Files
frigate/frigate/test/test_camera_watchdog.py
T
Josh HawkinsandGitHub 5d3b955bcb don't race the capture thread when stopping ffmpeg (#24884)
stop_all_ffmpeg called communicate() on the detect ffmpeg even when the capture thread was still reading stdout, which happened on every ffmpeg, record, and record_sub config restart and when a camera stalled during a stop. The two readers could buffer duplicate frames without bound or hang the stop. When the capture thread is still alive, it now terminates and waits, then kills, so the live thread drains the pipe and ffmpeg can finish its segment. Disable and shutdown still use stop_ffmpeg.
2026-11-07 06:32:25 -07:00

306 lines
12 KiB
Python

"""Tests for per stream recording health tracking in the camera watchdog."""
import subprocess as sp
import unittest
from datetime import UTC, datetime, timedelta
from unittest.mock import MagicMock, call, patch
from frigate.config import FrigateConfig
from frigate.const import STREAM_TYPE_MAIN, STREAM_TYPE_SUB
from frigate.video.ffmpeg import CameraWatchdog
def build_watchdog(
sub_enabled: bool = True, output_args: dict | None = None
) -> CameraWatchdog:
config = FrigateConfig(
**{
"mqtt": {"host": "mqtt"},
"cameras": {
"front_door": {
"ffmpeg": {
"output_args": output_args or {},
"inputs": [
{
"path": "rtsp://10.0.0.1:554/video",
"roles": ["record"],
},
{
"path": "rtsp://10.0.0.1:554/video2",
"roles": ["detect", "record_sub"],
},
],
},
"record": {
"enabled": True,
"sub": {"enabled": sub_enabled},
},
}
},
}
)
camera_config = config.cameras["front_door"]
with (
patch("frigate.video.ffmpeg.LogPipe"),
patch("frigate.video.ffmpeg.InterProcessRequestor"),
patch("frigate.video.ffmpeg.RecordingsDataSubscriber"),
patch("frigate.video.ffmpeg.CameraConfigUpdateSubscriber"),
):
watchdog = CameraWatchdog(
camera_config,
1,
MagicMock(),
MagicMock(),
MagicMock(),
MagicMock(),
MagicMock(),
MagicMock(),
MagicMock(),
MagicMock(),
)
return watchdog
class TestCameraWatchdogReset(unittest.TestCase):
def _build_resettable_watchdog(self) -> CameraWatchdog:
watchdog = build_watchdog()
watchdog.ffmpeg_detect_process = MagicMock()
watchdog.capture_thread = MagicMock()
watchdog.capture_thread.is_alive.return_value = False
watchdog.start_ffmpeg_detect = MagicMock()
return watchdog
def test_reset_does_not_buffer_raw_frames(self):
watchdog = self._build_resettable_watchdog()
process = watchdog.ffmpeg_detect_process
watchdog.reset_capture_thread()
process.terminate.assert_called_once()
process.wait.assert_called_once_with(timeout=30)
process.communicate.assert_not_called()
watchdog.start_ffmpeg_detect.assert_called_once()
def test_timed_out_reset_kills_and_reaps_without_reading_stdout(self):
watchdog = self._build_resettable_watchdog()
process = watchdog.ffmpeg_detect_process
process.wait.side_effect = [sp.TimeoutExpired("ffmpeg", 30), 0]
watchdog.reset_capture_thread()
process.kill.assert_called_once()
self.assertEqual(process.wait.call_args_list, [call(timeout=30), call()])
process.communicate.assert_not_called()
watchdog.start_ffmpeg_detect.assert_called_once()
class TestCameraWatchdogStop(unittest.TestCase):
def _build_stopping_watchdog(self, capture_reading: bool) -> CameraWatchdog:
watchdog = build_watchdog()
watchdog.ffmpeg_detect_process = MagicMock()
watchdog.capture_thread = MagicMock()
watchdog.capture_thread.is_alive.return_value = capture_reading
return watchdog
def test_live_capture_thread_terminates_without_reading_stdout(self):
watchdog = self._build_stopping_watchdog(capture_reading=True)
process = watchdog.ffmpeg_detect_process
with patch("frigate.video.ffmpeg.stop_ffmpeg") as stop_ffmpeg:
watchdog.stop_all_ffmpeg()
process.terminate.assert_called_once()
process.wait.assert_called_once_with(timeout=30)
process.kill.assert_not_called()
process.communicate.assert_not_called()
stop_ffmpeg.assert_not_called()
self.assertIsNone(watchdog.ffmpeg_detect_process)
def test_live_capture_thread_kills_ffmpeg_that_ignores_sigterm(self):
watchdog = self._build_stopping_watchdog(capture_reading=True)
process = watchdog.ffmpeg_detect_process
process.wait.side_effect = [sp.TimeoutExpired("ffmpeg", 30), 0]
watchdog.stop_all_ffmpeg()
process.kill.assert_called_once()
self.assertEqual(process.wait.call_args_list, [call(timeout=30), call()])
process.communicate.assert_not_called()
def test_joined_capture_thread_uses_stop_ffmpeg(self):
watchdog = self._build_stopping_watchdog(capture_reading=False)
process = watchdog.ffmpeg_detect_process
watchdog.capture_thread.is_alive.side_effect = [True, False]
with patch("frigate.video.ffmpeg.stop_ffmpeg") as stop_ffmpeg:
watchdog.stop_all_ffmpeg()
stop_ffmpeg.assert_called_once_with(process, watchdog.logger)
process.terminate.assert_not_called()
self.assertIsNone(watchdog.ffmpeg_detect_process)
class TestCameraWatchdogStreamHealth(unittest.TestCase):
def _build_watchdog(
self, sub_enabled: bool = True, output_args: dict | None = None
) -> CameraWatchdog:
return build_watchdog(sub_enabled, output_args)
def test_stale_sub_does_not_mark_main_stale(self):
watchdog = self._build_watchdog()
now = datetime.now().astimezone(UTC)
stale = (now - timedelta(hours=1)).timestamp()
watchdog.latest_cache_segment_time[STREAM_TYPE_MAIN] = now.timestamp()
watchdog.latest_valid_segment_time[STREAM_TYPE_MAIN] = now.timestamp()
watchdog.latest_cache_segment_time[STREAM_TYPE_SUB] = stale
watchdog.latest_valid_segment_time[STREAM_TYPE_SUB] = stale
assert watchdog._stream_staleness(STREAM_TYPE_MAIN, now) is None
assert watchdog._stream_staleness(STREAM_TYPE_SUB, now) is not None
def test_stale_main_does_not_mark_sub_stale(self):
watchdog = self._build_watchdog()
now = datetime.now().astimezone(UTC)
stale = (now - timedelta(hours=1)).timestamp()
watchdog.latest_cache_segment_time[STREAM_TYPE_MAIN] = stale
watchdog.latest_valid_segment_time[STREAM_TYPE_MAIN] = stale
watchdog.latest_cache_segment_time[STREAM_TYPE_SUB] = now.timestamp()
watchdog.latest_valid_segment_time[STREAM_TYPE_SUB] = now.timestamp()
assert watchdog._stream_staleness(STREAM_TYPE_MAIN, now) is not None
assert watchdog._stream_staleness(STREAM_TYPE_SUB, now) is None
def test_grace_period_suppresses_staleness(self):
watchdog = self._build_watchdog()
now = datetime.now().astimezone(UTC)
watchdog.record_enable_time = now - timedelta(seconds=10)
watchdog.latest_cache_segment_time[STREAM_TYPE_SUB] = (
now - timedelta(hours=1)
).timestamp()
assert watchdog._stream_staleness(STREAM_TYPE_SUB, now) is None
def test_status_goes_to_the_matching_role_topic(self):
watchdog = self._build_watchdog()
watchdog.record_status[STREAM_TYPE_MAIN].send("online", 100.0)
watchdog.record_status[STREAM_TYPE_SUB].send("offline", 100.0)
watchdog.requestor.send_data.assert_any_call(
"front_door/status/record", "online"
)
watchdog.requestor.send_data.assert_any_call(
"front_door/status/record_sub", "offline"
)
def test_status_is_cached_per_stream(self):
watchdog = self._build_watchdog()
watchdog.record_status[STREAM_TYPE_MAIN].send("online", 100.0)
watchdog.record_status[STREAM_TYPE_SUB].send("online", 100.0)
watchdog.record_status[STREAM_TYPE_MAIN].send("online", 100.0)
assert watchdog.requestor.send_data.call_count == 2
def test_recorded_streams_follows_config(self):
watchdog = self._build_watchdog()
assert watchdog._recorded_streams(["record"]) == [STREAM_TYPE_MAIN]
assert watchdog._recorded_streams(["detect", "record_sub"]) == [STREAM_TYPE_SUB]
assert watchdog._recorded_streams(["detect"]) == []
disabled = self._build_watchdog(sub_enabled=False)
assert disabled._recorded_streams(["detect", "record_sub"]) == []
def test_restart_grace_suppresses_repeat_staleness(self):
watchdog = self._build_watchdog()
now = datetime.now().astimezone(UTC)
stale = (now - timedelta(hours=1)).timestamp()
watchdog.latest_cache_segment_time[STREAM_TYPE_MAIN] = stale
watchdog.latest_valid_segment_time[STREAM_TYPE_MAIN] = stale
assert watchdog._stream_staleness(STREAM_TYPE_MAIN, now) is not None
watchdog._grant_restart_grace([STREAM_TYPE_MAIN], now)
assert watchdog._stream_staleness(STREAM_TYPE_MAIN, now) is None
assert (
watchdog._stream_staleness(STREAM_TYPE_MAIN, now + timedelta(seconds=89))
is None
)
assert (
watchdog._stream_staleness(STREAM_TYPE_MAIN, now + timedelta(seconds=91))
is not None
)
def test_restart_grace_is_per_stream(self):
watchdog = self._build_watchdog()
now = datetime.now().astimezone(UTC)
stale = (now - timedelta(hours=1)).timestamp()
for stream_type in (STREAM_TYPE_MAIN, STREAM_TYPE_SUB):
watchdog.latest_cache_segment_time[stream_type] = stale
watchdog.latest_valid_segment_time[stream_type] = stale
watchdog._grant_restart_grace([STREAM_TYPE_MAIN], now)
assert watchdog._stream_staleness(STREAM_TYPE_MAIN, now) is None
assert watchdog._stream_staleness(STREAM_TYPE_SUB, now) is not None
def test_detect_reset_grants_the_shared_sub_stream_grace(self):
watchdog = self._build_watchdog()
watchdog.detect_process_records_sub = True
watchdog.ffmpeg_detect_process = MagicMock()
watchdog.capture_thread = MagicMock()
watchdog.capture_thread.is_alive.return_value = False
watchdog.start_ffmpeg_detect = MagicMock()
now = datetime.now().astimezone(UTC)
stale = (now - timedelta(hours=1)).timestamp()
watchdog.latest_cache_segment_time[STREAM_TYPE_SUB] = stale
watchdog.latest_valid_segment_time[STREAM_TYPE_SUB] = stale
assert watchdog._stream_staleness(STREAM_TYPE_SUB, now) is not None
watchdog.reset_capture_thread(terminate=False)
# the sub check runs later in the same tick against a stale can_restart,
# so without this grace it would kill the just-restarted process again
assert (
watchdog._stream_staleness(STREAM_TYPE_SUB, datetime.now().astimezone(UTC))
is None
)
def test_detect_reset_leaves_sub_alone_when_not_shared(self):
watchdog = self._build_watchdog()
watchdog.detect_process_records_sub = False
watchdog.ffmpeg_detect_process = MagicMock()
watchdog.capture_thread = MagicMock()
watchdog.capture_thread.is_alive.return_value = False
watchdog.start_ffmpeg_detect = MagicMock()
now = datetime.now().astimezone(UTC)
stale = (now - timedelta(hours=1)).timestamp()
watchdog.latest_cache_segment_time[STREAM_TYPE_SUB] = stale
watchdog.latest_valid_segment_time[STREAM_TYPE_SUB] = stale
watchdog.reset_capture_thread(terminate=False)
assert (
watchdog._stream_staleness(STREAM_TYPE_SUB, datetime.now().astimezone(UTC))
is not None
)
def test_stale_threshold_follows_each_stream_segment_time(self):
watchdog = self._build_watchdog(
output_args={
"record": "-f segment -segment_time 10 -c copy",
"record_sub": "-f segment -segment_time 60 -c copy",
}
)
assert watchdog.record_stale_threshold[STREAM_TYPE_MAIN] == 120
assert watchdog.record_stale_threshold[STREAM_TYPE_SUB] == 150