feat(telemetry): expose compute pipeline stage metrics

2026-07-29 01:53:31 +03:00
parent 783aa444ac
commit 79eb4b46f7
19 changed files with 1000 additions and 24 deletions
@@ -274,6 +274,57 @@ def test_worker_telemetry_prefers_ndc_container_names_during_migration(
    assert triton["canonical_name"] == "ndc-mission-core-triton"


+def test_agent_pipeline_snapshot_merges_all_stage_series() -> None:
+    samples = []
+    for stage_id, stage_state, elapsed, activations, share in (
+        ("detector", "active", 2.5, 42, 62.5),
+        ("semantic-model", "waiting", 1.5, 11, 37.5),
+    ):
+        samples.append(
+            {
+                "node_id": EXPECTED_NODE_ID,
+                "kind": "pipeline",
+                "measurement": "missioncore_pipeline",
+                "observed_at_utc": "2026-07-28T12:00:00Z",
+                "payload": {
+                    "name": "missioncore_pipeline",
+                    "tags": {"stage_id": stage_id},
+                    "fields": {
+                        "service_state": "busy",
+                        "current_stage": "detector",
+                        "active_request_id": "request-006",
+                        "active_frame_index": 17,
+                        "completed_runs": 3,
+                        "failed_runs": 0,
+                        "model_load_seconds": 10.5,
+                        "collector_state": "live",
+                        "stage_state": stage_state,
+                        "elapsed_seconds": elapsed,
+                        "activations": activations,
+                        "share_percent": share,
+                    },
+                },
+            }
+        )
+
+    raw = _agent_raw_document({"samples": samples})
+
+    assert raw["perception"]["state"] == "busy"
+    assert raw["perception"]["active_stages"] == ["detector"]
+    assert raw["perception"]["stage_metrics"] == {
+        "detector": {
+            "elapsed_seconds": 2.5,
+            "activations": 42,
+            "share_percent": 62.5,
+        },
+        "semantic-model": {
+            "elapsed_seconds": 1.5,
+            "activations": 11,
+            "share_percent": 37.5,
+        },
+    }
+
+
 def test_worker_telemetry_history_keeps_one_row_per_agent_observation(
    tmp_path: Path,
 ) -> None: