Skip to content

Commit 625300d

Browse files
ib-steffenib-steffen
authored
Merge pull request #69 from sap-coutantm/master
Monitoring : Fixes and Node Monitoring
2 parents a559179 + f1d3bfd commit 625300d

12 files changed

Lines changed: 2467 additions & 167 deletions

File tree

‎deploy/infrabox/grafana/dashboards/build_inspector.json‎

Lines changed: 12 additions & 29 deletions
Original file line numberDiff line numberDiff line change
@@ -190,40 +190,23 @@
190190
"templating": {
191191
"list": [
192192
{
193-
"allValue": null,
194-
"current": {
195-
"text": "0006fa7e-b193-4dd2-a746-6c6be16f7d8c",
196-
"value": "0006fa7e-b193-4dd2-a746-6c6be16f7d8c"
197-
},
193+
"allValue": null,
194+
"datasource": "infrabox",
198195
"hide": 2,
199196
"includeAll": false,
200197
"label": "",
201198
"multi": false,
202199
"name": "bid",
203-
"options": [
204-
{
205-
"selected": true,
206-
"text": "0006fa7e-b193-4dd2-a746-6c6be16f7d8c",
207-
"value": "0006fa7e-b193-4dd2-a746-6c6be16f7d8c"
208-
},
209-
{
210-
"selected": false,
211-
"text": "0009ad79-6814-44c4-b33b-66326af7e849",
212-
"value": "0009ad79-6814-44c4-b33b-66326af7e849"
213-
},
214-
{
215-
"selected": false,
216-
"text": "00225b05-6417-49d6-846a-6aab6a105695",
217-
"value": "00225b05-6417-49d6-846a-6aab6a105695"
218-
},
219-
{
220-
"selected": false,
221-
"text": "005d6f95-68d7-41d3-88a9-8c323eb66f4f",
222-
"value": "005d6f95-68d7-41d3-88a9-8c323eb66f4f"
223-
}
224-
],
225-
"query": "0006fa7e-b193-4dd2-a746-6c6be16f7d8c,0009ad79-6814-44c4-b33b-66326af7e849,00225b05-6417-49d6-846a-6aab6a105695,005d6f95-68d7-41d3-88a9-8c323eb66f4f",
226-
"type": "custom"
200+
"options": [],
201+
"query": "SELECT id FROM build LIMIT 1",
202+
"refresh": 1,
203+
"regex": "",
204+
"sort": 0,
205+
"tagValuesQuery": "",
206+
"tags": [],
207+
"tagsQuery": "",
208+
"type": "query",
209+
"useTags": false
227210
}
228211
]
229212
},

‎deploy/infrabox/grafana/dashboards/cluster_inspector.json‎

Lines changed: 162 additions & 107 deletions
Original file line numberDiff line numberDiff line change
@@ -17,8 +17,8 @@
1717
"editable": true,
1818
"gnetId": null,
1919
"graphTooltip": 0,
20-
"id": 3,
21-
"iteration": 1529658438657,
20+
"id": 2,
21+
"iteration": 1531746409662,
2222
"links": [],
2323
"panels": [
2424
{
@@ -32,6 +32,102 @@
3232
"title": "Basic Information",
3333
"type": "row"
3434
},
35+
{
36+
"aliasColors": {},
37+
"bars": false,
38+
"dashLength": 10,
39+
"dashes": false,
40+
"datasource": "prometheus",
41+
"decimals": null,
42+
"description": "Amount of jobs of different states over time among the inspected cluster.",
43+
"fill": 1,
44+
"gridPos": {
45+
"h": 7,
46+
"w": 16,
47+
"x": 0,
48+
"y": 1
49+
},
50+
"id": 23,
51+
"interval": "5s",
52+
"legend": {
53+
"alignAsTable": true,
54+
"avg": true,
55+
"current": true,
56+
"max": true,
57+
"min": true,
58+
"rightSide": true,
59+
"show": true,
60+
"total": false,
61+
"values": true
62+
},
63+
"lines": true,
64+
"linewidth": 1,
65+
"links": [],
66+
"nullPointMode": "null",
67+
"percentage": false,
68+
"pointradius": 5,
69+
"points": false,
70+
"renderer": "flot",
71+
"seriesOverrides": [],
72+
"spaceLength": 10,
73+
"stack": false,
74+
"steppedLine": false,
75+
"targets": [
76+
{
77+
"$$hashKey": "object:3933",
78+
"expr": "job_active_current_count{cluster='[[cluster]]'}",
79+
"format": "time_series",
80+
"hide": false,
81+
"instant": false,
82+
"interval": "5s",
83+
"intervalFactor": 1,
84+
"legendFormat": "{{state}}",
85+
"refId": "B"
86+
}
87+
],
88+
"thresholds": [],
89+
"timeFrom": null,
90+
"timeShift": null,
91+
"title": "Job Count",
92+
"tooltip": {
93+
"shared": true,
94+
"sort": 0,
95+
"value_type": "individual"
96+
},
97+
"type": "graph",
98+
"xaxis": {
99+
"buckets": null,
100+
"mode": "time",
101+
"name": null,
102+
"show": true,
103+
"values": []
104+
},
105+
"yaxes": [
106+
{
107+
"$$hashKey": "object:3961",
108+
"decimals": 0,
109+
"format": "short",
110+
"label": "Jobs",
111+
"logBase": 1,
112+
"max": null,
113+
"min": null,
114+
"show": true
115+
},
116+
{
117+
"$$hashKey": "object:3962",
118+
"format": "short",
119+
"label": null,
120+
"logBase": 1,
121+
"max": null,
122+
"min": null,
123+
"show": true
124+
}
125+
],
126+
"yaxis": {
127+
"align": false,
128+
"alignLevel": null
129+
}
130+
},
35131
{
36132
"cacheTimeout": null,
37133
"colorBackground": false,
@@ -53,8 +149,8 @@
53149
},
54150
"gridPos": {
55151
"h": 7,
56-
"w": 3,
57-
"x": 0,
152+
"w": 2,
153+
"x": 16,
58154
"y": 1
59155
},
60156
"id": 6,
@@ -116,100 +212,77 @@
116212
"valueName": "avg"
117213
},
118214
{
119-
"aliasColors": {},
120-
"bars": false,
121-
"dashLength": 10,
122-
"dashes": false,
123-
"datasource": "prometheus",
124-
"decimals": null,
125-
"description": "Amount of jobs of different states over time among the inspected cluster.",
126-
"fill": 1,
215+
"columns": [],
216+
"datasource": "infrabox",
217+
"description": "List of the node_name referenced by the job table for this cluster.\n\nIf the Node has never been associated to any job, it can't be seen.",
218+
"fontSize": "100%",
127219
"gridPos": {
128220
"h": 7,
129-
"w": 21,
130-
"x": 3,
221+
"w": 6,
222+
"x": 18,
131223
"y": 1
132224
},
133-
"id": 23,
134-
"interval": "5s",
135-
"legend": {
136-
"alignAsTable": true,
137-
"avg": true,
138-
"current": true,
139-
"max": true,
140-
"min": true,
141-
"rightSide": true,
142-
"show": true,
143-
"total": false,
144-
"values": true
145-
},
146-
"lines": true,
147-
"linewidth": 1,
225+
"id": 25,
148226
"links": [],
149-
"nullPointMode": "null",
150-
"percentage": false,
151-
"pointradius": 5,
152-
"points": false,
153-
"renderer": "flot",
154-
"seriesOverrides": [],
155-
"spaceLength": 10,
156-
"stack": false,
157-
"steppedLine": false,
158-
"targets": [
159-
{
160-
"$$hashKey": "object:3933",
161-
"expr": "job_current_count{cluster='[[cluster]]'}",
162-
"format": "time_series",
163-
"hide": false,
164-
"instant": false,
165-
"interval": "5s",
166-
"intervalFactor": 1,
167-
"legendFormat": "{{state}}",
168-
"refId": "B"
169-
}
170-
],
171-
"thresholds": [],
172-
"timeFrom": null,
173-
"timeShift": null,
174-
"title": "Job Count",
175-
"tooltip": {
176-
"shared": true,
177-
"sort": 0,
178-
"value_type": "individual"
179-
},
180-
"type": "graph",
181-
"xaxis": {
182-
"buckets": null,
183-
"mode": "time",
184-
"name": null,
185-
"show": true,
186-
"values": []
227+
"pageSize": null,
228+
"scroll": true,
229+
"showHeader": true,
230+
"sort": {
231+
"col": 0,
232+
"desc": true
187233
},
188-
"yaxes": [
234+
"styles": [
235+
{
236+
"$$hashKey": "object:535",
237+
"alias": "Node",
238+
"colorMode": null,
239+
"colors": [
240+
"rgba(245, 54, 54, 0.9)",
241+
"rgba(237, 129, 40, 0.89)",
242+
"rgba(50, 172, 45, 0.97)"
243+
],
244+
"dateFormat": "YYYY-MM-DD HH:mm:ss",
245+
"decimals": 2,
246+
"link": true,
247+
"linkTooltip": "Inspect",
248+
"linkUrl": "d/09c0DMdiz/node-inspector?orgId=1&var-node_name=${__cell}",
249+
"mappingType": 1,
250+
"pattern": "node_name",
251+
"thresholds": [],
252+
"type": "string",
253+
"unit": "short"
254+
},
189255
{
190-
"$$hashKey": "object:3961",
256+
"$$hashKey": "object:565",
257+
"alias": "Running Jobs",
258+
"colorMode": null,
259+
"colors": [
260+
"rgba(245, 54, 54, 0.9)",
261+
"rgba(237, 129, 40, 0.89)",
262+
"rgba(50, 172, 45, 0.97)"
263+
],
264+
"dateFormat": "YYYY-MM-DD HH:mm:ss",
191265
"decimals": 0,
192-
"format": "short",
193-
"label": "Jobs",
194-
"logBase": 1,
195-
"max": null,
196-
"min": null,
197-
"show": true
198-
},
266+
"mappingType": 1,
267+
"pattern": "count",
268+
"thresholds": [],
269+
"type": "number",
270+
"unit": "short"
271+
}
272+
],
273+
"targets": [
199274
{
200-
"$$hashKey": "object:3962",
201-
"format": "short",
202-
"label": null,
203-
"logBase": 1,
204-
"max": null,
205-
"min": null,
206-
"show": true
275+
"$$hashKey": "object:282",
276+
"alias": "",
277+
"format": "table",
278+
"hide": false,
279+
"rawSql": "SELECT node_name, count(id) FILTER(WHERE state = 'running') FROM job WHERE node_name is not null AND cluster_name = '[[cluster]]' GROUP BY node_name",
280+
"refId": "A"
207281
}
208282
],
209-
"yaxis": {
210-
"align": false,
211-
"alignLevel": null
212-
}
283+
"title": "Node List",
284+
"transform": "table",
285+
"type": "table"
213286
},
214287
{
215288
"collapsed": false,
@@ -753,33 +826,15 @@
753826
"list": [
754827
{
755828
"allValue": null,
756-
"current": {
757-
"tags": [],
758-
"text": "calico",
759-
"value": "calico"
760-
},
761829
"datasource": "infrabox",
762830
"hide": 0,
763831
"includeAll": false,
764832
"label": "Inspected Cluster",
765833
"multi": false,
766834
"name": "cluster",
767-
"options": [
768-
{
769-
"$$hashKey": "object:3892",
770-
"selected": true,
771-
"text": "calico",
772-
"value": "calico"
773-
},
774-
{
775-
"$$hashKey": "object:3893",
776-
"selected": false,
777-
"text": "master",
778-
"value": "master"
779-
}
780-
],
835+
"options": [],
781836
"query": "select distinct name from cluster",
782-
"refresh": 0,
837+
"refresh": 1,
783838
"regex": "",
784839
"sort": 0,
785840
"tagValuesQuery": "",

‎deploy/infrabox/grafana/dashboards/global_dashboard.json‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -16,7 +16,7 @@
1616
"editable": true,
1717
"gnetId": null,
1818
"graphTooltip": 0,
19-
"id": 1,
19+
"id": 2,
2020
"links": [],
2121
"panels": [
2222
{
@@ -556,7 +556,7 @@
556556
"targets": [
557557
{
558558
"$$hashKey": "object:1147",
559-
"expr": "sum(job_current_count) by(state)",
559+
"expr": "job_active_current_count{cluster=\"'%'\"}",
560560
"format": "time_series",
561561
"hide": false,
562562
"interval": "",

0 commit comments

Comments
 (0)