From 8cd36416e2e52925fdf52fe6dae6ab3c232595c8 Mon Sep 17 00:00:00 2001
From: Li Xingyu <33626690+Li-Xingyu@users.noreply.github.com>
Date: Wed, 2 Sep 2026 22:24:01 +0800
Subject: [PATCH] Avoid killing recovered detector during restart (#24146)

---
 frigate/object_detection/base.py      | 18 ++++++-
 frigate/test/test_detector_restart.py | 76 +++++++++++++++++++++++++++
 2 files changed, 92 insertions(+), 2 deletions(-)
 create mode 100644 frigate/test/test_detector_restart.py

diff --git a/frigate/object_detection/base.py b/frigate/object_detection/base.py
index 858c8a5e4d..bda230348f 100644
--- a/frigate/object_detection/base.py
+++ b/frigate/object_detection/base.py
@@ -353,9 +353,23 @@ class ObjectDetectProcess:
         logging.info("Detection process has exited...")
 
     def start_or_restart(self) -> None:
-        self.detection_start.value = 0.0  # type: ignore[attr-defined]
         if (self.detect_process is not None) and self.detect_process.is_alive():
-            self.stop()
+            logging.info("Waiting for detection process to exit gracefully...")
+            self.detect_process.join(timeout=30)
+            if self.detect_process.exitcode is None:
+                # detection_start is set only after detection_queue.get()
+                # returns. If it was reset during the grace period, the process
+                # recovered and may be waiting on the shared queue again.
+                if self.detection_start.value == 0.0:  # type: ignore[attr-defined]
+                    logging.info("Detection process recovered before restart")
+                    return
+
+                logging.info("Detection process didn't exit. Force killing...")
+                self.detect_process.kill()
+                self.detect_process.join()
+            logging.info("Detection process has exited...")
+
+        self.detection_start.value = 0.0  # type: ignore[attr-defined]
 
         # Async path for MemryX
         if self.detector_config.type == "memryx":
diff --git a/frigate/test/test_detector_restart.py b/frigate/test/test_detector_restart.py
new file mode 100644
index 0000000000..ecb185d345
--- /dev/null
+++ b/frigate/test/test_detector_restart.py
@@ -0,0 +1,76 @@
+import unittest
+from types import SimpleNamespace
+from unittest.mock import Mock, call, patch
+
+from frigate.object_detection.base import ObjectDetectProcess
+
+
+def _make_detector(
+    detection_start: SimpleNamespace,
+    process: Mock,
+    stop_event: Mock,
+) -> ObjectDetectProcess:
+    detector = ObjectDetectProcess.__new__(ObjectDetectProcess)
+    detector.name = "test"
+    detector.cameras = ["front"]
+    detector.detection_queue = Mock()
+    detector.avg_inference_speed = Mock()
+    detector.detection_start = detection_start
+    detector.detect_process = process
+    detector.config = Mock()
+    detector.detector_config = SimpleNamespace(type="onnx")
+    detector.stop_event = stop_event
+    return detector
+
+
+class TestDetectorRestart(unittest.TestCase):
+    @patch("frigate.object_detection.base.DetectorRunner")
+    def test_restart_is_abandoned_when_inference_recovers(
+        self, detector_runner: Mock
+    ) -> None:
+        detection_start = SimpleNamespace(value=123.0)
+        process = Mock(exitcode=None)
+        process.is_alive.return_value = True
+        process.join.side_effect = lambda timeout=None: setattr(
+            detection_start, "value", 0.0
+        )
+        detector = _make_detector(detection_start, process, Mock())
+
+        detector.start_or_restart()
+
+        process.join.assert_called_once_with(timeout=30)
+        process.kill.assert_not_called()
+        self.assertIs(detector.detect_process, process)
+        detector_runner.assert_not_called()
+
+    @patch("frigate.object_detection.base.DetectorRunner")
+    def test_restart_force_kills_detector_that_remains_stuck(
+        self, detector_runner: Mock
+    ) -> None:
+        detection_start = SimpleNamespace(value=123.0)
+        process = Mock(exitcode=None)
+        process.is_alive.return_value = True
+        process.kill.side_effect = lambda: setattr(process, "exitcode", -9)
+        stop_event = Mock()
+        detector = _make_detector(detection_start, process, stop_event)
+
+        detector.start_or_restart()
+
+        process.join.assert_has_calls([call(timeout=30), call()])
+        process.kill.assert_called_once_with()
+        self.assertEqual(detection_start.value, 0.0)
+        detector_runner.assert_called_once_with(
+            "frigate.detector:test",
+            detector.detection_queue,
+            detector.cameras,
+            detector.avg_inference_speed,
+            detection_start,
+            detector.config,
+            detector.detector_config,
+            stop_event,
+        )
+        detector_runner.return_value.start.assert_called_once_with()
+
+
+if __name__ == "__main__":
+    unittest.main()