontSize": "100%",
327 "gridPos": {
328 "h": 8,
329 "w": 4,
330 "x": 20,
331 "y": 0
332 },
333 "id": 16,
334 "links": [],
335 "pageSize": null,
336 "scroll": true,
337 "showHeader": true,
338 "sort": {
339 "col": 2,
340 "desc": true
341 },
342 "styles": [
343 {
344 "alias": "OSD ID",
345 "colorMode": null,
346 "colors": [
347 "rgba(245, 54, 54, 0.9)",
348 "rgba(237, 129, 40, 0.89)",
349 "rgba(50, 172, 45, 0.97)"
350 ],
351 "dateFormat": "YYYY-MM-DD HH:mm:ss",
352 "decimals": 2,
353 "pattern": "ceph_daemon",
354 "thresholds": [],
355 "type": "string",
356 "unit": "short"
357 },
358 {
359 "alias": "Latency (ms)",
360 "colorMode": null,
361 "colors": [
362 "rgba(245, 54, 54, 0.9)",
363 "rgba(237, 129, 40, 0.89)",
364 "rgba(50, 172, 45, 0.97)"
365 ],
366 "dateFormat": "YYYY-MM-DD HH:mm:ss",
367 "decimals": 0,
368 "pattern": "Value",
369 "thresholds": [],
370 "type": "number",
371 "unit": "none"
372 },
373 {
374 "alias": "",
375 "colorMode": null,
376 "colors": [
377 "rgba(245, 54, 54, 0.9)",
378 "rgba(237, 129, 40, 0.89)",
379 "rgba(50, 172, 45, 0.97)"
380 ],
381 "dateFormat": "YYYY-MM-DD HH:mm:ss",
382 "decimals": 2,
383 "pattern": "/.*/",
384 "thresholds": [],
385 "type": "hidden",
386 "unit": "short"
387 }
388 ],
389 "targets": [
390 {
391 "expr": "topk(10,\n (sort(\n (irate(ceph_osd_op_w_latency_sum[1m]) / on (ceph_daemon) irate(ceph_osd_op_w_latency_count[1m]) * 1000)\n ))\n)\n\n",
392 "format": "table",
393 "instant": true,
394 "intervalFactor": 1,
395 "legendFormat": "",
396 "refId": "A"
397 }
398 ],
399 "title": "Highest WRITE Latencies",
400 "transform": "table",
401 "type": "table"
402 },
403 {
404 "aliasColors": {},
405 "breakPoint": "50%",
406 "cacheTimeout": null,
407 "combine": {
408 "label": "Others",
409 "threshold": 0
410 },
411 "datasource": "$datasource",
412 "fontSize": "80%",
413 "format": "none",
414 "gridPos": {
415 "h": 8,
416 "w": 4,
417 "x": 0,
418 "y": 8
419 },
420 "id": 2,
421 "interval": null,
422 "legend": {
423 "show": true,
424 "values": true
425 },
426 "legendType": "Under graph",
427 "links": [],
428 "maxDataPoints": 3,
429 "nullPointMode": "connected",
430 "pieType": "pie",
431 "strokeWidth": 1,
432 "targets": [
433 {
434 "expr": "count by(device_class) (ceph_osd_metadata)",
435 "format": "time_series",
436 "intervalFactor": 1,
437 "legendFormat": "{{device_class}}",
438 "refId": "A"
439 }
440 ],
441 "title": "OSD Types Summary",
442 "type": "grafana-piechart-panel",
443 "valueName": "current"
444 },
445 {
446 "aliasColors": {
447 "Non-Encrypted": "#E5AC0E"
448 },
449 "breakPoint": "50%",
450 "cacheTimeout": null,
451 "combine": {
452 "label": "Others",
453 "threshold": 0
454 },
455 "datasource": "$datasource",
456 "fontSize": "80%",
457 "format": "none",
458 "gridPos": {
459 "h": 8,
460 "w": 4,
461 "x": 4,
462 "y": 8
463 },
464 "height": "200px",
465 "hideTimeOverride": true,
466 "id": 4,
467 "interval": null,
468 "legend": {
469 "percentage": false,
470 "show": true,
471 "values": true
472 },
473 "legendType": "Under graph",
474 "links": [],
475 "maxDataPoints": "1",
476 "minSpan": 4,
477 "nullPointMode": "connected",
478 "pieType": "pie",
479 "strokeWidth": 1,
480 "targets": [
481 {
482 "expr": "count(ceph_bluefs_wal_total_bytes)",
483 "format": "time_series",
484 "intervalFactor": 2,
485 "legendFormat": "bluestore",
486 "refId": "A",
487 "step": 240
488 },
489 {
490 "expr": "count(ceph_osd_metadata) - count(ceph_bluefs_wal_total_bytes)",
491 "format": "time_series",
492 "intervalFactor": 2,
493 "legendFormat": "filestore",
494 "refId": "B",
495 "step": 240
496 },
497 {
498 "expr": "absent(ceph_bluefs_wal_total_bytes)*count(ceph_osd_metadata)",
499 "format": "time_series",
500 "intervalFactor": 2,
501 "legendFormat": "filestore",
502 "refId": "C",
503 "step": 240
504 }
505 ],
92f5a8d4 506 "timeFrom": null,
11fdf7f2
TL
507 "timeShift": null,
508 "title": "OSD Objectstore Types",
509 "type": "grafana-piechart-panel",
510 "valueName": "current"
511 },
512 {
513 "aliasColors": {},
514 "breakPoint": "50%",
515 "cacheTimeout": null,
516 "combine": {
517 "label": "Others",
518 "threshold": "0.05"
519 },
520 "datasource": "$datasource",
521 "description": "The pie chart shows the various OSD sizes used within the cluster",
522 "fontSize": "80%",
523 "format": "none",
524 "gridPos": {
525 "h": 8,
526 "w": 4,
527 "x": 8,
528 "y": 8
529 },
530 "height": "220",
531 "hideTimeOverride": true,
532 "id": 8,
533 "interval": null,
534 "legend": {
535 "header": "",
536 "percentage": false,
537 "show": true,
538 "sideWidth": null,
539 "sortDesc": true,
540 "values": true
541 },
542 "legendType": "Under graph",
543 "links": [],
544 "maxDataPoints": "",
545 "minSpan": 6,
546 "nullPointMode": "connected",
547 "pieType": "pie",
548 "strokeWidth": "1",
549 "targets": [
550 {
551 "expr": "count(ceph_osd_stat_bytes < 1099511627776)",
552 "format": "time_series",
553 "intervalFactor": 2,
554 "legendFormat": "<1 TB",
555 "refId": "A",
556 "step": 2
557 },
558 {
559 "expr": "count(ceph_osd_stat_bytes >= 1099511627776 < 2199023255552)",
560 "format": "time_series",
561 "intervalFactor": 2,
562 "legendFormat": "<2 TB",
563 "refId": "B",
564 "step": 2
565 },
566 {
567 "expr": "count(ceph_osd_stat_bytes >= 2199023255552 < 3298534883328)",
568 "format": "time_series",
569 "intervalFactor": 2,
570 "legendFormat": "<3TB",
571 "refId": "C",
572 "step": 2
573 },
574 {
575 "expr": "count(ceph_osd_stat_bytes >= 3298534883328 < 4398046511104)",
576 "format": "time_series",
577 "intervalFactor": 2,
578 "legendFormat": "<4TB",
579 "refId": "D",
580 "step": 2
581 },
582 {
583 "expr": "count(ceph_osd_stat_bytes >= 4398046511104 < 6597069766656)",
584 "format": "time_series",
585 "intervalFactor": 2,
586 "legendFormat": "<6TB",
587 "refId": "E",
588 "step": 2
589 },
590 {
591 "expr": "count(ceph_osd_stat_bytes >= 6597069766656 < 8796093022208)",
592 "format": "time_series",
593 "intervalFactor": 2,
594 "legendFormat": "<8TB",
595 "refId": "F",
596 "step": 2
597 },
598 {
599 "expr": "count(ceph_osd_stat_bytes >= 8796093022208 < 10995116277760)",
600 "format": "time_series",
601 "intervalFactor": 2,
602 "legendFormat": "<10TB",
603 "refId": "G",
604 "step": 2
605 },
606 {
607 "expr": "count(ceph_osd_stat_bytes >= 10995116277760 < 13194139533312)",
608 "format": "time_series",
609 "intervalFactor": 2,
610 "legendFormat": "<12TB",
611 "refId": "H",
612 "step": 2
613 },
614 {
615 "expr": "count(ceph_osd_stat_bytes >= 13194139533312)",
616 "format": "time_series",
617 "intervalFactor": 2,
618 "legendFormat": "12TB+",
619 "refId": "I",
620 "step": 2
621 }
622 ],
92f5a8d4 623 "timeFrom": null,
11fdf7f2
TL
624 "timeShift": null,
625 "title": "OSD Size Summary",
626 "type": "grafana-piechart-panel",
627 "valueName": "current"
628 },
629 {
630 "aliasColors": {},
631 "bars": true,
632 "dashLength": 10,
633 "dashes": false,
634 "datasource": "$datasource",
635 "description": "Each bar indicates the number of OSD's that have a PG count in a specific range as shown on the x axis.",
636 "fill": 1,
637 "gridPos": {
638 "h": 8,
639 "w": 12,
640 "x": 12,
641 "y": 8
642 },
643 "id": 6,
644 "legend": {
645 "alignAsTable": false,
646 "avg": false,
647 "current": false,
648 "hideEmpty": false,
649 "hideZero": false,
650 "max": false,
651 "min": false,
652 "rightSide": false,
653 "show": false,
654 "total": false,
655 "values": false
656 },
657 "lines": false,
658 "linewidth": 1,
659 "links": [],
660 "nullPointMode": "null",
661 "percentage": false,
662 "pointradius": 5,
663 "points": false,
664 "renderer": "flot",
665 "seriesOverrides": [],
666 "spaceLength": 10,
667 "stack": false,
668 "steppedLine": false,
669 "targets": [
670 {
671 "expr": "ceph_osd_numpg\n",
672 "format": "time_series",
673 "instant": true,
674 "intervalFactor": 1,
675 "legendFormat": "PGs per OSD",
676 "refId": "A"
677 }
678 ],
679 "thresholds": [],
680 "timeFrom": null,
681 "timeShift": null,
682 "title": "Distribution of PGs per OSD",
683 "tooltip": {
684 "shared": false,
685 "sort": 0,
686 "value_type": "individual"
687 },
688 "type": "graph",
689 "xaxis": {
690 "buckets": 20,
691 "mode": "histogram",
692 "name": null,
693 "show": true,
694 "values": [
695 "total"
696 ]
697 },
698 "yaxes": [
699 {
700 "decimals": 0,
701 "format": "short",
702 "label": "# of OSDs",
703 "logBase": 1,
704 "max": null,
705 "min": "0",
706 "show": true
707 },
708 {
709 "format": "short",
710 "label": null,
711 "logBase": 1,
712 "max": null,
713 "min": null,
714 "show": false
715 }
716 ]
717 },
718 {
719 "collapsed": false,
720 "gridPos": {
721 "h": 1,
722 "w": 24,
723 "x": 0,
724 "y": 16
725 },
726 "id": 20,
727 "panels": [],
728 "title": "R/W Profile",
729 "type": "row"
730 },
731 {
732 "aliasColors": {},
733 "bars": false,
734 "dashLength": 10,
735 "dashes": false,
736 "datasource": "$datasource",
737 "description": "Show the read/write workload profile overtime",
738 "fill": 1,
739 "gridPos": {
740 "h": 8,
741 "w": 24,
742 "x": 0,
743 "y": 17
744 },
745 "id": 10,
746 "legend": {
747 "avg": false,
748 "current": false,
749 "max": false,
750 "min": false,
751 "show": true,
752 "total": false,
753 "values": false
754 },
755 "lines": true,
756 "linewidth": 1,
757 "links": [],
758 "nullPointMode": "null",
759 "percentage": false,
760 "pointradius": 5,
761 "points": false,
762 "renderer": "flot",
763 "seriesOverrides": [],
764 "spaceLength": 10,
765 "stack": true,
766 "steppedLine": false,
767 "targets": [
768 {
769 "expr": "round(sum(irate(ceph_pool_rd[30s])))",
770 "format": "time_series",
771 "intervalFactor": 1,
772 "legendFormat": "Reads",
773 "refId": "A"
774 },
775 {
776 "expr": "round(sum(irate(ceph_pool_wr[30s])))",
777 "format": "time_series",
778 "intervalFactor": 1,
779 "legendFormat": "Writes",
780 "refId": "B"
781 }
782 ],
783 "thresholds": [],
92f5a8d4 784 "timeFrom": null,
11fdf7f2
TL
785 "timeShift": null,
786 "title": "Read/Write Profile",
787 "tooltip": {
788 "shared": true,
eafe8130 789 "sort": 2,
11fdf7f2
TL
790 "value_type": "individual"
791 },
792 "type": "graph",
793 "xaxis": {
794 "buckets": null,
795 "mode": "time",
796 "name": null,
797 "show": true,
798 "values": []
799 },
800 "yaxes": [
801 {
802 "format": "short",
803 "label": null,
804 "logBase": 1,
805 "max": null,
806 "min": null,
807 "show": true
808 },
809 {
810 "format": "short",
811 "label": null,
812 "logBase": 1,
813 "max": null,
814 "min": null,
815 "show": true
816 }
817 ]
818 }
819 ],
820 "refresh": "10s",
821 "schemaVersion": 16,
822 "style": "dark",
823 "tags": [],
824 "templating": {
825 "list": [
826 {
827 "current": {
828 "tags": [],
829 "text": "default",
830 "value": "default"
831 },
832 "hide": 0,
833 "label": "Data Source",
834 "name": "datasource",
835 "options": [],
836 "query": "prometheus",
837 "refresh": 1,
838 "regex": "",
839 "type": "datasource"
840 }
841 ]
842 },
843 "time": {
844 "from": "now-1h",
845 "to": "now"
846 },
847 "timepicker": {
848 "refresh_intervals": [
849 "5s",
850 "10s",
851 "30s",
852 "1m",
853 "5m",
854 "15m",
855 "30m",
856 "1h",
857 "2h",
858 "1d"
859 ],
860 "time_options": [
861 "5m",
862 "15m",
863 "1h",
864 "6h",
865 "12h",
866 "24h",
867 "2d",
868 "7d",
869 "30d"
870 ]
871 },
872 "timezone": "",
873 "title": "OSD Overview",
874 "uid": "lo02I1Aiz",
875 "version": 3
876}