Skip to content

Commit b71a76c

Browse files
Ji-minhyeokJi-minhyeok
authored
refactor: 모니터링 대시보드 통합 및 비즈니스 핵심 지표 고도화 (#217)
* refactor: 모니터링 대시보드 통합 및 비즈니스 핵심 지표 고도화
1 parent 7f6dae6 commit b71a76c

1 file changed

Lines changed: 204 additions & 0 deletions

File tree

‎prometheus-grafana/grafana/dashboards/trainus-dashboard.json‎

Lines changed: 204 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,204 @@
1+
{
2+
"annotations": {
3+
"list": [
4+
{
5+
"builtIn": 1,
6+
"datasource": { "type": "grafana", "uid": "-- Grafana --" },
7+
"enable": true,
8+
"hide": true,
9+
"iconColor": "rgba(0, 211, 255, 1)",
10+
"name": "Annotations & Alerts",
11+
"type": "dashboard"
12+
}
13+
]
14+
},
15+
"editable": true,
16+
"fiscalYearStartMonth": 0,
17+
"graphTooltip": 1,
18+
"id": null,
19+
"links": [],
20+
"panels": [
21+
{
22+
"collapsed": false,
23+
"gridPos": { "h": 1, "w": 24, "x": 0, "y": 0 },
24+
"id": 100,
25+
"panels": [],
26+
"title": "System Resources (Threads & Connections) [$env]",
27+
"type": "row"
28+
},
29+
{
30+
"datasource": { "type": "prometheus", "uid": "prometheus-default" },
31+
"description": "API 서버의 Tomcat 스레드 상태 (Busy, Current, Max)",
32+
"fieldConfig": {
33+
"defaults": {
34+
"color": { "mode": "palette-classic" },
35+
"custom": { "drawStyle": "line", "fillOpacity": 10, "lineWidth": 2 },
36+
"unit": "short"
37+
}
38+
},
39+
"gridPos": { "h": 8, "w": 12, "x": 0, "y": 1 },
40+
"id": 1,
41+
"options": {
42+
"legend": { "calcs": ["lastNotNull", "max"], "displayMode": "table", "placement": "bottom" },
43+
"tooltip": { "mode": "multi", "sort": "none" }
44+
},
45+
"targets": [
46+
{ "expr": "tomcat_threads_busy_threads{env=\"$env\", role=\"api\"}", "legendFormat": "Busy: {{instance}}", "refId": "A" },
47+
{ "expr": "tomcat_threads_current_threads{env=\"$env\", role=\"api\"}", "legendFormat": "Current: {{instance}}", "refId": "B" },
48+
{ "expr": "tomcat_threads_config_max_threads{env=\"$env\", role=\"api\"}", "legendFormat": "Max: {{instance}}", "refId": "C" }
49+
],
50+
"title": "API Tomcat Threads",
51+
"type": "timeseries"
52+
},
53+
{
54+
"datasource": { "type": "prometheus", "uid": "prometheus-default" },
55+
"description": "API 서버의 HikariCP 커넥션 풀 상태 (Active, Idle, Pending)",
56+
"fieldConfig": {
57+
"defaults": {
58+
"color": { "mode": "palette-classic" },
59+
"custom": { "drawStyle": "line", "fillOpacity": 15, "lineWidth": 2 },
60+
"unit": "short"
61+
}
62+
},
63+
"gridPos": { "h": 8, "w": 12, "x": 12, "y": 1 },
64+
"id": 3,
65+
"options": {
66+
"legend": { "calcs": ["lastNotNull", "max"], "displayMode": "table", "placement": "bottom" },
67+
"tooltip": { "mode": "multi", "sort": "none" }
68+
},
69+
"targets": [
70+
{ "expr": "hikaricp_connections_active{env=\"$env\", role=\"api\"}", "legendFormat": "Active: {{instance}}", "refId": "A" },
71+
{ "expr": "hikaricp_connections_idle{env=\"$env\", role=\"api\"}", "legendFormat": "Idle: {{instance}}", "refId": "B" },
72+
{ "expr": "hikaricp_connections_pending{env=\"$env\", role=\"api\"}", "legendFormat": "Pending: {{instance}}", "refId": "C" }
73+
],
74+
"title": "API HikariCP Connections",
75+
"type": "timeseries"
76+
},
77+
{
78+
"collapsed": false,
79+
"gridPos": { "h": 1, "w": 24, "x": 0, "y": 9 },
80+
"id": 101,
81+
"panels": [],
82+
"title": "Core Business Metrics (MQ & Defense) [$env]",
83+
"type": "row"
84+
},
85+
{
86+
"datasource": { "type": "prometheus", "uid": "prometheus-default" },
87+
"fieldConfig": {
88+
"defaults": {
89+
"color": { "mode": "palette-classic" },
90+
"custom": { "drawStyle": "line", "fillOpacity": 10, "lineWidth": 2 },
91+
"unit": "s"
92+
}
93+
},
94+
"gridPos": { "h": 8, "w": 16, "x": 0, "y": 10 },
95+
"id": 120,
96+
"options": {
97+
"legend": { "calcs": ["max", "lastNotNull"], "displayMode": "table", "placement": "bottom" },
98+
"tooltip": { "mode": "single" }
99+
},
100+
"targets": [
101+
{ "expr": "max(lesson_apply_latency_seconds_max{env=\"$env\"}) by (instance, instance_id)", "legendFormat": "Latency: {{instance}}", "refId": "A" }
102+
],
103+
"title": "Async Processing Latency (Max)",
104+
"type": "timeseries"
105+
},
106+
{
107+
"datasource": { "type": "prometheus", "uid": "prometheus-default" },
108+
"fieldConfig": {
109+
"defaults": {
110+
"color": { "mode": "thresholds" },
111+
"mappings": [],
112+
"thresholds": {
113+
"mode": "absolute",
114+
"steps": [
115+
{ "color": "green", "value": null },
116+
{ "color": "red", "value": 1 }
117+
]
118+
}
119+
}
120+
},
121+
"gridPos": { "h": 8, "w": 8, "x": 16, "y": 10 },
122+
"id": 121,
123+
"options": {
124+
"colorMode": "value",
125+
"graphMode": "area",
126+
"justifyMode": "auto",
127+
"orientation": "auto",
128+
"reduceOptions": { "calcs": ["sum"], "fields": "", "values": false },
129+
"textMode": "auto"
130+
},
131+
"targets": [
132+
{ "expr": "sum(lesson_apply_total{env=\"$env\", result=\"pre_filter_reject\"})", "legendFormat": "Rejected Total", "refId": "A" }
133+
],
134+
"title": "[Defend] Pre-filter Reject Count",
135+
"type": "stat"
136+
},
137+
{
138+
"collapsed": false,
139+
"gridPos": { "h": 1, "w": 24, "x": 0, "y": 18 },
140+
"id": 102,
141+
"panels": [],
142+
"title": "Throughput & Latency (HTTP) [$env]",
143+
"type": "row"
144+
},
145+
{
146+
"datasource": { "type": "prometheus", "uid": "prometheus-default" },
147+
"fieldConfig": {
148+
"defaults": {
149+
"color": { "mode": "palette-classic" },
150+
"custom": { "drawStyle": "line", "lineWidth": 2 },
151+
"unit": "reqps"
152+
}
153+
},
154+
"gridPos": { "h": 8, "w": 12, "x": 0, "y": 19 },
155+
"id": 4,
156+
"targets": [
157+
{ "expr": "sum(rate(http_server_requests_seconds_count{env=\"$env\"}[1m])) by (uri, instance)", "legendFormat": "{{uri}} ({{instance}})", "refId": "A" }
158+
],
159+
"title": "Request Throughput (RPS)",
160+
"type": "timeseries"
161+
},
162+
{
163+
"datasource": { "type": "prometheus", "uid": "prometheus-default" },
164+
"fieldConfig": {
165+
"defaults": {
166+
"color": { "mode": "palette-classic" },
167+
"custom": { "drawStyle": "line", "lineWidth": 2 },
168+
"unit": "s"
169+
}
170+
},
171+
"gridPos": { "h": 8, "w": 12, "x": 12, "y": 19 },
172+
"id": 5,
173+
"targets": [
174+
{ "expr": "rate(http_server_requests_seconds_sum{env=\"$env\"}[1m]) / rate(http_server_requests_seconds_count{env=\"$env\"}[1m])", "legendFormat": "{{uri}} ({{instance}})", "refId": "A" }
175+
],
176+
"title": "Average Response Time",
177+
"type": "timeseries"
178+
}
179+
],
180+
"refresh": "5s",
181+
"schemaVersion": 40,
182+
"templating": {
183+
"list": [
184+
{
185+
"current": { "selected": true, "text": "production", "value": "production" },
186+
"hide": 0,
187+
"includeAll": false,
188+
"label": "Environment",
189+
"multi": false,
190+
"name": "env",
191+
"options": [
192+
{ "selected": true, "text": "production", "value": "production" },
193+
{ "selected": false, "text": "baseline", "value": "baseline" }
194+
],
195+
"query": "production,baseline",
196+
"type": "custom"
197+
}
198+
]
199+
},
200+
"time": { "from": "now-15m", "to": "now" },
201+
"title": "TrainUs Master Dashboard",
202+
"uid": "trainus-master-v1",
203+
"version": 4
204+
}

0 commit comments

Comments
 (0)