feat(telemetry): add portable local telemetry plane

2026-07-28 00:54:11 +03:00
parent 9fc002d86f
commit b8a008b50e
16 changed files with 1486 additions and 0 deletions
@@ -0,0 +1,57 @@
+[agent]
+  interval = "2s"
+  round_interval = true
+  omit_hostname = false
+
+[global_tags]
+  node_id = "${MISSIONCORE_NODE_ID}"
+  contour_id = "${MISSIONCORE_CONTOUR_ID}"
+  agent_id = "${MISSIONCORE_AGENT_ID}"
+
+[[inputs.cpu]]
+  percpu = false
+  totalcpu = true
+  report_active = true
+
+[[inputs.mem]]
+[[inputs.system]]
+[[inputs.disk]]
+  ignore_fs = ["tmpfs", "devtmpfs", "devfs"]
+[[inputs.net]]
+[[inputs.win_perf_counters]]
+  [[inputs.win_perf_counters.object]]
+    ObjectName = "System"
+    Counters = ["System Up Time"]
+    Instances = ["------"]
+    Measurement = "win_system"
+
+[[inputs.nvidia_smi]]
+
+[[inputs.docker]]
+  endpoint = "npipe:////./pipe/docker_engine"
+  container_name_include = []
+
+[[inputs.http_response]]
+  urls = ["http://127.0.0.1:8000/v2/health/ready"]
+  response_timeout = "2s"
+  response_status_code = 200
+  name_override = "missioncore_triton_health"
+
+[[inputs.prometheus]]
+  urls = ["http://127.0.0.1:8002/metrics"]
+  metric_version = 1
+  response_timeout = "2s"
+  namepass = [
+    "nv_inference_request_success",
+    "nv_inference_request_failure",
+    "nv_inference_count",
+  ]
+
+[[outputs.mqtt]]
+  servers = ["tcp://${MISSIONCORE_MQTT_HOST}:${MISSIONCORE_MQTT_PORT}"]
+  topic = "mission-core/v1/contours/${MISSIONCORE_CONTOUR_ID}/agents/${MISSIONCORE_AGENT_ID}/host"
+  username = "${MISSIONCORE_MQTT_USERNAME}"
+  password = "${MISSIONCORE_MQTT_PASSWORD}"
+  qos = 1
+  data_format = "json"
+  json_timestamp_units = "1s"