@smithers-orchestrator/observability 0.23.0 → 0.24.0
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/docker-compose.otel.yml +104 -0
- package/grafana/dashboards/smithers-dashboard.json +453 -0
- package/grafana/provisioning/dashboards/dashboards.yml +12 -0
- package/grafana/provisioning/datasources/datasources.yml +49 -0
- package/loki/loki-config.yaml +35 -0
- package/otel-collector-config.yml +38 -0
- package/package.json +9 -3
- package/prometheus/prometheus.yml +8 -0
- package/src/_otelLogBuilders.js +11 -5
- package/src/index.d.ts +9 -1
- package/src/index.js +1 -1
- package/tempo/tempo.yml +17 -0
|
@@ -0,0 +1,104 @@
|
|
|
1
|
+
version: "3.9"
|
|
2
|
+
|
|
3
|
+
services:
|
|
4
|
+
otel-collector:
|
|
5
|
+
image: otel/opentelemetry-collector-contrib:0.148.0
|
|
6
|
+
restart: unless-stopped
|
|
7
|
+
command: ["--config", "/etc/otelcol/config.yml"]
|
|
8
|
+
volumes:
|
|
9
|
+
- ./otel-collector-config.yml:/etc/otelcol/config.yml:ro
|
|
10
|
+
ports:
|
|
11
|
+
- "4317:4317" # gRPC
|
|
12
|
+
- "4318:4318" # HTTP (Smithers sends here)
|
|
13
|
+
- "8889:8889" # Prometheus exporter scrape endpoint
|
|
14
|
+
depends_on:
|
|
15
|
+
tempo:
|
|
16
|
+
condition: service_healthy
|
|
17
|
+
loki:
|
|
18
|
+
condition: service_healthy
|
|
19
|
+
healthcheck:
|
|
20
|
+
test: ["CMD", "/otelcol-contrib", "--version"]
|
|
21
|
+
interval: 10s
|
|
22
|
+
timeout: 5s
|
|
23
|
+
retries: 12
|
|
24
|
+
start_period: 10s
|
|
25
|
+
|
|
26
|
+
loki:
|
|
27
|
+
image: grafana/loki:3.3.2
|
|
28
|
+
restart: unless-stopped
|
|
29
|
+
command: ["-config.file=/etc/loki/local-config.yaml"]
|
|
30
|
+
volumes:
|
|
31
|
+
- ./loki/loki-config.yaml:/etc/loki/local-config.yaml:ro
|
|
32
|
+
- ./data/loki:/loki
|
|
33
|
+
ports:
|
|
34
|
+
- "3100:3100"
|
|
35
|
+
healthcheck:
|
|
36
|
+
test: ["CMD", "wget", "-q", "-O-", "http://127.0.0.1:3100/ready"]
|
|
37
|
+
interval: 10s
|
|
38
|
+
timeout: 5s
|
|
39
|
+
retries: 18
|
|
40
|
+
start_period: 10s
|
|
41
|
+
|
|
42
|
+
prometheus:
|
|
43
|
+
image: prom/prometheus:v3.2.0
|
|
44
|
+
restart: unless-stopped
|
|
45
|
+
volumes:
|
|
46
|
+
- ./prometheus/prometheus.yml:/etc/prometheus/prometheus.yml:ro
|
|
47
|
+
- ./data/prometheus:/prometheus
|
|
48
|
+
ports:
|
|
49
|
+
- "9090:9090"
|
|
50
|
+
depends_on:
|
|
51
|
+
otel-collector:
|
|
52
|
+
condition: service_healthy
|
|
53
|
+
healthcheck:
|
|
54
|
+
test: ["CMD", "wget", "-q", "-O-", "http://127.0.0.1:9090/-/ready"]
|
|
55
|
+
interval: 10s
|
|
56
|
+
timeout: 5s
|
|
57
|
+
retries: 18
|
|
58
|
+
start_period: 10s
|
|
59
|
+
|
|
60
|
+
tempo:
|
|
61
|
+
image: grafana/tempo:2.7.0
|
|
62
|
+
restart: unless-stopped
|
|
63
|
+
command: ["-config.file=/etc/tempo/tempo.yml"]
|
|
64
|
+
volumes:
|
|
65
|
+
- ./tempo/tempo.yml:/etc/tempo/tempo.yml:ro
|
|
66
|
+
- ./data/tempo:/tmp/tempo
|
|
67
|
+
ports:
|
|
68
|
+
- "3200:3200" # Tempo query API
|
|
69
|
+
- "4317" # OTLP gRPC (internal)
|
|
70
|
+
healthcheck:
|
|
71
|
+
test: ["CMD", "wget", "-q", "-O-", "http://127.0.0.1:3200/api/search"]
|
|
72
|
+
interval: 10s
|
|
73
|
+
timeout: 5s
|
|
74
|
+
retries: 24
|
|
75
|
+
start_period: 20s
|
|
76
|
+
|
|
77
|
+
grafana:
|
|
78
|
+
image: grafana/grafana:11.5.0
|
|
79
|
+
restart: unless-stopped
|
|
80
|
+
environment:
|
|
81
|
+
# Local dev default is "admin"; override via env for anything shared.
|
|
82
|
+
- GF_SECURITY_ADMIN_PASSWORD=${GF_SECURITY_ADMIN_PASSWORD:-admin}
|
|
83
|
+
# Anonymous browsing stays on for frictionless local dev, but read-only.
|
|
84
|
+
- GF_AUTH_ANONYMOUS_ENABLED=${GF_AUTH_ANONYMOUS_ENABLED:-true}
|
|
85
|
+
- GF_AUTH_ANONYMOUS_ORG_ROLE=${GF_AUTH_ANONYMOUS_ORG_ROLE:-Viewer}
|
|
86
|
+
volumes:
|
|
87
|
+
- ./grafana/provisioning:/etc/grafana/provisioning:ro
|
|
88
|
+
- ./grafana/dashboards:/var/lib/grafana/dashboards:ro
|
|
89
|
+
- ./data/grafana:/var/lib/grafana
|
|
90
|
+
ports:
|
|
91
|
+
- "3001:3000"
|
|
92
|
+
depends_on:
|
|
93
|
+
prometheus:
|
|
94
|
+
condition: service_healthy
|
|
95
|
+
tempo:
|
|
96
|
+
condition: service_healthy
|
|
97
|
+
loki:
|
|
98
|
+
condition: service_healthy
|
|
99
|
+
healthcheck:
|
|
100
|
+
test: ["CMD", "wget", "-q", "-O-", "http://127.0.0.1:3000/api/health"]
|
|
101
|
+
interval: 10s
|
|
102
|
+
timeout: 5s
|
|
103
|
+
retries: 24
|
|
104
|
+
start_period: 20s
|
|
@@ -0,0 +1,453 @@
|
|
|
1
|
+
{
|
|
2
|
+
"uid": "smithers-overview",
|
|
3
|
+
"title": "Smithers Overview",
|
|
4
|
+
"tags": ["smithers", "ai", "orchestration"],
|
|
5
|
+
"timezone": "browser",
|
|
6
|
+
"schemaVersion": 39,
|
|
7
|
+
"version": 1,
|
|
8
|
+
"refresh": "10s",
|
|
9
|
+
"time": { "from": "now-1h", "to": "now" },
|
|
10
|
+
"editable": true,
|
|
11
|
+
"panels": [
|
|
12
|
+
{
|
|
13
|
+
"type": "row",
|
|
14
|
+
"title": "Overview",
|
|
15
|
+
"collapsed": false,
|
|
16
|
+
"gridPos": { "h": 1, "w": 24, "x": 0, "y": 0 },
|
|
17
|
+
"id": 1
|
|
18
|
+
},
|
|
19
|
+
{
|
|
20
|
+
"type": "stat",
|
|
21
|
+
"title": "Active Runs",
|
|
22
|
+
"id": 2,
|
|
23
|
+
"gridPos": { "h": 4, "w": 4, "x": 0, "y": 1 },
|
|
24
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
25
|
+
"targets": [{ "expr": "smithers_smithers_runs_active", "refId": "A" }],
|
|
26
|
+
"fieldConfig": {
|
|
27
|
+
"defaults": {
|
|
28
|
+
"thresholds": {
|
|
29
|
+
"mode": "absolute",
|
|
30
|
+
"steps": [
|
|
31
|
+
{ "color": "green", "value": null },
|
|
32
|
+
{ "color": "yellow", "value": 5 },
|
|
33
|
+
{ "color": "red", "value": 10 }
|
|
34
|
+
]
|
|
35
|
+
},
|
|
36
|
+
"color": { "mode": "thresholds" }
|
|
37
|
+
}
|
|
38
|
+
},
|
|
39
|
+
"options": {
|
|
40
|
+
"reduceOptions": { "calcs": ["lastNotNull"] },
|
|
41
|
+
"colorMode": "background",
|
|
42
|
+
"graphMode": "area"
|
|
43
|
+
}
|
|
44
|
+
},
|
|
45
|
+
{
|
|
46
|
+
"type": "stat",
|
|
47
|
+
"title": "Total Runs",
|
|
48
|
+
"id": 3,
|
|
49
|
+
"gridPos": { "h": 4, "w": 4, "x": 4, "y": 1 },
|
|
50
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
51
|
+
"targets": [{ "expr": "smithers_smithers_runs_total", "refId": "A" }],
|
|
52
|
+
"fieldConfig": { "defaults": { "color": { "mode": "fixed", "fixedColor": "blue" } } },
|
|
53
|
+
"options": {
|
|
54
|
+
"reduceOptions": { "calcs": ["lastNotNull"] },
|
|
55
|
+
"colorMode": "value",
|
|
56
|
+
"graphMode": "area"
|
|
57
|
+
}
|
|
58
|
+
},
|
|
59
|
+
{
|
|
60
|
+
"type": "stat",
|
|
61
|
+
"title": "Active Nodes",
|
|
62
|
+
"id": 4,
|
|
63
|
+
"gridPos": { "h": 4, "w": 4, "x": 8, "y": 1 },
|
|
64
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
65
|
+
"targets": [{ "expr": "smithers_smithers_nodes_active", "refId": "A" }],
|
|
66
|
+
"fieldConfig": {
|
|
67
|
+
"defaults": {
|
|
68
|
+
"thresholds": {
|
|
69
|
+
"mode": "absolute",
|
|
70
|
+
"steps": [
|
|
71
|
+
{ "color": "green", "value": null },
|
|
72
|
+
{ "color": "yellow", "value": 10 },
|
|
73
|
+
{ "color": "red", "value": 20 }
|
|
74
|
+
]
|
|
75
|
+
},
|
|
76
|
+
"color": { "mode": "thresholds" }
|
|
77
|
+
}
|
|
78
|
+
},
|
|
79
|
+
"options": {
|
|
80
|
+
"reduceOptions": { "calcs": ["lastNotNull"] },
|
|
81
|
+
"colorMode": "background",
|
|
82
|
+
"graphMode": "area"
|
|
83
|
+
}
|
|
84
|
+
},
|
|
85
|
+
{
|
|
86
|
+
"type": "stat",
|
|
87
|
+
"title": "Node Success Rate",
|
|
88
|
+
"id": 5,
|
|
89
|
+
"gridPos": { "h": 4, "w": 4, "x": 12, "y": 1 },
|
|
90
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
91
|
+
"targets": [{
|
|
92
|
+
"expr": "(smithers_smithers_nodes_finished / (smithers_smithers_nodes_finished + smithers_smithers_nodes_failed > 0)) * 100 or vector(100)",
|
|
93
|
+
"refId": "A"
|
|
94
|
+
}],
|
|
95
|
+
"fieldConfig": {
|
|
96
|
+
"defaults": {
|
|
97
|
+
"unit": "percent",
|
|
98
|
+
"thresholds": {
|
|
99
|
+
"mode": "absolute",
|
|
100
|
+
"steps": [
|
|
101
|
+
{ "color": "red", "value": null },
|
|
102
|
+
{ "color": "yellow", "value": 90 },
|
|
103
|
+
{ "color": "green", "value": 99 }
|
|
104
|
+
]
|
|
105
|
+
},
|
|
106
|
+
"color": { "mode": "thresholds" },
|
|
107
|
+
"min": 0,
|
|
108
|
+
"max": 100
|
|
109
|
+
}
|
|
110
|
+
},
|
|
111
|
+
"options": {
|
|
112
|
+
"reduceOptions": { "calcs": ["lastNotNull"] },
|
|
113
|
+
"colorMode": "background",
|
|
114
|
+
"graphMode": "none"
|
|
115
|
+
}
|
|
116
|
+
},
|
|
117
|
+
{
|
|
118
|
+
"type": "stat",
|
|
119
|
+
"title": "Cache Hit Rate",
|
|
120
|
+
"id": 6,
|
|
121
|
+
"gridPos": { "h": 4, "w": 4, "x": 16, "y": 1 },
|
|
122
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
123
|
+
"targets": [{
|
|
124
|
+
"expr": "(smithers_smithers_cache_hits / (smithers_smithers_cache_hits + smithers_smithers_cache_misses > 0)) * 100 or vector(0)",
|
|
125
|
+
"refId": "A"
|
|
126
|
+
}],
|
|
127
|
+
"fieldConfig": {
|
|
128
|
+
"defaults": {
|
|
129
|
+
"unit": "percent",
|
|
130
|
+
"thresholds": {
|
|
131
|
+
"mode": "absolute",
|
|
132
|
+
"steps": [
|
|
133
|
+
{ "color": "red", "value": null },
|
|
134
|
+
{ "color": "yellow", "value": 30 },
|
|
135
|
+
{ "color": "green", "value": 70 }
|
|
136
|
+
]
|
|
137
|
+
},
|
|
138
|
+
"color": { "mode": "thresholds" },
|
|
139
|
+
"min": 0,
|
|
140
|
+
"max": 100
|
|
141
|
+
}
|
|
142
|
+
},
|
|
143
|
+
"options": {
|
|
144
|
+
"reduceOptions": { "calcs": ["lastNotNull"] },
|
|
145
|
+
"colorMode": "background",
|
|
146
|
+
"graphMode": "none"
|
|
147
|
+
}
|
|
148
|
+
},
|
|
149
|
+
{
|
|
150
|
+
"type": "stat",
|
|
151
|
+
"title": "DB Retries",
|
|
152
|
+
"id": 7,
|
|
153
|
+
"gridPos": { "h": 4, "w": 4, "x": 20, "y": 1 },
|
|
154
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
155
|
+
"targets": [{ "expr": "smithers_smithers_db_retries", "refId": "A" }],
|
|
156
|
+
"fieldConfig": {
|
|
157
|
+
"defaults": {
|
|
158
|
+
"thresholds": {
|
|
159
|
+
"mode": "absolute",
|
|
160
|
+
"steps": [
|
|
161
|
+
{ "color": "green", "value": null },
|
|
162
|
+
{ "color": "yellow", "value": 5 },
|
|
163
|
+
{ "color": "red", "value": 20 }
|
|
164
|
+
]
|
|
165
|
+
},
|
|
166
|
+
"color": { "mode": "thresholds" }
|
|
167
|
+
}
|
|
168
|
+
},
|
|
169
|
+
"options": {
|
|
170
|
+
"reduceOptions": { "calcs": ["lastNotNull"] },
|
|
171
|
+
"colorMode": "value",
|
|
172
|
+
"graphMode": "area"
|
|
173
|
+
}
|
|
174
|
+
},
|
|
175
|
+
|
|
176
|
+
{
|
|
177
|
+
"type": "row",
|
|
178
|
+
"title": "Runs & Nodes",
|
|
179
|
+
"collapsed": false,
|
|
180
|
+
"gridPos": { "h": 1, "w": 24, "x": 0, "y": 5 },
|
|
181
|
+
"id": 10
|
|
182
|
+
},
|
|
183
|
+
{
|
|
184
|
+
"type": "timeseries",
|
|
185
|
+
"title": "Node Throughput",
|
|
186
|
+
"id": 11,
|
|
187
|
+
"gridPos": { "h": 8, "w": 12, "x": 0, "y": 6 },
|
|
188
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
189
|
+
"targets": [
|
|
190
|
+
{ "expr": "rate(smithers_smithers_nodes_started[5m])", "legendFormat": "Started", "refId": "A" },
|
|
191
|
+
{ "expr": "rate(smithers_smithers_nodes_finished[5m])", "legendFormat": "Finished", "refId": "B" },
|
|
192
|
+
{ "expr": "rate(smithers_smithers_nodes_failed[5m])", "legendFormat": "Failed", "refId": "C" }
|
|
193
|
+
],
|
|
194
|
+
"fieldConfig": {
|
|
195
|
+
"defaults": {
|
|
196
|
+
"custom": { "drawStyle": "line", "lineWidth": 2, "fillOpacity": 10, "gradientMode": "scheme" },
|
|
197
|
+
"color": { "mode": "palette-classic" },
|
|
198
|
+
"unit": "ops"
|
|
199
|
+
}
|
|
200
|
+
},
|
|
201
|
+
"options": {
|
|
202
|
+
"tooltip": { "mode": "multi", "sort": "desc" },
|
|
203
|
+
"legend": { "displayMode": "list", "placement": "bottom", "calcs": ["mean", "max"] }
|
|
204
|
+
}
|
|
205
|
+
},
|
|
206
|
+
{
|
|
207
|
+
"type": "heatmap",
|
|
208
|
+
"title": "Node Duration",
|
|
209
|
+
"id": 12,
|
|
210
|
+
"gridPos": { "h": 8, "w": 12, "x": 12, "y": 6 },
|
|
211
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
212
|
+
"targets": [{
|
|
213
|
+
"expr": "sum(increase(smithers_smithers_node_duration_ms_bucket[5m])) by (le)",
|
|
214
|
+
"legendFormat": "{{le}}",
|
|
215
|
+
"refId": "A",
|
|
216
|
+
"format": "heatmap"
|
|
217
|
+
}],
|
|
218
|
+
"options": {
|
|
219
|
+
"calculate": false,
|
|
220
|
+
"yAxis": { "unit": "ms" },
|
|
221
|
+
"color": { "mode": "scheme", "scheme": "Oranges", "steps": 64 },
|
|
222
|
+
"cellGap": 1
|
|
223
|
+
}
|
|
224
|
+
},
|
|
225
|
+
|
|
226
|
+
{
|
|
227
|
+
"type": "row",
|
|
228
|
+
"title": "Tools",
|
|
229
|
+
"collapsed": false,
|
|
230
|
+
"gridPos": { "h": 1, "w": 24, "x": 0, "y": 14 },
|
|
231
|
+
"id": 20
|
|
232
|
+
},
|
|
233
|
+
{
|
|
234
|
+
"type": "timeseries",
|
|
235
|
+
"title": "Tool Call Rate",
|
|
236
|
+
"id": 21,
|
|
237
|
+
"gridPos": { "h": 8, "w": 12, "x": 0, "y": 15 },
|
|
238
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
239
|
+
"targets": [{
|
|
240
|
+
"expr": "rate(smithers_smithers_tool_calls_total[5m])",
|
|
241
|
+
"legendFormat": "Tool Calls / sec",
|
|
242
|
+
"refId": "A"
|
|
243
|
+
}],
|
|
244
|
+
"fieldConfig": {
|
|
245
|
+
"defaults": {
|
|
246
|
+
"custom": { "drawStyle": "line", "lineWidth": 2, "fillOpacity": 10 },
|
|
247
|
+
"color": { "mode": "fixed", "fixedColor": "purple" },
|
|
248
|
+
"unit": "ops"
|
|
249
|
+
}
|
|
250
|
+
},
|
|
251
|
+
"options": {
|
|
252
|
+
"tooltip": { "mode": "single" },
|
|
253
|
+
"legend": { "displayMode": "list", "placement": "bottom" }
|
|
254
|
+
}
|
|
255
|
+
},
|
|
256
|
+
{
|
|
257
|
+
"type": "heatmap",
|
|
258
|
+
"title": "Tool Duration",
|
|
259
|
+
"id": 22,
|
|
260
|
+
"gridPos": { "h": 8, "w": 12, "x": 12, "y": 15 },
|
|
261
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
262
|
+
"targets": [{
|
|
263
|
+
"expr": "sum(increase(smithers_smithers_tool_duration_ms_bucket[5m])) by (le)",
|
|
264
|
+
"legendFormat": "{{le}}",
|
|
265
|
+
"refId": "A",
|
|
266
|
+
"format": "heatmap"
|
|
267
|
+
}],
|
|
268
|
+
"options": {
|
|
269
|
+
"calculate": false,
|
|
270
|
+
"yAxis": { "unit": "ms" },
|
|
271
|
+
"color": { "mode": "scheme", "scheme": "Blues", "steps": 64 },
|
|
272
|
+
"cellGap": 1
|
|
273
|
+
}
|
|
274
|
+
},
|
|
275
|
+
|
|
276
|
+
{
|
|
277
|
+
"type": "row",
|
|
278
|
+
"title": "Infrastructure",
|
|
279
|
+
"collapsed": false,
|
|
280
|
+
"gridPos": { "h": 1, "w": 24, "x": 0, "y": 23 },
|
|
281
|
+
"id": 30
|
|
282
|
+
},
|
|
283
|
+
{
|
|
284
|
+
"type": "timeseries",
|
|
285
|
+
"title": "HTTP Request Latency",
|
|
286
|
+
"id": 31,
|
|
287
|
+
"gridPos": { "h": 8, "w": 8, "x": 0, "y": 24 },
|
|
288
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
289
|
+
"targets": [
|
|
290
|
+
{
|
|
291
|
+
"expr": "histogram_quantile(0.50, sum(rate(smithers_smithers_http_request_duration_ms_bucket[5m])) by (le))",
|
|
292
|
+
"legendFormat": "p50",
|
|
293
|
+
"refId": "A"
|
|
294
|
+
},
|
|
295
|
+
{
|
|
296
|
+
"expr": "histogram_quantile(0.95, sum(rate(smithers_smithers_http_request_duration_ms_bucket[5m])) by (le))",
|
|
297
|
+
"legendFormat": "p95",
|
|
298
|
+
"refId": "B"
|
|
299
|
+
},
|
|
300
|
+
{
|
|
301
|
+
"expr": "histogram_quantile(0.99, sum(rate(smithers_smithers_http_request_duration_ms_bucket[5m])) by (le))",
|
|
302
|
+
"legendFormat": "p99",
|
|
303
|
+
"refId": "C"
|
|
304
|
+
}
|
|
305
|
+
],
|
|
306
|
+
"fieldConfig": {
|
|
307
|
+
"defaults": {
|
|
308
|
+
"custom": { "drawStyle": "line", "lineWidth": 2, "fillOpacity": 5 },
|
|
309
|
+
"color": { "mode": "palette-classic" },
|
|
310
|
+
"unit": "ms"
|
|
311
|
+
}
|
|
312
|
+
},
|
|
313
|
+
"options": {
|
|
314
|
+
"tooltip": { "mode": "multi", "sort": "desc" },
|
|
315
|
+
"legend": { "displayMode": "list", "placement": "bottom" }
|
|
316
|
+
}
|
|
317
|
+
},
|
|
318
|
+
{
|
|
319
|
+
"type": "heatmap",
|
|
320
|
+
"title": "DB Query Duration",
|
|
321
|
+
"id": 32,
|
|
322
|
+
"gridPos": { "h": 8, "w": 8, "x": 8, "y": 24 },
|
|
323
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
324
|
+
"targets": [{
|
|
325
|
+
"expr": "sum(increase(smithers_smithers_db_query_ms_bucket[5m])) by (le)",
|
|
326
|
+
"legendFormat": "{{le}}",
|
|
327
|
+
"refId": "A",
|
|
328
|
+
"format": "heatmap"
|
|
329
|
+
}],
|
|
330
|
+
"options": {
|
|
331
|
+
"calculate": false,
|
|
332
|
+
"yAxis": { "unit": "ms" },
|
|
333
|
+
"color": { "mode": "scheme", "scheme": "Greens", "steps": 64 },
|
|
334
|
+
"cellGap": 1
|
|
335
|
+
}
|
|
336
|
+
},
|
|
337
|
+
{
|
|
338
|
+
"type": "timeseries",
|
|
339
|
+
"title": "VCS & Hot Reload",
|
|
340
|
+
"id": 33,
|
|
341
|
+
"gridPos": { "h": 8, "w": 8, "x": 16, "y": 24 },
|
|
342
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
343
|
+
"targets": [
|
|
344
|
+
{
|
|
345
|
+
"expr": "rate(smithers_smithers_hot_reloads[5m])",
|
|
346
|
+
"legendFormat": "Hot Reloads / sec",
|
|
347
|
+
"refId": "A"
|
|
348
|
+
},
|
|
349
|
+
{
|
|
350
|
+
"expr": "rate(smithers_smithers_hot_reload_failures[5m])",
|
|
351
|
+
"legendFormat": "Reload Failures / sec",
|
|
352
|
+
"refId": "B"
|
|
353
|
+
},
|
|
354
|
+
{
|
|
355
|
+
"expr": "histogram_quantile(0.95, sum(rate(smithers_smithers_vcs_duration_ms_bucket[5m])) by (le))",
|
|
356
|
+
"legendFormat": "VCS p95 (ms)",
|
|
357
|
+
"refId": "C"
|
|
358
|
+
}
|
|
359
|
+
],
|
|
360
|
+
"fieldConfig": {
|
|
361
|
+
"defaults": {
|
|
362
|
+
"custom": { "drawStyle": "line", "lineWidth": 2, "fillOpacity": 10 },
|
|
363
|
+
"color": { "mode": "palette-classic" }
|
|
364
|
+
}
|
|
365
|
+
},
|
|
366
|
+
"options": {
|
|
367
|
+
"tooltip": { "mode": "multi", "sort": "desc" },
|
|
368
|
+
"legend": { "displayMode": "list", "placement": "bottom" }
|
|
369
|
+
}
|
|
370
|
+
},
|
|
371
|
+
|
|
372
|
+
{
|
|
373
|
+
"type": "row",
|
|
374
|
+
"title": "Approvals",
|
|
375
|
+
"collapsed": false,
|
|
376
|
+
"gridPos": { "h": 1, "w": 24, "x": 0, "y": 32 },
|
|
377
|
+
"id": 40
|
|
378
|
+
},
|
|
379
|
+
{
|
|
380
|
+
"type": "timeseries",
|
|
381
|
+
"title": "Approval Flow",
|
|
382
|
+
"id": 41,
|
|
383
|
+
"gridPos": { "h": 8, "w": 12, "x": 0, "y": 33 },
|
|
384
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
385
|
+
"targets": [
|
|
386
|
+
{ "expr": "rate(smithers_smithers_approvals_requested[5m])", "legendFormat": "Requested", "refId": "A" },
|
|
387
|
+
{ "expr": "rate(smithers_smithers_approvals_granted[5m])", "legendFormat": "Granted", "refId": "B" },
|
|
388
|
+
{ "expr": "rate(smithers_smithers_approvals_denied[5m])", "legendFormat": "Denied", "refId": "C" }
|
|
389
|
+
],
|
|
390
|
+
"fieldConfig": {
|
|
391
|
+
"defaults": {
|
|
392
|
+
"custom": { "drawStyle": "bars", "lineWidth": 1, "fillOpacity": 50 },
|
|
393
|
+
"color": { "mode": "palette-classic" },
|
|
394
|
+
"unit": "ops"
|
|
395
|
+
}
|
|
396
|
+
},
|
|
397
|
+
"options": {
|
|
398
|
+
"tooltip": { "mode": "multi" },
|
|
399
|
+
"legend": { "displayMode": "list", "placement": "bottom" }
|
|
400
|
+
}
|
|
401
|
+
},
|
|
402
|
+
{
|
|
403
|
+
"type": "timeseries",
|
|
404
|
+
"title": "HTTP Request Rate",
|
|
405
|
+
"id": 42,
|
|
406
|
+
"gridPos": { "h": 8, "w": 12, "x": 12, "y": 33 },
|
|
407
|
+
"datasource": { "type": "prometheus", "uid": "prometheus" },
|
|
408
|
+
"targets": [{
|
|
409
|
+
"expr": "rate(smithers_smithers_http_requests[5m])",
|
|
410
|
+
"legendFormat": "Requests / sec",
|
|
411
|
+
"refId": "A"
|
|
412
|
+
}],
|
|
413
|
+
"fieldConfig": {
|
|
414
|
+
"defaults": {
|
|
415
|
+
"custom": { "drawStyle": "line", "lineWidth": 2, "fillOpacity": 15, "gradientMode": "scheme" },
|
|
416
|
+
"color": { "mode": "fixed", "fixedColor": "green" },
|
|
417
|
+
"unit": "reqps"
|
|
418
|
+
}
|
|
419
|
+
},
|
|
420
|
+
"options": {
|
|
421
|
+
"tooltip": { "mode": "single" },
|
|
422
|
+
"legend": { "displayMode": "list", "placement": "bottom" }
|
|
423
|
+
}
|
|
424
|
+
},
|
|
425
|
+
|
|
426
|
+
{
|
|
427
|
+
"type": "row",
|
|
428
|
+
"title": "Traces",
|
|
429
|
+
"collapsed": true,
|
|
430
|
+
"gridPos": { "h": 1, "w": 24, "x": 0, "y": 41 },
|
|
431
|
+
"id": 50,
|
|
432
|
+
"panels": [
|
|
433
|
+
{
|
|
434
|
+
"type": "traces",
|
|
435
|
+
"title": "Recent Traces",
|
|
436
|
+
"id": 51,
|
|
437
|
+
"gridPos": { "h": 12, "w": 24, "x": 0, "y": 42 },
|
|
438
|
+
"datasource": { "type": "tempo", "uid": "tempo" },
|
|
439
|
+
"targets": [{
|
|
440
|
+
"datasource": { "type": "tempo", "uid": "tempo" },
|
|
441
|
+
"queryType": "traceqlSearch",
|
|
442
|
+
"limit": 20,
|
|
443
|
+
"tableType": "traces",
|
|
444
|
+
"filters": [
|
|
445
|
+
{ "id": "service-name", "tag": "service.name", "operator": "=", "value": ["smithers"], "scope": "resource" }
|
|
446
|
+
],
|
|
447
|
+
"refId": "A"
|
|
448
|
+
}]
|
|
449
|
+
}
|
|
450
|
+
]
|
|
451
|
+
}
|
|
452
|
+
]
|
|
453
|
+
}
|
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
apiVersion: 1
|
|
2
|
+
|
|
3
|
+
datasources:
|
|
4
|
+
- name: Prometheus
|
|
5
|
+
type: prometheus
|
|
6
|
+
uid: prometheus
|
|
7
|
+
access: proxy
|
|
8
|
+
url: http://prometheus:9090
|
|
9
|
+
isDefault: true
|
|
10
|
+
editable: false
|
|
11
|
+
jsonData:
|
|
12
|
+
httpMethod: POST
|
|
13
|
+
exemplarTraceIdDestinations:
|
|
14
|
+
- name: traceID
|
|
15
|
+
datasourceUid: tempo
|
|
16
|
+
|
|
17
|
+
- name: Loki
|
|
18
|
+
type: loki
|
|
19
|
+
uid: loki
|
|
20
|
+
access: proxy
|
|
21
|
+
url: http://loki:3100
|
|
22
|
+
editable: false
|
|
23
|
+
jsonData:
|
|
24
|
+
maxLines: 1000
|
|
25
|
+
|
|
26
|
+
- name: Tempo
|
|
27
|
+
type: tempo
|
|
28
|
+
uid: tempo
|
|
29
|
+
access: proxy
|
|
30
|
+
url: http://tempo:3200
|
|
31
|
+
editable: false
|
|
32
|
+
jsonData:
|
|
33
|
+
tracesToLogs:
|
|
34
|
+
datasourceUid: loki
|
|
35
|
+
tags: ["service.name"]
|
|
36
|
+
mappedTags:
|
|
37
|
+
- key: service.name
|
|
38
|
+
value: service.name
|
|
39
|
+
mapTagNamesEnabled: false
|
|
40
|
+
spanStartTimeShift: "-1h"
|
|
41
|
+
spanEndTimeShift: "1h"
|
|
42
|
+
filterByTraceID: false
|
|
43
|
+
filterBySpanID: false
|
|
44
|
+
tracesToMetrics:
|
|
45
|
+
datasourceUid: prometheus
|
|
46
|
+
spanStartTimeShift: "-1h"
|
|
47
|
+
spanEndTimeShift: "1h"
|
|
48
|
+
nodeGraph:
|
|
49
|
+
enabled: true
|
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
auth_enabled: false
|
|
2
|
+
|
|
3
|
+
server:
|
|
4
|
+
http_listen_port: 3100
|
|
5
|
+
|
|
6
|
+
common:
|
|
7
|
+
instance_addr: 127.0.0.1
|
|
8
|
+
path_prefix: /loki
|
|
9
|
+
storage:
|
|
10
|
+
filesystem:
|
|
11
|
+
chunks_directory: /loki/chunks
|
|
12
|
+
rules_directory: /loki/rules
|
|
13
|
+
replication_factor: 1
|
|
14
|
+
ring:
|
|
15
|
+
kvstore:
|
|
16
|
+
store: inmemory
|
|
17
|
+
|
|
18
|
+
schema_config:
|
|
19
|
+
configs:
|
|
20
|
+
- from: 2020-10-24
|
|
21
|
+
store: tsdb
|
|
22
|
+
object_store: filesystem
|
|
23
|
+
schema: v13
|
|
24
|
+
index:
|
|
25
|
+
prefix: index_
|
|
26
|
+
period: 24h
|
|
27
|
+
|
|
28
|
+
limits_config:
|
|
29
|
+
allow_structured_metadata: true
|
|
30
|
+
|
|
31
|
+
ruler:
|
|
32
|
+
alertmanager_url: http://localhost:9093
|
|
33
|
+
|
|
34
|
+
analytics:
|
|
35
|
+
reporting_enabled: false
|
|
@@ -0,0 +1,38 @@
|
|
|
1
|
+
receivers:
|
|
2
|
+
otlp:
|
|
3
|
+
protocols:
|
|
4
|
+
grpc:
|
|
5
|
+
endpoint: "0.0.0.0:4317"
|
|
6
|
+
http:
|
|
7
|
+
endpoint: "0.0.0.0:4318"
|
|
8
|
+
|
|
9
|
+
exporters:
|
|
10
|
+
prometheus:
|
|
11
|
+
endpoint: "0.0.0.0:8889"
|
|
12
|
+
namespace: smithers
|
|
13
|
+
otlp_grpc/tempo:
|
|
14
|
+
endpoint: "tempo:4317"
|
|
15
|
+
tls:
|
|
16
|
+
insecure: true
|
|
17
|
+
otlp_http/loki:
|
|
18
|
+
endpoint: "http://loki:3100/otlp"
|
|
19
|
+
|
|
20
|
+
processors:
|
|
21
|
+
batch:
|
|
22
|
+
timeout: 5s
|
|
23
|
+
send_batch_size: 1024
|
|
24
|
+
|
|
25
|
+
service:
|
|
26
|
+
pipelines:
|
|
27
|
+
metrics:
|
|
28
|
+
receivers: [otlp]
|
|
29
|
+
processors: [batch]
|
|
30
|
+
exporters: [prometheus]
|
|
31
|
+
traces:
|
|
32
|
+
receivers: [otlp]
|
|
33
|
+
processors: [batch]
|
|
34
|
+
exporters: [otlp_grpc/tempo]
|
|
35
|
+
logs:
|
|
36
|
+
receivers: [otlp]
|
|
37
|
+
processors: [batch]
|
|
38
|
+
exporters: [otlp_http/loki]
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@smithers-orchestrator/observability",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.24.0",
|
|
4
4
|
"description": "Concrete Smithers metrics, logging, tracing, and observability integrations",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"sideEffects": false,
|
|
@@ -22,14 +22,20 @@
|
|
|
22
22
|
}
|
|
23
23
|
},
|
|
24
24
|
"files": [
|
|
25
|
-
"src/"
|
|
25
|
+
"src/",
|
|
26
|
+
"docker-compose.otel.yml",
|
|
27
|
+
"otel-collector-config.yml",
|
|
28
|
+
"prometheus/",
|
|
29
|
+
"tempo/",
|
|
30
|
+
"loki/",
|
|
31
|
+
"grafana/"
|
|
26
32
|
],
|
|
27
33
|
"dependencies": {
|
|
28
34
|
"@effect/opentelemetry": "^0.63.0",
|
|
29
35
|
"@effect/platform": "^0.96.0",
|
|
30
36
|
"@effect/platform-bun": "^0.89.0",
|
|
31
37
|
"effect": "^3.21.1",
|
|
32
|
-
"@smithers-orchestrator/agents": "0.
|
|
38
|
+
"@smithers-orchestrator/agents": "0.24.0"
|
|
33
39
|
},
|
|
34
40
|
"devDependencies": {
|
|
35
41
|
"@types/bun": "latest",
|
package/src/_otelLogBuilders.js
CHANGED
|
@@ -58,11 +58,17 @@ export function buildOtelLogRecord(body, attributes, severity) {
|
|
|
58
58
|
* @returns {OtelLogSeverity}
|
|
59
59
|
*/
|
|
60
60
|
export function inferCanonicalSeverity(event) {
|
|
61
|
-
|
|
62
|
-
|
|
63
|
-
|
|
64
|
-
|
|
65
|
-
|
|
61
|
+
if (event.event.kind === "capture.error") {
|
|
62
|
+
// Truncated output is a non-fatal capture issue, not an error
|
|
63
|
+
if (event.payload?.reason === "truncated-json-stream") {
|
|
64
|
+
return "WARN";
|
|
65
|
+
}
|
|
66
|
+
return "ERROR";
|
|
67
|
+
}
|
|
68
|
+
if (event.event.kind === "capture.warning" || event.event.kind === "stderr") {
|
|
69
|
+
return "WARN";
|
|
70
|
+
}
|
|
71
|
+
return "INFO";
|
|
66
72
|
}
|
|
67
73
|
|
|
68
74
|
/**
|
package/src/index.d.ts
CHANGED
|
@@ -838,6 +838,12 @@ declare const toolOutputTruncatedTotal: Metric.Metric.Counter<number>;
|
|
|
838
838
|
|
|
839
839
|
declare const eventsEmittedTotal: Metric.Metric.Counter<number>;
|
|
840
840
|
|
|
841
|
+
declare const snapshotsCaptured: Metric.Metric.Counter<number>;
|
|
842
|
+
|
|
843
|
+
declare const runForksCreated: Metric.Metric.Counter<number>;
|
|
844
|
+
|
|
845
|
+
declare const replaysStarted: Metric.Metric.Counter<number>;
|
|
846
|
+
|
|
841
847
|
declare const activeRuns: Metric.Metric.Gauge<number>;
|
|
842
848
|
|
|
843
849
|
declare const activeNodes: Metric.Metric.Gauge<number>;
|
|
@@ -906,6 +912,8 @@ declare const sandboxTransportDurationMs: Metric.Metric<effect_MetricKeyType.Met
|
|
|
906
912
|
|
|
907
913
|
declare const sandboxPatchCount: Metric.Metric<effect_MetricKeyType.MetricKeyType.Histogram, number, effect_MetricState.MetricState.Histogram>;
|
|
908
914
|
|
|
915
|
+
declare const snapshotDuration: Metric.Metric<effect_MetricKeyType.MetricKeyType.Histogram, number, effect_MetricState.MetricState.Histogram>;
|
|
916
|
+
|
|
909
917
|
declare const rewindTotal: Metric.Metric.Counter<number>;
|
|
910
918
|
|
|
911
919
|
declare const rewindRollbackTotal: Metric.Metric.Counter<number>;
|
|
@@ -1031,4 +1039,4 @@ type SmithersMetricDefinition = SmithersMetricDefinition$2;
|
|
|
1031
1039
|
type SmithersObservabilityOptions = SmithersObservabilityOptions$4;
|
|
1032
1040
|
type SmithersObservabilityService = SmithersObservabilityService$1;
|
|
1033
1041
|
|
|
1034
|
-
export { type CorrelationContext, CorrelationContextLive, type CorrelationContextPatch, CorrelationContextService, type CorrelationPatch, type MetricLabels, MetricsService, MetricsServiceLive, type MetricsServiceShape, type MetricsSnapshot, type ResolvedSmithersObservabilityOptions, type SmithersEvent, type SmithersLogFormat, type SmithersMetricDefinition, SmithersObservability, type SmithersObservabilityOptions, type SmithersObservabilityService, TracingService, TracingServiceLive, activeNodes, activeRuns, annotateSmithersTrace, approvalPending, approvalWaitDuration, approvalsDenied, approvalsGranted, approvalsRequested, attemptDuration, cacheHits, cacheMisses, correlationContextFiberRef, correlationContextToLogAnnotations, createSmithersObservabilityLayer, createSmithersOtelLayer, createSmithersRuntimeLayer, dbQueryDuration, dbRetries, dbTransactionDuration, dbTransactionRetries, dbTransactionRollbacks, errorsTotal, eventsEmittedTotal, externalWaitAsyncPending, getCurrentCorrelationContext, getCurrentCorrelationContextEffect, getCurrentSmithersTraceAnnotations, getCurrentSmithersTraceSpan, hotReloadDuration, hotReloadFailures, hotReloads, httpRequestDuration, httpRequests, logDebug, logError, logInfo, logWarning, makeSmithersSpanAttributes, mergeCorrelationContext, metricsServiceAdapter, nodeDuration, nodeRetriesTotal, nodesFailed, nodesFinished, nodesStarted, processHeapUsedBytes, processMemoryRssBytes, processUptimeSeconds, prometheusContentType, promptSizeBytes, renderPrometheusMetrics, resolveSmithersObservabilityOptions, responseSizeBytes, rewindDurationMs, rewindFramesDeleted, rewindRollbackTotal, rewindSandboxesReverted, rewindTotal, runDuration, runWithCorrelationContext, runsAncestryDepth, runsCancelledTotal, runsCarriedStateBytes, runsContinuedTotal, runsFailedTotal, runsFinishedTotal, runsResumedTotal, runsTotal, sandboxActive, sandboxBundleSizeBytes, sandboxCompletedTotal, sandboxCreatedTotal, sandboxDurationMs, sandboxPatchCount, sandboxTransportDurationMs, schedulerConcurrencyUtilization, schedulerQueueDepth, schedulerWaitDuration, scorerEventsFailed, scorerEventsFinished, scorerEventsStarted, smithersMetricCatalog, smithersMetrics, smithersSpanNames, timerDelayDuration, timersCancelled, timersCreated, timersFired, timersPending, toPrometheusMetricName, tokensCacheReadTotal, tokensCacheWriteTotal, tokensContextWindowBucketTotal, tokensContextWindowPerCall, tokensInputPerCall, tokensInputTotal, tokensOutputPerCall, tokensOutputTotal, tokensReasoningTotal, toolCallErrorsTotal, toolCallsTotal, toolDuration, toolOutputTruncatedTotal, trackEvent as trackSmithersEvent, updateCurrentCorrelationContext, updateProcessMetrics, vcsDuration, withCorrelationContext, withCurrentCorrelationContext, withSmithersSpan };
|
|
1042
|
+
export { type CorrelationContext, CorrelationContextLive, type CorrelationContextPatch, CorrelationContextService, type CorrelationPatch, type MetricLabels, MetricsService, MetricsServiceLive, type MetricsServiceShape, type MetricsSnapshot, type ResolvedSmithersObservabilityOptions, type SmithersEvent, type SmithersLogFormat, type SmithersMetricDefinition, SmithersObservability, type SmithersObservabilityOptions, type SmithersObservabilityService, TracingService, TracingServiceLive, activeNodes, activeRuns, annotateSmithersTrace, approvalPending, approvalWaitDuration, approvalsDenied, approvalsGranted, approvalsRequested, attemptDuration, cacheHits, cacheMisses, correlationContextFiberRef, correlationContextToLogAnnotations, createSmithersObservabilityLayer, createSmithersOtelLayer, createSmithersRuntimeLayer, dbQueryDuration, dbRetries, dbTransactionDuration, dbTransactionRetries, dbTransactionRollbacks, errorsTotal, eventsEmittedTotal, externalWaitAsyncPending, getCurrentCorrelationContext, getCurrentCorrelationContextEffect, getCurrentSmithersTraceAnnotations, getCurrentSmithersTraceSpan, hotReloadDuration, hotReloadFailures, hotReloads, httpRequestDuration, httpRequests, logDebug, logError, logInfo, logWarning, makeSmithersSpanAttributes, mergeCorrelationContext, metricsServiceAdapter, nodeDuration, nodeRetriesTotal, nodesFailed, nodesFinished, nodesStarted, processHeapUsedBytes, processMemoryRssBytes, processUptimeSeconds, prometheusContentType, promptSizeBytes, renderPrometheusMetrics, replaysStarted, resolveSmithersObservabilityOptions, responseSizeBytes, rewindDurationMs, rewindFramesDeleted, rewindRollbackTotal, rewindSandboxesReverted, rewindTotal, runDuration, runForksCreated, runWithCorrelationContext, runsAncestryDepth, runsCancelledTotal, runsCarriedStateBytes, runsContinuedTotal, runsFailedTotal, runsFinishedTotal, runsResumedTotal, runsTotal, sandboxActive, sandboxBundleSizeBytes, sandboxCompletedTotal, sandboxCreatedTotal, sandboxDurationMs, sandboxPatchCount, sandboxTransportDurationMs, schedulerConcurrencyUtilization, schedulerQueueDepth, schedulerWaitDuration, scorerEventsFailed, scorerEventsFinished, scorerEventsStarted, smithersMetricCatalog, smithersMetrics, smithersSpanNames, snapshotDuration, snapshotsCaptured, timerDelayDuration, timersCancelled, timersCreated, timersFired, timersPending, toPrometheusMetricName, tokensCacheReadTotal, tokensCacheWriteTotal, tokensContextWindowBucketTotal, tokensContextWindowPerCall, tokensInputPerCall, tokensInputTotal, tokensOutputPerCall, tokensOutputTotal, tokensReasoningTotal, toolCallErrorsTotal, toolCallsTotal, toolDuration, toolOutputTruncatedTotal, trackEvent as trackSmithersEvent, updateCurrentCorrelationContext, updateProcessMetrics, vcsDuration, withCorrelationContext, withCurrentCorrelationContext, withSmithersSpan };
|
package/src/index.js
CHANGED
|
@@ -29,7 +29,7 @@ export { createSmithersOtelLayer } from "./createSmithersOtelLayer.js";
|
|
|
29
29
|
export { createSmithersObservabilityLayer } from "./createSmithersObservabilityLayer.js";
|
|
30
30
|
export { createSmithersRuntimeLayer } from "./createSmithersRuntimeLayer.js";
|
|
31
31
|
export { rewindTotal, rewindRollbackTotal, rewindDurationMs, rewindFramesDeleted, rewindSandboxesReverted, } from "./metrics/index.js";
|
|
32
|
-
export { activeNodes, activeRuns, approvalPending, externalWaitAsyncPending, approvalsDenied, approvalsGranted, approvalsRequested, approvalWaitDuration, timerDelayDuration, timersCancelled, timersCreated, timersFired, timersPending, attemptDuration, cacheHits, cacheMisses, dbQueryDuration, dbRetries, dbTransactionDuration, dbTransactionRetries, dbTransactionRollbacks, errorsTotal, eventsEmittedTotal, hotReloadDuration, hotReloadFailures, hotReloads, httpRequestDuration, httpRequests, nodeDuration, nodeRetriesTotal, nodesFailed, nodesFinished, nodesStarted, processHeapUsedBytes, processMemoryRssBytes, processUptimeSeconds, promptSizeBytes, responseSizeBytes, runDuration, runsCancelledTotal, runsContinuedTotal, runsFailedTotal, runsFinishedTotal, runsResumedTotal, runsAncestryDepth, runsCarriedStateBytes, sandboxActive, sandboxBundleSizeBytes, sandboxCompletedTotal, sandboxCreatedTotal, sandboxDurationMs, sandboxPatchCount, sandboxTransportDurationMs, runsTotal, schedulerConcurrencyUtilization, schedulerQueueDepth, schedulerWaitDuration, tokensCacheReadTotal, tokensCacheWriteTotal, tokensContextWindowBucketTotal, tokensContextWindowPerCall, tokensInputPerCall, tokensInputTotal, tokensOutputPerCall, tokensOutputTotal, tokensReasoningTotal, toolCallErrorsTotal, toolCallsTotal, toolDuration, toolOutputTruncatedTotal, scorerEventsStarted, scorerEventsFinished, scorerEventsFailed, trackEvent as trackSmithersEvent, updateProcessMetrics, vcsDuration, toPrometheusMetricName, smithersMetricCatalog, metricsServiceAdapter, } from "./metrics/index.js";
|
|
32
|
+
export { activeNodes, activeRuns, approvalPending, externalWaitAsyncPending, approvalsDenied, approvalsGranted, approvalsRequested, approvalWaitDuration, timerDelayDuration, timersCancelled, timersCreated, timersFired, timersPending, attemptDuration, cacheHits, cacheMisses, dbQueryDuration, dbRetries, dbTransactionDuration, dbTransactionRetries, dbTransactionRollbacks, errorsTotal, eventsEmittedTotal, hotReloadDuration, hotReloadFailures, hotReloads, httpRequestDuration, httpRequests, nodeDuration, nodeRetriesTotal, nodesFailed, nodesFinished, nodesStarted, processHeapUsedBytes, processMemoryRssBytes, processUptimeSeconds, promptSizeBytes, responseSizeBytes, runDuration, runsCancelledTotal, runsContinuedTotal, runsFailedTotal, runsFinishedTotal, runsResumedTotal, runsAncestryDepth, runsCarriedStateBytes, sandboxActive, sandboxBundleSizeBytes, sandboxCompletedTotal, sandboxCreatedTotal, sandboxDurationMs, sandboxPatchCount, sandboxTransportDurationMs, runsTotal, schedulerConcurrencyUtilization, schedulerQueueDepth, schedulerWaitDuration, tokensCacheReadTotal, tokensCacheWriteTotal, tokensContextWindowBucketTotal, tokensContextWindowPerCall, tokensInputPerCall, tokensInputTotal, tokensOutputPerCall, tokensOutputTotal, tokensReasoningTotal, toolCallErrorsTotal, toolCallsTotal, toolDuration, toolOutputTruncatedTotal, scorerEventsStarted, scorerEventsFinished, scorerEventsFailed, snapshotsCaptured, runForksCreated, replaysStarted, snapshotDuration, trackEvent as trackSmithersEvent, updateProcessMetrics, vcsDuration, toPrometheusMetricName, smithersMetricCatalog, metricsServiceAdapter, } from "./metrics/index.js";
|
|
33
33
|
export { correlationContextFiberRef, correlationContextToLogAnnotations, CorrelationContextLive, CorrelationContextService, getCurrentCorrelationContext, getCurrentCorrelationContextEffect, mergeCorrelationContext, runWithCorrelationContext, withCorrelationContext, withCurrentCorrelationContext, } from "./correlation.js";
|
|
34
34
|
export { updateCurrentCorrelationContext } from "./correlation.js";
|
|
35
35
|
export { logDebug, logInfo, logWarning, logError, } from "./logging.js";
|
package/tempo/tempo.yml
ADDED