Skip to content

Commit cff66dc

Browse files
s0neas0nea
authored
Ceph dashboard performance optimizations (#47)
* ceph-capacity-quick-view: optimize queries Remove unnecessary multiplier to optimize queries. Signed-off-by: Tatjana Dehler <tatjana.dehler@clyso.com> * ceph-rgw-overview-additions: reduce amount of data Reduce amount of data shown on some panels to prevent performance issues. Signed-off-by: Tatjana Dehler <tatjana.dehler@clyso.com> * ceph-rgw-overview: reduce amout of data Reduce amount of data shown on some panels to prevent performance issues. Signed-off-by: Tatjana Dehler <tatjana.dehler@clyso.com> --------- Signed-off-by: Tatjana Dehler <tatjana.dehler@clyso.com>
1 parent f7cb0b0 commit cff66dc

5 files changed

Lines changed: 396 additions & 604 deletions

File tree

‎charts/ceph-operations/Chart.yaml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@ apiVersion: v2
22
name: ceph-operations
33
description: Ceph operations bundle
44
type: application
5-
version: 1.6.8
5+
version: 1.6.9
66
maintainers:
77
- name: sumitarora2786
88
- name: richardtief

‎charts/ceph-operations/perses-dashboards-global/ceph-capacity-quick-view.json‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -371,7 +371,7 @@
371371
"kind": "PrometheusTimeSeriesQuery",
372372
"spec": {
373373
"minStep": "",
374-
"query": "max(global:limes_consolidated_cluster_capacity{full_resource=~\"ceph/(.*)\"}) by (region, full_resource) *\nmax(global:limes_consolidated_unit_multiplier{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)",
374+
"query": "max(global:limes_consolidated_cluster_capacity{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)",
375375
"seriesNameFormat": ""
376376
}
377377
}
@@ -387,7 +387,7 @@
387387
"kind": "PrometheusDatasource"
388388
},
389389
"minStep": "",
390-
"query": "sum(global:limes_consolidated_domain_quota{full_resource=~\"ceph/(.*)\"}) by (region, full_resource) *\nmax(global:limes_consolidated_unit_multiplier{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)",
390+
"query": "sum(global:limes_consolidated_domain_quota{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)",
391391
"seriesNameFormat": ""
392392
}
393393
}
@@ -403,7 +403,7 @@
403403
"kind": "PrometheusDatasource"
404404
},
405405
"minStep": "",
406-
"query": "(\n max(global:limes_consolidated_cluster_capacity{full_resource=~\"ceph/(.*)\"}) by (region, full_resource) *\n max(global:limes_consolidated_unit_multiplier{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n) - (\n sum(global:limes_consolidated_domain_quota{full_resource=~\"ceph/(.*)\"}) by (region, full_resource) *\n max(global:limes_consolidated_unit_multiplier{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n)",
406+
"query": "(\n max(global:limes_consolidated_cluster_capacity{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n) - (\n sum(global:limes_consolidated_domain_quota{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n)",
407407
"seriesNameFormat": ""
408408
}
409409
}
@@ -419,7 +419,7 @@
419419
"kind": "PrometheusDatasource"
420420
},
421421
"minStep": "",
422-
"query": "sum(global:limes_consolidated_domain_usage{full_resource=~\"ceph/(.*)\"}) by (region, full_resource) *\nmax(global:limes_consolidated_unit_multiplier{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)",
422+
"query": "sum(global:limes_consolidated_domain_usage{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)",
423423
"seriesNameFormat": ""
424424
}
425425
}
@@ -435,7 +435,7 @@
435435
"kind": "PrometheusDatasource"
436436
},
437437
"minStep": "",
438-
"query": "(\n sum(global:limes_consolidated_domain_quota{full_resource=~\"ceph/(.*)\"}) by (region, full_resource) *\n max(global:limes_consolidated_unit_multiplier{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n) / (\n max(global:limes_consolidated_cluster_capacity{full_resource=~\"ceph/(.*)\"}) by (region, full_resource) *\n max(global:limes_consolidated_unit_multiplier{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n)",
438+
"query": "(\n sum(global:limes_consolidated_domain_quota{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n) / (\n max(global:limes_consolidated_cluster_capacity{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n)",
439439
"seriesNameFormat": ""
440440
}
441441
}
@@ -451,7 +451,7 @@
451451
"kind": "PrometheusDatasource"
452452
},
453453
"minStep": "",
454-
"query": "(\n sum(global:limes_consolidated_domain_usage{full_resource=~\"ceph/(.*)\"}) by (region, full_resource) *\n max(global:limes_consolidated_unit_multiplier{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n) / (\n max(global:limes_consolidated_cluster_capacity{full_resource=~\"ceph/(.*)\"}) by (region, full_resource) *\n max(global:limes_consolidated_unit_multiplier{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n)",
454+
"query": "(\n sum(global:limes_consolidated_domain_usage{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n) / (\n max(global:limes_consolidated_cluster_capacity{full_resource=~\"ceph/(.*)\"}) by (region, full_resource)\n)",
455455
"seriesNameFormat": ""
456456
}
457457
}

‎charts/ceph-operations/perses-dashboards/ceph-rgw-overview-additions.json‎

Lines changed: 38 additions & 66 deletions
Original file line numberDiff line numberDiff line change
@@ -19,17 +19,10 @@
1919
"kind": "TimeSeriesChart",
2020
"spec": {
2121
"legend": {
22-
"mode": "table",
23-
"position": "bottom",
24-
"size": "small",
25-
"values": [
26-
"mean",
27-
"last",
28-
"max"
29-
]
22+
"position": "bottom"
3023
},
3124
"visual": {
32-
"areaOpacity": 0.1,
25+
"areaOpacity": 0,
3326
"connectNulls": false,
3427
"lineWidth": 1.25,
3528
"pointRadius": 2.75
@@ -63,23 +56,16 @@
6356
"kind": "Panel",
6457
"spec": {
6558
"display": {
66-
"name": "4er Errors By Bucket"
59+
"name": "Top 10 - 4er Errors By Bucket"
6760
},
6861
"plugin": {
6962
"kind": "TimeSeriesChart",
7063
"spec": {
7164
"legend": {
72-
"mode": "table",
73-
"position": "bottom",
74-
"size": "small",
75-
"values": [
76-
"mean",
77-
"last",
78-
"max"
79-
]
65+
"position": "bottom"
8066
},
8167
"visual": {
82-
"areaOpacity": 0.1,
68+
"areaOpacity": 0,
8369
"connectNulls": false,
8470
"lineWidth": 1.25,
8571
"pointRadius": 2.75
@@ -100,7 +86,7 @@
10086
"plugin": {
10187
"kind": "PrometheusTimeSeriesQuery",
10288
"spec": {
103-
"query": "sum by (bucket, http_status) (\n rate(radosgw_errors_detailed{http_status=~\"4[0-9][0-9]\", http_status=~\"$errors\", bucket=~\"$bucket\"}[$__rate_interval])\n)",
89+
"query": "topk(10,\n sum by (bucket, http_status) (\n rate(radosgw_errors_detailed{http_status=~\"4[0-9][0-9]\", http_status=~\"$errors\", bucket=~\"$bucket\"}[$__rate_interval])\n )\n)",
10490
"seriesNameFormat": "{{bucket}} - {{http_status}}"
10591
}
10692
}
@@ -119,17 +105,10 @@
119105
"kind": "TimeSeriesChart",
120106
"spec": {
121107
"legend": {
122-
"mode": "table",
123-
"position": "bottom",
124-
"size": "small",
125-
"values": [
126-
"mean",
127-
"last",
128-
"max"
129-
]
108+
"position": "bottom"
130109
},
131110
"visual": {
132-
"areaOpacity": 0.1,
111+
"areaOpacity": 0,
133112
"connectNulls": false,
134113
"lineWidth": 1.25,
135114
"pointRadius": 2.75
@@ -169,17 +148,10 @@
169148
"kind": "TimeSeriesChart",
170149
"spec": {
171150
"legend": {
172-
"mode": "table",
173-
"position": "bottom",
174-
"size": "small",
175-
"values": [
176-
"mean",
177-
"last",
178-
"max"
179-
]
151+
"position": "bottom"
180152
},
181153
"visual": {
182-
"areaOpacity": 0.1,
154+
"areaOpacity": 0,
183155
"connectNulls": false,
184156
"lineWidth": 1.25,
185157
"pointRadius": 2.75
@@ -213,7 +185,7 @@
213185
"kind": "Panel",
214186
"spec": {
215187
"display": {
216-
"name": "Bucket Requests DELETE"
188+
"name": "Top 10 - Bucket Requests DELETE"
217189
},
218190
"plugin": {
219191
"kind": "TimeSeriesChart",
@@ -225,7 +197,7 @@
225197
"values": []
226198
},
227199
"visual": {
228-
"areaOpacity": 0.1,
200+
"areaOpacity": 0,
229201
"connectNulls": false,
230202
"lineWidth": 1.25,
231203
"pointRadius": 2.75
@@ -246,7 +218,7 @@
246218
"plugin": {
247219
"kind": "PrometheusTimeSeriesQuery",
248220
"spec": {
249-
"query": "sum by (ip) (\n rate(radosgw_requests_by_ip_bucket_method_tenant{ip=~\"$ip\", method=\"DELETE\"}[$__rate_interval])\n)",
221+
"query": "topk(10,\n sum by (ip) (\n rate(radosgw_requests_by_ip_bucket_method_tenant{ip=~\"$ip\", method=\"DELETE\"}[$__rate_interval])\n )\n)",
250222
"seriesNameFormat": "{{ip}}"
251223
}
252224
}
@@ -259,7 +231,7 @@
259231
"kind": "Panel",
260232
"spec": {
261233
"display": {
262-
"name": "Bucket Requests GET"
234+
"name": "Top 10 - Bucket Requests GET"
263235
},
264236
"plugin": {
265237
"kind": "TimeSeriesChart",
@@ -271,7 +243,7 @@
271243
"values": []
272244
},
273245
"visual": {
274-
"areaOpacity": 0.1,
246+
"areaOpacity": 0,
275247
"connectNulls": false,
276248
"lineWidth": 1.25,
277249
"pointRadius": 2.75
@@ -292,7 +264,7 @@
292264
"plugin": {
293265
"kind": "PrometheusTimeSeriesQuery",
294266
"spec": {
295-
"query": "sum by (ip) (\n rate(radosgw_requests_by_ip_bucket_method_tenant{ip=~\"$ip\", method=\"GET\"}[$__rate_interval])\n)",
267+
"query": "topk(10,\n sum by (ip) (\n rate(radosgw_requests_by_ip_bucket_method_tenant{ip=~\"$ip\", method=\"GET\"}[$__rate_interval])\n )\n)",
296268
"seriesNameFormat": "{{ip}}"
297269
}
298270
}
@@ -305,7 +277,7 @@
305277
"kind": "Panel",
306278
"spec": {
307279
"display": {
308-
"name": "Bucket Requests PUT"
280+
"name": "Top 10 - Bucket Requests PUT"
309281
},
310282
"plugin": {
311283
"kind": "TimeSeriesChart",
@@ -317,7 +289,7 @@
317289
"values": []
318290
},
319291
"visual": {
320-
"areaOpacity": 0.1,
292+
"areaOpacity": 0,
321293
"connectNulls": false,
322294
"lineWidth": 1.25,
323295
"pointRadius": 2.75
@@ -338,7 +310,7 @@
338310
"plugin": {
339311
"kind": "PrometheusTimeSeriesQuery",
340312
"spec": {
341-
"query": "sum by (ip) (\n rate(radosgw_requests_by_ip_bucket_method_tenant{ip=~\"$ip\", method=\"PUT\"}[$__rate_interval])\n)",
313+
"query": "topk(10,\n sum by (ip) (\n rate(radosgw_requests_by_ip_bucket_method_tenant{ip=~\"$ip\", method=\"PUT\"}[$__rate_interval])\n )\n)",
342314
"seriesNameFormat": "{{ip}}"
343315
}
344316
}
@@ -351,7 +323,7 @@
351323
"kind": "Panel",
352324
"spec": {
353325
"display": {
354-
"name": "Bucket Requests per Client IP"
326+
"name": "Top 10 - Bucket Requests per Client IP"
355327
},
356328
"plugin": {
357329
"kind": "TimeSeriesChart",
@@ -363,7 +335,7 @@
363335
"values": []
364336
},
365337
"visual": {
366-
"areaOpacity": 0.1,
338+
"areaOpacity": 0,
367339
"connectNulls": false,
368340
"lineWidth": 1.25,
369341
"pointRadius": 2.75
@@ -384,7 +356,7 @@
384356
"plugin": {
385357
"kind": "PrometheusTimeSeriesQuery",
386358
"spec": {
387-
"query": "sum by (ip, bucket, method) (\n rate(radosgw_requests_by_ip_bucket_method_tenant{ip=~\"$ip\", bucket=~\"$bucket\"}[$__rate_interval])\n)",
359+
"query": "topk(10,\n sum by (ip, bucket, method) (\n rate(radosgw_requests_by_ip_bucket_method_tenant{ip=~\"$ip\", bucket=~\"$bucket\"}[$__rate_interval])\n )\n)",
388360
"seriesNameFormat": "{{ip}} - {{bucket}}"
389361
}
390362
}
@@ -409,7 +381,7 @@
409381
"values": []
410382
},
411383
"visual": {
412-
"areaOpacity": 0.1,
384+
"areaOpacity": 0,
413385
"connectNulls": false,
414386
"lineWidth": 1.25,
415387
"pointRadius": 2.75
@@ -455,7 +427,7 @@
455427
"values": []
456428
},
457429
"visual": {
458-
"areaOpacity": 0.1,
430+
"areaOpacity": 0,
459431
"connectNulls": false,
460432
"lineWidth": 1.25,
461433
"pointRadius": 2.75
@@ -498,7 +470,7 @@
498470
"position": "bottom"
499471
},
500472
"visual": {
501-
"areaOpacity": 0.1,
473+
"areaOpacity": 0,
502474
"connectNulls": false,
503475
"lineWidth": 1.25,
504476
"pointRadius": 2.75
@@ -532,7 +504,7 @@
532504
"kind": "Panel",
533505
"spec": {
534506
"display": {
535-
"name": "Requests per User and IP"
507+
"name": "Top 10 - Requests per User and IP"
536508
},
537509
"plugin": {
538510
"kind": "TimeSeriesChart",
@@ -541,7 +513,7 @@
541513
"position": "right"
542514
},
543515
"visual": {
544-
"areaOpacity": 0.1,
516+
"areaOpacity": 0,
545517
"connectNulls": false,
546518
"lineWidth": 1.25,
547519
"pointRadius": 2.75
@@ -562,7 +534,7 @@
562534
"plugin": {
563535
"kind": "PrometheusTimeSeriesQuery",
564536
"spec": {
565-
"query": "sum by (user, ip) (\n rate(radosgw_requests_by_ip{ip=~\"$ip\", user=~\"$user\"}[$__rate_interval])\n)",
537+
"query": "topk(10,\n sum by (user, ip) (\n rate(radosgw_requests_by_ip{ip=~\"$ip\", user=~\"$user\"}[$__rate_interval])\n )\n)",
566538
"seriesNameFormat": "{{ip}} - {{user}}"
567539
}
568540
}
@@ -586,34 +558,34 @@
586558
{
587559
"x": 0,
588560
"y": 0,
589-
"width": 12,
590-
"height": 10,
561+
"width": 9,
562+
"height": 9,
591563
"content": {
592564
"$ref": "#/spec/panels/4erErrors"
593565
}
594566
},
595567
{
596-
"x": 12,
597-
"y": 10,
598-
"width": 12,
568+
"x": 9,
569+
"y": 9,
570+
"width": 8,
599571
"height": 10,
600572
"content": {
601573
"$ref": "#/spec/panels/4erErrors-3"
602574
}
603575
},
604576
{
605-
"x": 12,
577+
"x": 9,
606578
"y": 0,
607-
"width": 12,
608-
"height": 10,
579+
"width": 8,
580+
"height": 9,
609581
"content": {
610582
"$ref": "#/spec/panels/4erErrors-4"
611583
}
612584
},
613585
{
614586
"x": 0,
615-
"y": 10,
616-
"width": 12,
587+
"y": 9,
588+
"width": 9,
617589
"height": 10,
618590
"content": {
619591
"$ref": "#/spec/panels/4erErrors-2"

0 commit comments

Comments
 (0)