Calls: store every radio heard on a call; call_end writes srcaddr (server-26#304)

P1 of the radio-ID plan. The node now reports `radios`, one
{radio, start, end} per transmission, on /upload and on call_end. Both go
through app/internal/radios.py into `radios` + `radio_ids` on the call doc.
/upload writes them before the pipeline starts, because call_end arrives
after it.

call_end also writes srcaddr now. Only call_start used to, and the node
usually learns the talker after call_start, so most calls kept None and the
correlator's srcaddr match rarely fired. /upload fills srcaddr from the
latest radio when the doc has none.

Older nodes send neither field; nothing is blanked.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

drb-c2-core/app/internal/mqtt_handler.py

@@ -6,6 +6,7 @@ import paho.mqtt.client as mqtt
 from app.config import settings
 from app.internal.logger import logger
 from app.internal import firestore as fstore
+from app.internal import radios
 from app.internal import talkgroups
 from app.internal.tenancy import FOUNDING_ORG_ID
 
@@ -307,6 +308,12 @@ class MQTTHandler:
             updates["org_id"] = node["org_id"]
         if payload.get("audio_url"):
             updates["audio_url"] = payload["audio_url"]
+        # The node usually learns the talker after call_start went out; only
+        # call_start used to write srcaddr, so most calls kept None
+        # (server-26#304).
+        if payload.get("srcaddr") is not None:
+            updates["srcaddr"] = payload["srcaddr"]
+        updates.update(radios.call_fields(payload.get("radios")))
 
         await fstore.doc_set("calls", call_id, updates)
         logger.info(f"Call end: {call_id}")

drb-c2-core/app/internal/radios.py

@@ -0,0 +1,64 @@
+"""
+Every radio heard on a call, as the edge node reports it (server-26#304).
+
+The node sends `radios`: a list of {radio, start, end}, one entry per
+transmission, offsets in seconds into the uploaded audio file, so a talker's
+wall-clock time is audio_start_epoch + offset (the same mapping the node uses
+for the audio itself). It arrives twice — on
+/upload (as a JSON string form field, because the pipeline starts there) and
+on the MQTT call_end — and both go through `call_fields` so the call doc gets
+the same shape either way.
+
+`radio_ids` is the distinct IDs in first-heard order, kept alongside so a
+correlator or query can ask "was radio X on this call" without walking the list.
+"""
+import json
+from typing import Any, Dict, List, Optional
+
+from app.internal.logger import logger
+
+# A 600 s recording polled every 0.5 s cannot honestly hold more than this.
+MAX_TRANSMISSIONS = 200
+
+
+def _num(value: Any) -> Optional[float]:
+    try:
+        number = float(value)
+    except (TypeError, ValueError):
+        return None
+    return number if number >= 0 else None
+
+
+def parse(value: Any) -> List[Dict[str, Any]]:
+    """The node's radio list, validated. Accepts a list or its JSON string; junk entries are dropped."""
+    if value is None or value == "":
+        return []
+    if isinstance(value, str):
+        try:
+            value = json.loads(value)
+        except ValueError:
+            logger.warning(f"radios: not JSON, ignored ({value[:80]!r})")
+            return []
+    if not isinstance(value, list):
+        return []
+    out: List[Dict[str, Any]] = []
+    for item in value[:MAX_TRANSMISSIONS]:
+        if not isinstance(item, dict):
+            continue
+        try:
+            radio = int(item.get("radio"))
+        except (TypeError, ValueError):
+            continue
+        start, end = _num(item.get("start")), _num(item.get("end"))
+        if radio <= 0 or start is None or end is None:
+            continue
+        out.append({"radio": radio, "start": start, "end": max(start, end)})
+    return out
+
+
+def call_fields(value: Any) -> Dict[str, Any]:
+    """Call-doc fields for a radio list; empty when the node sent none (older nodes, analog)."""
+    radios = parse(value)
+    if not radios:
+        return {}
+    return {"radios": radios, "radio_ids": list(dict.fromkeys(r["radio"] for r in radios))}

drb-c2-core/app/routers/upload.py

@@ -5,6 +5,7 @@ from fastapi import APIRouter, BackgroundTasks, UploadFile, File, Form, HTTPExce
 from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
 from app.internal.storage import upload_audio
 from app.internal import dedup
+from app.internal import radios
 from app.internal import firestore as fstore
 from app.internal import clock
 from app.internal.logger import logger
@@ -24,6 +25,7 @@ async def upload_call_audio(
     talkgroup_id: Optional[int] = Form(None),
     talkgroup_name: Optional[str] = Form(None),
     system_id: Optional[str] = Form(None),
+    radios_json: Optional[str] = Form(None, alias="radios"),
     credentials: Optional[HTTPAuthorizationCredentials] = Security(_bearer),
 ):
     """
@@ -101,6 +103,14 @@ async def upload_call_audio(
             updates = {"audio_gcs_uri": gcs_uri}
             if node and node.get("org_id"):
                 updates["org_id"] = node["org_id"]
+            # Every radio heard on the call (server-26#304). Written here, not
+            # only on call_end, because the pipeline below starts before the
+            # node publishes call_end.
+            radio_fields = radios.call_fields(radios_json)
+            updates.update(radio_fields)
+            if radio_fields and existing.get("srcaddr") is None:
+                # Same meaning call_end gives it: the latest talker.
+                updates["srcaddr"] = radio_fields["radios"][-1]["radio"]
             await fstore.doc_set("calls", call_id, updates)
         except Exception as e:
             logger.warning(f"Could not update call {call_id} with audio_gcs_uri: {e}")

drb-c2-core/tests/test_radios.py

@@ -0,0 +1,100 @@
+"""
+Every radio on a call reaches the call doc (server-26#304) — from /upload,
+which starts the pipeline, and from call_end, which used to drop srcaddr.
+"""
+import json
+from unittest.mock import AsyncMock, patch
+
+import pytest
+from fastapi.testclient import TestClient
+
+from app.internal import radios
+from app.internal.mqtt_handler import MQTTHandler
+from app.main import app
+from app.routers import upload
+
+LIST = [{"radio": 555, "start": 0.0, "end": 1.0}, {"radio": 777, "start": 2.0, "end": 2.5},
+        {"radio": 555, "start": 3.5, "end": 4.0}]
+
+
+def test_parse_accepts_json_and_drops_junk():
+    raw = json.dumps(LIST + [{"radio": 0, "start": 1, "end": 2}, {"radio": "x"}, "junk",
+                             {"radio": 9, "start": -1, "end": 2}, {"radio": 8, "start": 5, "end": 4}])
+    assert radios.parse(raw) == LIST + [{"radio": 8, "start": 5.0, "end": 5.0}]
+    assert radios.parse("not json") == []
+    assert radios.parse({"radio": 1}) == []
+    assert radios.parse(None) == []
+
+
+def test_call_fields_keeps_distinct_ids_in_first_heard_order():
+    assert radios.call_fields(LIST)["radio_ids"] == [555, 777]
+    assert radios.call_fields([]) == {}
+
+
+@pytest.mark.asyncio
+async def test_call_end_writes_srcaddr_and_radios():
+    """call_start used to be the only writer of srcaddr; the talker is usually learned later."""
+    handler = MQTTHandler()
+    with patch("app.internal.mqtt_handler.fstore") as fs:
+        fs.doc_get_cached = AsyncMock(return_value=None)
+        fs.doc_set = AsyncMock()
+        await handler._on_call_end("node-01", {"call_id": "c1", "srcaddr": 555, "radios": LIST})
+    updates = fs.doc_set.call_args[0][2]
+    assert updates["srcaddr"] == 555
+    assert updates["radios"] == LIST
+    assert updates["radio_ids"] == [555, 777]
+
+
+@pytest.mark.asyncio
+async def test_call_end_without_radios_leaves_them_alone():
+    """An older node sends neither field; a merge write must not blank what /upload stored."""
+    handler = MQTTHandler()
+    with patch("app.internal.mqtt_handler.fstore") as fs:
+        fs.doc_get_cached = AsyncMock(return_value=None)
+        fs.doc_set = AsyncMock()
+        await handler._on_call_end("node-01", {"call_id": "c1", "srcaddr": None})
+    updates = fs.doc_set.call_args[0][2]
+    assert "srcaddr" not in updates and "radios" not in updates
+
+
+def _upload(existing_call, form_radios):
+    store = {
+        "node_keys": {"api_key": "k"},
+        "nodes": {"approval_status": "approved", "org_id": "org"},
+        "calls": existing_call,
+    }
+    writes = []
+
+    async def doc_get(c, i):
+        return store.get(c)
+
+    async def doc_set(c, i, data, merge=True):
+        writes.append((c, data))
+
+    data = {"call_id": "c1", "node_id": "n1"}
+    if form_radios is not None:
+        data["radios"] = form_radios
+    with patch.object(upload.fstore, "doc_get", doc_get), \
+         patch.object(upload.fstore, "doc_set", doc_set), \
+         patch.object(upload, "upload_audio", AsyncMock(return_value="gs://b/c1.mp3")), \
+         patch.object(upload.dedup, "find_duplicate_of", AsyncMock(return_value="other")):
+        resp = TestClient(app).post(
+            "/upload", data=data,
+            files={"file": ("a.mp3", b"\x00\x01", "audio/mpeg")},
+            headers={"Authorization": "Bearer k"},
+        )
+    assert resp.status_code == 200
+    return writes[0][1]
+
+
+def test_upload_stores_the_radio_list_before_the_pipeline_runs():
+    first = _upload({}, json.dumps(LIST))
+    assert first["radios"] == LIST
+    assert first["radio_ids"] == [555, 777]
+    # call_start had no talker yet: take the latest one, as call_end would.
+    assert first["srcaddr"] == 555
+
+
+def test_upload_keeps_an_existing_srcaddr_and_tolerates_no_list():
+    assert "srcaddr" not in _upload({"srcaddr": 999}, json.dumps(LIST))
+    assert "radios" not in _upload({}, None)