wip

src/backend/core/recording/worker/services.py

@@ -2,6 +2,8 @@
 
 # pylint: disable=no-member
 
+import logging
+
 from asgiref.sync import async_to_sync
 from livekit import api as livekit_api
 
@@ -10,6 +12,8 @@ from ..enums import FileExtension
 from .exceptions import WorkerConnectionError, WorkerResponseError
 from .factories import WorkerServiceConfig
 
+logger = logging.getLogger(__name__)
+
 
 class BaseEgressService:
     """Base egress defining common methods to manage and interact with LiveKit egress processes."""
@@ -49,6 +53,23 @@ class BaseEgressService:
         finally:
             await lkapi.aclose()
 
+    @staticmethod
+    def _log_egress_error(response, event: str):
+        """Log the reason LiveKit reported an unsuccessful egress on stop.
+
+        Mirrors the logging done in the 'egress_ended' webhook. The
+        StopEgress response carries the same error fields.
+        """
+
+        logger.error(
+            "Egress %s on stop (egress_id=%s, status=%s): %s (error_code=%s)",
+            event,
+            response.egress_id,
+            livekit_api.EgressStatus.Name(response.status),
+            response.error or "no error reported",
+            response.error_code or "no error_code reported",
+        )
+
     def stop(self, worker_id: str) -> str:
         """Stop an ongoing egress worker.
         The StopEgressRequest is shared among all types of egress,
@@ -69,14 +90,17 @@ class BaseEgressService:
         # To avoid exposing EgressStatus values and coupling with LiveKit outside of this class,
         # the response status is mapped to simpler "ABORTED", "STOPPED" or "FAILED_TO_STOP" strings.
         if response.status == livekit_api.EgressStatus.EGRESS_ABORTED:
+            self._log_egress_error(response, "aborted")
             return "ABORTED"
 
         if response.status == livekit_api.EgressStatus.EGRESS_FAILED:
+            self._log_egress_error(response, "failed")
             return "FAILED"
 
         if response.status == livekit_api.EgressStatus.EGRESS_ENDING:
             return "STOPPED"
 
+        self._log_egress_error(response, "failed to stop")
         return "FAILED_TO_STOP"
 
     def start(self, room_name, recording_id):

src/backend/core/tests/recording/worker/test_services.py

@@ -4,6 +4,7 @@ Test worker service classes.
 
 # pylint: disable=protected-access,redefined-outer-name,unused-argument,no-member
 
+import logging
 from unittest.mock import AsyncMock, Mock, patch
 
 import pytest
@@ -175,6 +176,32 @@ def test_base_egress_stop_with_status(service, response_status, expected_result)
     assert result == expected_result
 
 
+@pytest.mark.parametrize(
+    "response_status,event",
+    [
+        (livekit_api.EgressStatus.EGRESS_ABORTED, "aborted"),
+        (livekit_api.EgressStatus.EGRESS_FAILED, "failed"),
+        (livekit_api.EgressStatus.EGRESS_COMPLETE, "failed to stop"),
+    ],
+)
+def test_base_egress_stop_logs_livekit_error(service, response_status, event, caplog):
+    """Should log the reason LiveKit reported for an unsuccessful stop."""
+    mock_response = Mock(
+        status=response_status,
+        egress_id="test_worker_id",
+        error="could not connect to the room",
+        error_code=500,
+    )
+    service._handle_request = Mock(return_value=mock_response)
+
+    with caplog.at_level(logging.ERROR):
+        service.stop("test_worker_id")
+
+    assert f"Egress {event} on stop (egress_id=test_worker_id" in caplog.text
+    assert "could not connect to the room" in caplog.text
+    assert "error_code=500" in caplog.text
+
+
 def test_base_egress_stop_missing_status(service):
     """Test stop method when response is missing status"""
     # Mock _handle_request with missing status