toad-eye 2.4.4 → 2.5.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +1 -1
- package/dist/agent.js +6 -2
- package/dist/context/error.d.ts +15 -0
- package/dist/context/error.js +17 -0
- package/dist/context/guard.d.ts +8 -0
- package/dist/context/guard.js +41 -0
- package/dist/context/index.d.ts +2 -0
- package/dist/context/index.js +2 -0
- package/dist/core/metrics.d.ts +4 -1
- package/dist/core/metrics.js +32 -2
- package/dist/core/pricing.d.ts +2 -0
- package/dist/core/pricing.js +51 -15
- package/dist/core/spans.js +27 -2
- package/dist/core/tracer.js +2 -0
- package/dist/index.d.ts +1 -0
- package/dist/index.js +1 -0
- package/dist/instrumentations/anthropic.js +25 -2
- package/dist/instrumentations/create.js +61 -3
- package/dist/instrumentations/gemini.js +12 -0
- package/dist/instrumentations/openai.js +20 -3
- package/dist/instrumentations/types.d.ts +9 -0
- package/dist/types/attributes.d.ts +1 -0
- package/dist/types/attributes.js +1 -0
- package/dist/types/config.d.ts +7 -0
- package/dist/types/metrics.d.ts +3 -0
- package/dist/types/metrics.js +4 -0
- package/dist/types/spans.d.ts +4 -0
- package/package.json +1 -1
- package/templates/grafana/dashboards/agent-workflow.json +154 -31
- package/templates/grafana/dashboards/cost-breakdown.json +168 -40
- package/templates/grafana/dashboards/error-drilldown.json +155 -37
- package/templates/grafana/dashboards/finops-attribution.json +192 -43
- package/templates/grafana/dashboards/latency-analysis.json +103 -24
- package/templates/grafana/dashboards/llm-overview.json +268 -66
- package/templates/grafana/dashboards/model-comparison.json +119 -26
- package/templates/grafana/dashboards/provider-health.json +137 -31
|
@@ -1,5 +1,7 @@
|
|
|
1
1
|
{
|
|
2
|
-
"annotations": {
|
|
2
|
+
"annotations": {
|
|
3
|
+
"list": []
|
|
4
|
+
},
|
|
3
5
|
"editable": true,
|
|
4
6
|
"fiscalYearStartMonth": 0,
|
|
5
7
|
"graphTooltip": 1,
|
|
@@ -9,11 +11,19 @@
|
|
|
9
11
|
{
|
|
10
12
|
"title": "Total Cost",
|
|
11
13
|
"type": "stat",
|
|
12
|
-
"gridPos": {
|
|
13
|
-
|
|
14
|
+
"gridPos": {
|
|
15
|
+
"h": 4,
|
|
16
|
+
"w": 8,
|
|
17
|
+
"x": 0,
|
|
18
|
+
"y": 0
|
|
19
|
+
},
|
|
20
|
+
"datasource": {
|
|
21
|
+
"type": "prometheus",
|
|
22
|
+
"uid": "prometheus"
|
|
23
|
+
},
|
|
14
24
|
"targets": [
|
|
15
25
|
{
|
|
16
|
-
"expr": "sum(gen_ai_client_request_cost_USD_sum{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"})",
|
|
26
|
+
"expr": "sum(gen_ai_client_request_cost_USD_sum{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"})",
|
|
17
27
|
"refId": "A"
|
|
18
28
|
}
|
|
19
29
|
],
|
|
@@ -22,9 +32,18 @@
|
|
|
22
32
|
"unit": "currencyUSD",
|
|
23
33
|
"thresholds": {
|
|
24
34
|
"steps": [
|
|
25
|
-
{
|
|
26
|
-
|
|
27
|
-
|
|
35
|
+
{
|
|
36
|
+
"color": "green",
|
|
37
|
+
"value": null
|
|
38
|
+
},
|
|
39
|
+
{
|
|
40
|
+
"color": "yellow",
|
|
41
|
+
"value": 50
|
|
42
|
+
},
|
|
43
|
+
{
|
|
44
|
+
"color": "red",
|
|
45
|
+
"value": 500
|
|
46
|
+
}
|
|
28
47
|
]
|
|
29
48
|
}
|
|
30
49
|
},
|
|
@@ -34,11 +53,19 @@
|
|
|
34
53
|
{
|
|
35
54
|
"title": "Cost Today",
|
|
36
55
|
"type": "stat",
|
|
37
|
-
"gridPos": {
|
|
38
|
-
|
|
56
|
+
"gridPos": {
|
|
57
|
+
"h": 4,
|
|
58
|
+
"w": 8,
|
|
59
|
+
"x": 8,
|
|
60
|
+
"y": 0
|
|
61
|
+
},
|
|
62
|
+
"datasource": {
|
|
63
|
+
"type": "prometheus",
|
|
64
|
+
"uid": "prometheus"
|
|
65
|
+
},
|
|
39
66
|
"targets": [
|
|
40
67
|
{
|
|
41
|
-
"expr": "sum(increase(gen_ai_client_request_cost_USD_sum{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1d]))",
|
|
68
|
+
"expr": "sum(increase(gen_ai_client_request_cost_USD_sum{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1d]))",
|
|
42
69
|
"refId": "A"
|
|
43
70
|
}
|
|
44
71
|
],
|
|
@@ -47,9 +74,18 @@
|
|
|
47
74
|
"unit": "currencyUSD",
|
|
48
75
|
"thresholds": {
|
|
49
76
|
"steps": [
|
|
50
|
-
{
|
|
51
|
-
|
|
52
|
-
|
|
77
|
+
{
|
|
78
|
+
"color": "green",
|
|
79
|
+
"value": null
|
|
80
|
+
},
|
|
81
|
+
{
|
|
82
|
+
"color": "yellow",
|
|
83
|
+
"value": 10
|
|
84
|
+
},
|
|
85
|
+
{
|
|
86
|
+
"color": "red",
|
|
87
|
+
"value": 50
|
|
88
|
+
}
|
|
53
89
|
]
|
|
54
90
|
}
|
|
55
91
|
},
|
|
@@ -59,11 +95,19 @@
|
|
|
59
95
|
{
|
|
60
96
|
"title": "Projected Monthly Cost",
|
|
61
97
|
"type": "stat",
|
|
62
|
-
"gridPos": {
|
|
63
|
-
|
|
98
|
+
"gridPos": {
|
|
99
|
+
"h": 4,
|
|
100
|
+
"w": 8,
|
|
101
|
+
"x": 16,
|
|
102
|
+
"y": 0
|
|
103
|
+
},
|
|
104
|
+
"datasource": {
|
|
105
|
+
"type": "prometheus",
|
|
106
|
+
"uid": "prometheus"
|
|
107
|
+
},
|
|
64
108
|
"targets": [
|
|
65
109
|
{
|
|
66
|
-
"expr": "sum(rate(gen_ai_client_request_cost_USD_sum{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1d])) * 86400 * 30",
|
|
110
|
+
"expr": "sum(rate(gen_ai_client_request_cost_USD_sum{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1d])) * 86400 * 30",
|
|
67
111
|
"refId": "A"
|
|
68
112
|
}
|
|
69
113
|
],
|
|
@@ -72,9 +116,18 @@
|
|
|
72
116
|
"unit": "currencyUSD",
|
|
73
117
|
"thresholds": {
|
|
74
118
|
"steps": [
|
|
75
|
-
{
|
|
76
|
-
|
|
77
|
-
|
|
119
|
+
{
|
|
120
|
+
"color": "green",
|
|
121
|
+
"value": null
|
|
122
|
+
},
|
|
123
|
+
{
|
|
124
|
+
"color": "yellow",
|
|
125
|
+
"value": 100
|
|
126
|
+
},
|
|
127
|
+
{
|
|
128
|
+
"color": "red",
|
|
129
|
+
"value": 1000
|
|
130
|
+
}
|
|
78
131
|
]
|
|
79
132
|
}
|
|
80
133
|
},
|
|
@@ -84,11 +137,19 @@
|
|
|
84
137
|
{
|
|
85
138
|
"title": "Cost by Provider",
|
|
86
139
|
"type": "timeseries",
|
|
87
|
-
"gridPos": {
|
|
88
|
-
|
|
140
|
+
"gridPos": {
|
|
141
|
+
"h": 8,
|
|
142
|
+
"w": 12,
|
|
143
|
+
"x": 0,
|
|
144
|
+
"y": 4
|
|
145
|
+
},
|
|
146
|
+
"datasource": {
|
|
147
|
+
"type": "prometheus",
|
|
148
|
+
"uid": "prometheus"
|
|
149
|
+
},
|
|
89
150
|
"targets": [
|
|
90
151
|
{
|
|
91
|
-
"expr": "sum(rate(gen_ai_client_request_cost_USD_sum{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) by (gen_ai_provider_name) * 60",
|
|
152
|
+
"expr": "sum(rate(gen_ai_client_request_cost_USD_sum{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) by (gen_ai_provider_name) * 60",
|
|
92
153
|
"legendFormat": "{{ gen_ai_provider_name }}",
|
|
93
154
|
"refId": "A"
|
|
94
155
|
}
|
|
@@ -100,7 +161,9 @@
|
|
|
100
161
|
"drawStyle": "bars",
|
|
101
162
|
"fillOpacity": 50,
|
|
102
163
|
"lineWidth": 1,
|
|
103
|
-
"stacking": {
|
|
164
|
+
"stacking": {
|
|
165
|
+
"mode": "normal"
|
|
166
|
+
}
|
|
104
167
|
}
|
|
105
168
|
},
|
|
106
169
|
"overrides": []
|
|
@@ -109,11 +172,19 @@
|
|
|
109
172
|
{
|
|
110
173
|
"title": "Cost by Model",
|
|
111
174
|
"type": "timeseries",
|
|
112
|
-
"gridPos": {
|
|
113
|
-
|
|
175
|
+
"gridPos": {
|
|
176
|
+
"h": 8,
|
|
177
|
+
"w": 12,
|
|
178
|
+
"x": 12,
|
|
179
|
+
"y": 4
|
|
180
|
+
},
|
|
181
|
+
"datasource": {
|
|
182
|
+
"type": "prometheus",
|
|
183
|
+
"uid": "prometheus"
|
|
184
|
+
},
|
|
114
185
|
"targets": [
|
|
115
186
|
{
|
|
116
|
-
"expr": "sum(rate(gen_ai_client_request_cost_USD_sum{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) by (gen_ai_request_model) * 60",
|
|
187
|
+
"expr": "sum(rate(gen_ai_client_request_cost_USD_sum{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) by (gen_ai_request_model) * 60",
|
|
117
188
|
"legendFormat": "{{ gen_ai_request_model }}",
|
|
118
189
|
"refId": "A"
|
|
119
190
|
}
|
|
@@ -125,7 +196,9 @@
|
|
|
125
196
|
"drawStyle": "bars",
|
|
126
197
|
"fillOpacity": 50,
|
|
127
198
|
"lineWidth": 1,
|
|
128
|
-
"stacking": {
|
|
199
|
+
"stacking": {
|
|
200
|
+
"mode": "normal"
|
|
201
|
+
}
|
|
129
202
|
}
|
|
130
203
|
},
|
|
131
204
|
"overrides": []
|
|
@@ -134,11 +207,19 @@
|
|
|
134
207
|
{
|
|
135
208
|
"title": "Cost per Request",
|
|
136
209
|
"type": "timeseries",
|
|
137
|
-
"gridPos": {
|
|
138
|
-
|
|
210
|
+
"gridPos": {
|
|
211
|
+
"h": 8,
|
|
212
|
+
"w": 12,
|
|
213
|
+
"x": 0,
|
|
214
|
+
"y": 12
|
|
215
|
+
},
|
|
216
|
+
"datasource": {
|
|
217
|
+
"type": "prometheus",
|
|
218
|
+
"uid": "prometheus"
|
|
219
|
+
},
|
|
139
220
|
"targets": [
|
|
140
221
|
{
|
|
141
|
-
"expr": "sum(rate(gen_ai_client_request_cost_USD_sum{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[5m])) / sum(rate(gen_ai_client_request_cost_USD_count{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[5m]))",
|
|
222
|
+
"expr": "sum(rate(gen_ai_client_request_cost_USD_sum{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[5m])) / sum(rate(gen_ai_client_request_cost_USD_count{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[5m]))",
|
|
142
223
|
"legendFormat": "avg cost/request",
|
|
143
224
|
"refId": "A"
|
|
144
225
|
}
|
|
@@ -146,7 +227,11 @@
|
|
|
146
227
|
"fieldConfig": {
|
|
147
228
|
"defaults": {
|
|
148
229
|
"unit": "currencyUSD",
|
|
149
|
-
"custom": {
|
|
230
|
+
"custom": {
|
|
231
|
+
"drawStyle": "line",
|
|
232
|
+
"fillOpacity": 15,
|
|
233
|
+
"lineWidth": 2
|
|
234
|
+
}
|
|
150
235
|
},
|
|
151
236
|
"overrides": []
|
|
152
237
|
}
|
|
@@ -154,11 +239,19 @@
|
|
|
154
239
|
{
|
|
155
240
|
"title": "Daily Cost Trend",
|
|
156
241
|
"type": "timeseries",
|
|
157
|
-
"gridPos": {
|
|
158
|
-
|
|
242
|
+
"gridPos": {
|
|
243
|
+
"h": 8,
|
|
244
|
+
"w": 12,
|
|
245
|
+
"x": 12,
|
|
246
|
+
"y": 12
|
|
247
|
+
},
|
|
248
|
+
"datasource": {
|
|
249
|
+
"type": "prometheus",
|
|
250
|
+
"uid": "prometheus"
|
|
251
|
+
},
|
|
159
252
|
"targets": [
|
|
160
253
|
{
|
|
161
|
-
"expr": "sum(increase(gen_ai_client_request_cost_USD_sum{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1h]))",
|
|
254
|
+
"expr": "sum(increase(gen_ai_client_request_cost_USD_sum{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1h]))",
|
|
162
255
|
"legendFormat": "hourly cost",
|
|
163
256
|
"refId": "A"
|
|
164
257
|
}
|
|
@@ -166,7 +259,11 @@
|
|
|
166
259
|
"fieldConfig": {
|
|
167
260
|
"defaults": {
|
|
168
261
|
"unit": "currencyUSD",
|
|
169
|
-
"custom": {
|
|
262
|
+
"custom": {
|
|
263
|
+
"drawStyle": "line",
|
|
264
|
+
"fillOpacity": 15,
|
|
265
|
+
"lineWidth": 2
|
|
266
|
+
}
|
|
170
267
|
},
|
|
171
268
|
"overrides": []
|
|
172
269
|
}
|
|
@@ -177,29 +274,60 @@
|
|
|
177
274
|
"tags": ["llm", "observability", "toad-eye"],
|
|
178
275
|
"templating": {
|
|
179
276
|
"list": [
|
|
277
|
+
{
|
|
278
|
+
"name": "service",
|
|
279
|
+
"type": "query",
|
|
280
|
+
"datasource": {
|
|
281
|
+
"type": "prometheus",
|
|
282
|
+
"uid": "prometheus"
|
|
283
|
+
},
|
|
284
|
+
"query": "label_values(gen_ai_client_requests_total, job)",
|
|
285
|
+
"includeAll": true,
|
|
286
|
+
"allValue": ".*",
|
|
287
|
+
"current": {
|
|
288
|
+
"text": "All",
|
|
289
|
+
"value": "$__all"
|
|
290
|
+
},
|
|
291
|
+
"multi": true
|
|
292
|
+
},
|
|
180
293
|
{
|
|
181
294
|
"name": "provider",
|
|
182
295
|
"type": "query",
|
|
183
|
-
"datasource": {
|
|
296
|
+
"datasource": {
|
|
297
|
+
"type": "prometheus",
|
|
298
|
+
"uid": "prometheus"
|
|
299
|
+
},
|
|
184
300
|
"query": "label_values(gen_ai_client_requests_total, gen_ai_provider_name)",
|
|
185
301
|
"includeAll": true,
|
|
186
302
|
"allValue": ".*",
|
|
187
|
-
"current": {
|
|
303
|
+
"current": {
|
|
304
|
+
"text": "All",
|
|
305
|
+
"value": "$__all"
|
|
306
|
+
},
|
|
188
307
|
"multi": true
|
|
189
308
|
},
|
|
190
309
|
{
|
|
191
310
|
"name": "model",
|
|
192
311
|
"type": "query",
|
|
193
|
-
"datasource": {
|
|
312
|
+
"datasource": {
|
|
313
|
+
"type": "prometheus",
|
|
314
|
+
"uid": "prometheus"
|
|
315
|
+
},
|
|
194
316
|
"query": "label_values(gen_ai_client_requests_total{gen_ai_provider_name=~\"$provider\"}, gen_ai_request_model)",
|
|
195
317
|
"includeAll": true,
|
|
196
318
|
"allValue": ".*",
|
|
197
|
-
"current": {
|
|
319
|
+
"current": {
|
|
320
|
+
"text": "All",
|
|
321
|
+
"value": "$__all"
|
|
322
|
+
},
|
|
198
323
|
"multi": true
|
|
199
324
|
}
|
|
200
325
|
]
|
|
201
326
|
},
|
|
202
|
-
"time": {
|
|
327
|
+
"time": {
|
|
328
|
+
"from": "now-15m",
|
|
329
|
+
"to": "now"
|
|
330
|
+
},
|
|
203
331
|
"timepicker": {},
|
|
204
332
|
"timezone": "browser",
|
|
205
333
|
"title": "toad-eye: Cost Breakdown",
|
|
@@ -1,5 +1,7 @@
|
|
|
1
1
|
{
|
|
2
|
-
"annotations": {
|
|
2
|
+
"annotations": {
|
|
3
|
+
"list": []
|
|
4
|
+
},
|
|
3
5
|
"editable": true,
|
|
4
6
|
"fiscalYearStartMonth": 0,
|
|
5
7
|
"graphTooltip": 1,
|
|
@@ -9,11 +11,19 @@
|
|
|
9
11
|
{
|
|
10
12
|
"title": "Total Errors",
|
|
11
13
|
"type": "stat",
|
|
12
|
-
"gridPos": {
|
|
13
|
-
|
|
14
|
+
"gridPos": {
|
|
15
|
+
"h": 4,
|
|
16
|
+
"w": 8,
|
|
17
|
+
"x": 0,
|
|
18
|
+
"y": 0
|
|
19
|
+
},
|
|
20
|
+
"datasource": {
|
|
21
|
+
"type": "prometheus",
|
|
22
|
+
"uid": "prometheus"
|
|
23
|
+
},
|
|
14
24
|
"targets": [
|
|
15
25
|
{
|
|
16
|
-
"expr": "sum(gen_ai_client_errors_total{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"})",
|
|
26
|
+
"expr": "sum(gen_ai_client_errors_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"})",
|
|
17
27
|
"refId": "A"
|
|
18
28
|
}
|
|
19
29
|
],
|
|
@@ -22,9 +32,18 @@
|
|
|
22
32
|
"unit": "short",
|
|
23
33
|
"thresholds": {
|
|
24
34
|
"steps": [
|
|
25
|
-
{
|
|
26
|
-
|
|
27
|
-
|
|
35
|
+
{
|
|
36
|
+
"color": "green",
|
|
37
|
+
"value": null
|
|
38
|
+
},
|
|
39
|
+
{
|
|
40
|
+
"color": "yellow",
|
|
41
|
+
"value": 10
|
|
42
|
+
},
|
|
43
|
+
{
|
|
44
|
+
"color": "red",
|
|
45
|
+
"value": 100
|
|
46
|
+
}
|
|
28
47
|
]
|
|
29
48
|
}
|
|
30
49
|
},
|
|
@@ -34,11 +53,19 @@
|
|
|
34
53
|
{
|
|
35
54
|
"title": "Error Rate %",
|
|
36
55
|
"type": "stat",
|
|
37
|
-
"gridPos": {
|
|
38
|
-
|
|
56
|
+
"gridPos": {
|
|
57
|
+
"h": 4,
|
|
58
|
+
"w": 8,
|
|
59
|
+
"x": 8,
|
|
60
|
+
"y": 0
|
|
61
|
+
},
|
|
62
|
+
"datasource": {
|
|
63
|
+
"type": "prometheus",
|
|
64
|
+
"uid": "prometheus"
|
|
65
|
+
},
|
|
39
66
|
"targets": [
|
|
40
67
|
{
|
|
41
|
-
"expr": "sum(gen_ai_client_errors_total{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}) / sum(gen_ai_client_requests_total{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}) * 100",
|
|
68
|
+
"expr": "sum(gen_ai_client_errors_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}) / sum(gen_ai_client_requests_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}) * 100",
|
|
42
69
|
"refId": "A"
|
|
43
70
|
}
|
|
44
71
|
],
|
|
@@ -47,9 +74,18 @@
|
|
|
47
74
|
"unit": "percent",
|
|
48
75
|
"thresholds": {
|
|
49
76
|
"steps": [
|
|
50
|
-
{
|
|
51
|
-
|
|
52
|
-
|
|
77
|
+
{
|
|
78
|
+
"color": "green",
|
|
79
|
+
"value": null
|
|
80
|
+
},
|
|
81
|
+
{
|
|
82
|
+
"color": "yellow",
|
|
83
|
+
"value": 5
|
|
84
|
+
},
|
|
85
|
+
{
|
|
86
|
+
"color": "red",
|
|
87
|
+
"value": 20
|
|
88
|
+
}
|
|
53
89
|
]
|
|
54
90
|
}
|
|
55
91
|
},
|
|
@@ -59,11 +95,19 @@
|
|
|
59
95
|
{
|
|
60
96
|
"title": "Errors Last Hour",
|
|
61
97
|
"type": "stat",
|
|
62
|
-
"gridPos": {
|
|
63
|
-
|
|
98
|
+
"gridPos": {
|
|
99
|
+
"h": 4,
|
|
100
|
+
"w": 8,
|
|
101
|
+
"x": 16,
|
|
102
|
+
"y": 0
|
|
103
|
+
},
|
|
104
|
+
"datasource": {
|
|
105
|
+
"type": "prometheus",
|
|
106
|
+
"uid": "prometheus"
|
|
107
|
+
},
|
|
64
108
|
"targets": [
|
|
65
109
|
{
|
|
66
|
-
"expr": "sum(increase(gen_ai_client_errors_total{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1h]))",
|
|
110
|
+
"expr": "sum(increase(gen_ai_client_errors_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1h]))",
|
|
67
111
|
"refId": "A"
|
|
68
112
|
}
|
|
69
113
|
],
|
|
@@ -72,9 +116,18 @@
|
|
|
72
116
|
"unit": "short",
|
|
73
117
|
"thresholds": {
|
|
74
118
|
"steps": [
|
|
75
|
-
{
|
|
76
|
-
|
|
77
|
-
|
|
119
|
+
{
|
|
120
|
+
"color": "green",
|
|
121
|
+
"value": null
|
|
122
|
+
},
|
|
123
|
+
{
|
|
124
|
+
"color": "yellow",
|
|
125
|
+
"value": 5
|
|
126
|
+
},
|
|
127
|
+
{
|
|
128
|
+
"color": "red",
|
|
129
|
+
"value": 50
|
|
130
|
+
}
|
|
78
131
|
]
|
|
79
132
|
}
|
|
80
133
|
},
|
|
@@ -84,11 +137,19 @@
|
|
|
84
137
|
{
|
|
85
138
|
"title": "Error Rate by Provider",
|
|
86
139
|
"type": "timeseries",
|
|
87
|
-
"gridPos": {
|
|
88
|
-
|
|
140
|
+
"gridPos": {
|
|
141
|
+
"h": 8,
|
|
142
|
+
"w": 12,
|
|
143
|
+
"x": 0,
|
|
144
|
+
"y": 4
|
|
145
|
+
},
|
|
146
|
+
"datasource": {
|
|
147
|
+
"type": "prometheus",
|
|
148
|
+
"uid": "prometheus"
|
|
149
|
+
},
|
|
89
150
|
"targets": [
|
|
90
151
|
{
|
|
91
|
-
"expr": "sum(rate(gen_ai_client_errors_total{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) by (gen_ai_provider_name) * 60",
|
|
152
|
+
"expr": "sum(rate(gen_ai_client_errors_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) by (gen_ai_provider_name) * 60",
|
|
92
153
|
"legendFormat": "{{ gen_ai_provider_name }}",
|
|
93
154
|
"refId": "A"
|
|
94
155
|
}
|
|
@@ -96,7 +157,11 @@
|
|
|
96
157
|
"fieldConfig": {
|
|
97
158
|
"defaults": {
|
|
98
159
|
"unit": "reqpm",
|
|
99
|
-
"custom": {
|
|
160
|
+
"custom": {
|
|
161
|
+
"drawStyle": "line",
|
|
162
|
+
"fillOpacity": 15,
|
|
163
|
+
"lineWidth": 2
|
|
164
|
+
}
|
|
100
165
|
},
|
|
101
166
|
"overrides": []
|
|
102
167
|
}
|
|
@@ -104,11 +169,19 @@
|
|
|
104
169
|
{
|
|
105
170
|
"title": "Error Rate by Model",
|
|
106
171
|
"type": "timeseries",
|
|
107
|
-
"gridPos": {
|
|
108
|
-
|
|
172
|
+
"gridPos": {
|
|
173
|
+
"h": 8,
|
|
174
|
+
"w": 12,
|
|
175
|
+
"x": 12,
|
|
176
|
+
"y": 4
|
|
177
|
+
},
|
|
178
|
+
"datasource": {
|
|
179
|
+
"type": "prometheus",
|
|
180
|
+
"uid": "prometheus"
|
|
181
|
+
},
|
|
109
182
|
"targets": [
|
|
110
183
|
{
|
|
111
|
-
"expr": "sum(rate(gen_ai_client_errors_total{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) by (gen_ai_request_model) * 60",
|
|
184
|
+
"expr": "sum(rate(gen_ai_client_errors_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) by (gen_ai_request_model) * 60",
|
|
112
185
|
"legendFormat": "{{ gen_ai_request_model }}",
|
|
113
186
|
"refId": "A"
|
|
114
187
|
}
|
|
@@ -116,7 +189,11 @@
|
|
|
116
189
|
"fieldConfig": {
|
|
117
190
|
"defaults": {
|
|
118
191
|
"unit": "reqpm",
|
|
119
|
-
"custom": {
|
|
192
|
+
"custom": {
|
|
193
|
+
"drawStyle": "line",
|
|
194
|
+
"fillOpacity": 15,
|
|
195
|
+
"lineWidth": 2
|
|
196
|
+
}
|
|
120
197
|
},
|
|
121
198
|
"overrides": []
|
|
122
199
|
}
|
|
@@ -124,16 +201,24 @@
|
|
|
124
201
|
{
|
|
125
202
|
"title": "Error vs Success Rate",
|
|
126
203
|
"type": "timeseries",
|
|
127
|
-
"gridPos": {
|
|
128
|
-
|
|
204
|
+
"gridPos": {
|
|
205
|
+
"h": 8,
|
|
206
|
+
"w": 24,
|
|
207
|
+
"x": 0,
|
|
208
|
+
"y": 12
|
|
209
|
+
},
|
|
210
|
+
"datasource": {
|
|
211
|
+
"type": "prometheus",
|
|
212
|
+
"uid": "prometheus"
|
|
213
|
+
},
|
|
129
214
|
"targets": [
|
|
130
215
|
{
|
|
131
|
-
"expr": "sum(rate(gen_ai_client_errors_total{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) * 60",
|
|
216
|
+
"expr": "sum(rate(gen_ai_client_errors_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) * 60",
|
|
132
217
|
"legendFormat": "errors/min",
|
|
133
218
|
"refId": "A"
|
|
134
219
|
},
|
|
135
220
|
{
|
|
136
|
-
"expr": "(sum(rate(gen_ai_client_requests_total{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) - sum(rate(gen_ai_client_errors_total{gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m]))) * 60",
|
|
221
|
+
"expr": "(sum(rate(gen_ai_client_requests_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m])) - sum(rate(gen_ai_client_errors_total{job=~\"$service\", gen_ai_provider_name=~\"$provider\", gen_ai_request_model=~\"$model\"}[1m]))) * 60",
|
|
137
222
|
"legendFormat": "success/min",
|
|
138
223
|
"refId": "B"
|
|
139
224
|
}
|
|
@@ -145,7 +230,9 @@
|
|
|
145
230
|
"drawStyle": "line",
|
|
146
231
|
"fillOpacity": 15,
|
|
147
232
|
"lineWidth": 2,
|
|
148
|
-
"stacking": {
|
|
233
|
+
"stacking": {
|
|
234
|
+
"mode": "normal"
|
|
235
|
+
}
|
|
149
236
|
}
|
|
150
237
|
},
|
|
151
238
|
"overrides": []
|
|
@@ -157,29 +244,60 @@
|
|
|
157
244
|
"tags": ["llm", "observability", "toad-eye"],
|
|
158
245
|
"templating": {
|
|
159
246
|
"list": [
|
|
247
|
+
{
|
|
248
|
+
"name": "service",
|
|
249
|
+
"type": "query",
|
|
250
|
+
"datasource": {
|
|
251
|
+
"type": "prometheus",
|
|
252
|
+
"uid": "prometheus"
|
|
253
|
+
},
|
|
254
|
+
"query": "label_values(gen_ai_client_requests_total, job)",
|
|
255
|
+
"includeAll": true,
|
|
256
|
+
"allValue": ".*",
|
|
257
|
+
"current": {
|
|
258
|
+
"text": "All",
|
|
259
|
+
"value": "$__all"
|
|
260
|
+
},
|
|
261
|
+
"multi": true
|
|
262
|
+
},
|
|
160
263
|
{
|
|
161
264
|
"name": "provider",
|
|
162
265
|
"type": "query",
|
|
163
|
-
"datasource": {
|
|
266
|
+
"datasource": {
|
|
267
|
+
"type": "prometheus",
|
|
268
|
+
"uid": "prometheus"
|
|
269
|
+
},
|
|
164
270
|
"query": "label_values(gen_ai_client_requests_total, gen_ai_provider_name)",
|
|
165
271
|
"includeAll": true,
|
|
166
272
|
"allValue": ".*",
|
|
167
|
-
"current": {
|
|
273
|
+
"current": {
|
|
274
|
+
"text": "All",
|
|
275
|
+
"value": "$__all"
|
|
276
|
+
},
|
|
168
277
|
"multi": true
|
|
169
278
|
},
|
|
170
279
|
{
|
|
171
280
|
"name": "model",
|
|
172
281
|
"type": "query",
|
|
173
|
-
"datasource": {
|
|
282
|
+
"datasource": {
|
|
283
|
+
"type": "prometheus",
|
|
284
|
+
"uid": "prometheus"
|
|
285
|
+
},
|
|
174
286
|
"query": "label_values(gen_ai_client_requests_total{gen_ai_provider_name=~\"$provider\"}, gen_ai_request_model)",
|
|
175
287
|
"includeAll": true,
|
|
176
288
|
"allValue": ".*",
|
|
177
|
-
"current": {
|
|
289
|
+
"current": {
|
|
290
|
+
"text": "All",
|
|
291
|
+
"value": "$__all"
|
|
292
|
+
},
|
|
178
293
|
"multi": true
|
|
179
294
|
}
|
|
180
295
|
]
|
|
181
296
|
},
|
|
182
|
-
"time": {
|
|
297
|
+
"time": {
|
|
298
|
+
"from": "now-15m",
|
|
299
|
+
"to": "now"
|
|
300
|
+
},
|
|
183
301
|
"timepicker": {},
|
|
184
302
|
"timezone": "browser",
|
|
185
303
|
"title": "toad-eye: Error Drill-down",
|