toad-eye 2.4.5 → 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 +8 -2
- package/dist/instrumentations/create.js +47 -3
- package/dist/instrumentations/types.d.ts +2 -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,
|
|
@@ -9,8 +11,16 @@
|
|
|
9
11
|
{
|
|
10
12
|
"title": "Avg Steps per Query",
|
|
11
13
|
"type": "stat",
|
|
12
|
-
"gridPos": {
|
|
13
|
-
|
|
14
|
+
"gridPos": {
|
|
15
|
+
"h": 4,
|
|
16
|
+
"w": 6,
|
|
17
|
+
"x": 0,
|
|
18
|
+
"y": 0
|
|
19
|
+
},
|
|
20
|
+
"datasource": {
|
|
21
|
+
"type": "prometheus",
|
|
22
|
+
"uid": "prometheus"
|
|
23
|
+
},
|
|
14
24
|
"targets": [
|
|
15
25
|
{
|
|
16
26
|
"expr": "sum(gen_ai_agent_steps_per_query_sum) / sum(gen_ai_agent_steps_per_query_count) or vector(0)",
|
|
@@ -23,9 +33,18 @@
|
|
|
23
33
|
"decimals": 1,
|
|
24
34
|
"thresholds": {
|
|
25
35
|
"steps": [
|
|
26
|
-
{
|
|
27
|
-
|
|
28
|
-
|
|
36
|
+
{
|
|
37
|
+
"color": "green",
|
|
38
|
+
"value": null
|
|
39
|
+
},
|
|
40
|
+
{
|
|
41
|
+
"color": "yellow",
|
|
42
|
+
"value": 10
|
|
43
|
+
},
|
|
44
|
+
{
|
|
45
|
+
"color": "red",
|
|
46
|
+
"value": 20
|
|
47
|
+
}
|
|
29
48
|
]
|
|
30
49
|
}
|
|
31
50
|
},
|
|
@@ -35,8 +54,16 @@
|
|
|
35
54
|
{
|
|
36
55
|
"title": "Total Agent Queries",
|
|
37
56
|
"type": "stat",
|
|
38
|
-
"gridPos": {
|
|
39
|
-
|
|
57
|
+
"gridPos": {
|
|
58
|
+
"h": 4,
|
|
59
|
+
"w": 6,
|
|
60
|
+
"x": 6,
|
|
61
|
+
"y": 0
|
|
62
|
+
},
|
|
63
|
+
"datasource": {
|
|
64
|
+
"type": "prometheus",
|
|
65
|
+
"uid": "prometheus"
|
|
66
|
+
},
|
|
40
67
|
"targets": [
|
|
41
68
|
{
|
|
42
69
|
"expr": "sum(gen_ai_agent_steps_per_query_count)",
|
|
@@ -44,15 +71,25 @@
|
|
|
44
71
|
}
|
|
45
72
|
],
|
|
46
73
|
"fieldConfig": {
|
|
47
|
-
"defaults": {
|
|
74
|
+
"defaults": {
|
|
75
|
+
"unit": "short"
|
|
76
|
+
},
|
|
48
77
|
"overrides": []
|
|
49
78
|
}
|
|
50
79
|
},
|
|
51
80
|
{
|
|
52
81
|
"title": "Total Tool Invocations",
|
|
53
82
|
"type": "stat",
|
|
54
|
-
"gridPos": {
|
|
55
|
-
|
|
83
|
+
"gridPos": {
|
|
84
|
+
"h": 4,
|
|
85
|
+
"w": 6,
|
|
86
|
+
"x": 12,
|
|
87
|
+
"y": 0
|
|
88
|
+
},
|
|
89
|
+
"datasource": {
|
|
90
|
+
"type": "prometheus",
|
|
91
|
+
"uid": "prometheus"
|
|
92
|
+
},
|
|
56
93
|
"targets": [
|
|
57
94
|
{
|
|
58
95
|
"expr": "sum(gen_ai_agent_tool_usage_total)",
|
|
@@ -60,15 +97,25 @@
|
|
|
60
97
|
}
|
|
61
98
|
],
|
|
62
99
|
"fieldConfig": {
|
|
63
|
-
"defaults": {
|
|
100
|
+
"defaults": {
|
|
101
|
+
"unit": "short"
|
|
102
|
+
},
|
|
64
103
|
"overrides": []
|
|
65
104
|
}
|
|
66
105
|
},
|
|
67
106
|
{
|
|
68
107
|
"title": "Unique Tools Used",
|
|
69
108
|
"type": "stat",
|
|
70
|
-
"gridPos": {
|
|
71
|
-
|
|
109
|
+
"gridPos": {
|
|
110
|
+
"h": 4,
|
|
111
|
+
"w": 6,
|
|
112
|
+
"x": 18,
|
|
113
|
+
"y": 0
|
|
114
|
+
},
|
|
115
|
+
"datasource": {
|
|
116
|
+
"type": "prometheus",
|
|
117
|
+
"uid": "prometheus"
|
|
118
|
+
},
|
|
72
119
|
"targets": [
|
|
73
120
|
{
|
|
74
121
|
"expr": "count(count by (gen_ai_agent_tool_name) (gen_ai_agent_tool_usage_total))",
|
|
@@ -76,7 +123,9 @@
|
|
|
76
123
|
}
|
|
77
124
|
],
|
|
78
125
|
"fieldConfig": {
|
|
79
|
-
"defaults": {
|
|
126
|
+
"defaults": {
|
|
127
|
+
"unit": "short"
|
|
128
|
+
},
|
|
80
129
|
"overrides": []
|
|
81
130
|
}
|
|
82
131
|
},
|
|
@@ -84,8 +133,16 @@
|
|
|
84
133
|
"title": "Agent Step Type Breakdown",
|
|
85
134
|
"description": "Distribution of think / act / observe / answer steps",
|
|
86
135
|
"type": "piechart",
|
|
87
|
-
"gridPos": {
|
|
88
|
-
|
|
136
|
+
"gridPos": {
|
|
137
|
+
"h": 8,
|
|
138
|
+
"w": 12,
|
|
139
|
+
"x": 0,
|
|
140
|
+
"y": 4
|
|
141
|
+
},
|
|
142
|
+
"datasource": {
|
|
143
|
+
"type": "prometheus",
|
|
144
|
+
"uid": "prometheus"
|
|
145
|
+
},
|
|
89
146
|
"targets": [
|
|
90
147
|
{
|
|
91
148
|
"expr": "sum by (gen_ai_agent_step_type) (gen_ai_agent_steps_per_query_sum)",
|
|
@@ -106,8 +163,16 @@
|
|
|
106
163
|
"title": "Tool Usage Frequency",
|
|
107
164
|
"description": "Invocations per tool name",
|
|
108
165
|
"type": "barchart",
|
|
109
|
-
"gridPos": {
|
|
110
|
-
|
|
166
|
+
"gridPos": {
|
|
167
|
+
"h": 8,
|
|
168
|
+
"w": 12,
|
|
169
|
+
"x": 12,
|
|
170
|
+
"y": 4
|
|
171
|
+
},
|
|
172
|
+
"datasource": {
|
|
173
|
+
"type": "prometheus",
|
|
174
|
+
"uid": "prometheus"
|
|
175
|
+
},
|
|
111
176
|
"targets": [
|
|
112
177
|
{
|
|
113
178
|
"expr": "sort_desc(sum by (gen_ai_agent_tool_name) (gen_ai_agent_tool_usage_total))",
|
|
@@ -119,8 +184,16 @@
|
|
|
119
184
|
{
|
|
120
185
|
"title": "Steps per Query Over Time",
|
|
121
186
|
"type": "timeseries",
|
|
122
|
-
"gridPos": {
|
|
123
|
-
|
|
187
|
+
"gridPos": {
|
|
188
|
+
"h": 8,
|
|
189
|
+
"w": 24,
|
|
190
|
+
"x": 0,
|
|
191
|
+
"y": 12
|
|
192
|
+
},
|
|
193
|
+
"datasource": {
|
|
194
|
+
"type": "prometheus",
|
|
195
|
+
"uid": "prometheus"
|
|
196
|
+
},
|
|
124
197
|
"targets": [
|
|
125
198
|
{
|
|
126
199
|
"expr": "sum(rate(gen_ai_agent_steps_per_query_sum[5m])) / sum(rate(gen_ai_agent_steps_per_query_count[5m]))",
|
|
@@ -136,7 +209,9 @@
|
|
|
136
209
|
"fieldConfig": {
|
|
137
210
|
"defaults": {
|
|
138
211
|
"unit": "short",
|
|
139
|
-
"custom": {
|
|
212
|
+
"custom": {
|
|
213
|
+
"fillOpacity": 10
|
|
214
|
+
}
|
|
140
215
|
},
|
|
141
216
|
"overrides": []
|
|
142
217
|
}
|
|
@@ -144,8 +219,16 @@
|
|
|
144
219
|
{
|
|
145
220
|
"title": "Tool Invocations Over Time",
|
|
146
221
|
"type": "timeseries",
|
|
147
|
-
"gridPos": {
|
|
148
|
-
|
|
222
|
+
"gridPos": {
|
|
223
|
+
"h": 8,
|
|
224
|
+
"w": 12,
|
|
225
|
+
"x": 0,
|
|
226
|
+
"y": 20
|
|
227
|
+
},
|
|
228
|
+
"datasource": {
|
|
229
|
+
"type": "prometheus",
|
|
230
|
+
"uid": "prometheus"
|
|
231
|
+
},
|
|
149
232
|
"targets": [
|
|
150
233
|
{
|
|
151
234
|
"expr": "sum by (gen_ai_agent_tool_name) (rate(gen_ai_agent_tool_usage_total[5m]))",
|
|
@@ -156,7 +239,12 @@
|
|
|
156
239
|
"fieldConfig": {
|
|
157
240
|
"defaults": {
|
|
158
241
|
"unit": "reqps",
|
|
159
|
-
"custom": {
|
|
242
|
+
"custom": {
|
|
243
|
+
"fillOpacity": 20,
|
|
244
|
+
"stacking": {
|
|
245
|
+
"mode": "normal"
|
|
246
|
+
}
|
|
247
|
+
}
|
|
160
248
|
},
|
|
161
249
|
"overrides": []
|
|
162
250
|
}
|
|
@@ -165,8 +253,16 @@
|
|
|
165
253
|
"title": "Steps Distribution Histogram",
|
|
166
254
|
"description": "How many steps do agent queries typically take?",
|
|
167
255
|
"type": "histogram",
|
|
168
|
-
"gridPos": {
|
|
169
|
-
|
|
256
|
+
"gridPos": {
|
|
257
|
+
"h": 8,
|
|
258
|
+
"w": 12,
|
|
259
|
+
"x": 12,
|
|
260
|
+
"y": 20
|
|
261
|
+
},
|
|
262
|
+
"datasource": {
|
|
263
|
+
"type": "prometheus",
|
|
264
|
+
"uid": "prometheus"
|
|
265
|
+
},
|
|
170
266
|
"targets": [
|
|
171
267
|
{
|
|
172
268
|
"expr": "sum by (le) (gen_ai_agent_steps_per_query_bucket)",
|
|
@@ -176,7 +272,9 @@
|
|
|
176
272
|
],
|
|
177
273
|
"fieldConfig": {
|
|
178
274
|
"defaults": {
|
|
179
|
-
"custom": {
|
|
275
|
+
"custom": {
|
|
276
|
+
"fillOpacity": 80
|
|
277
|
+
}
|
|
180
278
|
},
|
|
181
279
|
"overrides": []
|
|
182
280
|
}
|
|
@@ -186,19 +284,44 @@
|
|
|
186
284
|
"tags": ["toad-eye", "agent"],
|
|
187
285
|
"templating": {
|
|
188
286
|
"list": [
|
|
287
|
+
{
|
|
288
|
+
"name": "service",
|
|
289
|
+
"type": "query",
|
|
290
|
+
"datasource": {
|
|
291
|
+
"type": "prometheus",
|
|
292
|
+
"uid": "prometheus"
|
|
293
|
+
},
|
|
294
|
+
"query": "label_values(gen_ai_client_requests_total, job)",
|
|
295
|
+
"includeAll": true,
|
|
296
|
+
"allValue": ".*",
|
|
297
|
+
"current": {
|
|
298
|
+
"text": "All",
|
|
299
|
+
"value": "$__all"
|
|
300
|
+
},
|
|
301
|
+
"multi": true
|
|
302
|
+
},
|
|
189
303
|
{
|
|
190
304
|
"name": "tool",
|
|
191
305
|
"type": "query",
|
|
192
|
-
"datasource": {
|
|
306
|
+
"datasource": {
|
|
307
|
+
"type": "prometheus",
|
|
308
|
+
"uid": "prometheus"
|
|
309
|
+
},
|
|
193
310
|
"query": "label_values(gen_ai_agent_tool_usage_total, gen_ai_agent_tool_name)",
|
|
194
311
|
"includeAll": true,
|
|
195
312
|
"multi": true,
|
|
196
|
-
"current": {
|
|
313
|
+
"current": {
|
|
314
|
+
"text": "All",
|
|
315
|
+
"value": "$__all"
|
|
316
|
+
},
|
|
197
317
|
"regex": ""
|
|
198
318
|
}
|
|
199
319
|
]
|
|
200
320
|
},
|
|
201
|
-
"time": {
|
|
321
|
+
"time": {
|
|
322
|
+
"from": "now-6h",
|
|
323
|
+
"to": "now"
|
|
324
|
+
},
|
|
202
325
|
"timepicker": {},
|
|
203
326
|
"timezone": "utc",
|
|
204
327
|
"title": "toad-eye / Agent Workflow",
|
|
@@ -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",
|