On / test_source_reconnect.py
Reoo88's picture
Fix live source handover without interrupting the YouTube publisher
8ede3fd verified
Raw History Blame Contribute Delete
13.5 kB
#!/usr/bin/env python3
"""Regression for live source handover; no real TikTok/YouTube requests.
Starting a replacement pump while swap_source=True made the new thread exit before
reading a single byte. FFmpeg was kept alive but YouTube received no replacement media.
"""
import io
import http.server
import json
import shutil
import subprocess
import tempfile
import os
import pathlib
import sys
import threading
import time
import unittest
from unittest.mock import Mock, patch
ROOT = pathlib.Path(sys.argv[1]).resolve() if len(sys.argv) > 1 else pathlib.Path(__file__).parent
sys.path.insert(0, str(ROOT))
os.environ.setdefault('DATA_DIR', tempfile.mkdtemp(prefix='handover-test-data-'))
from relay import Sender
from youtube_manager import ManagedYouTube
class InlineThread:
"""Deterministic scheduler: the new pump runs immediately when start() is called."""
def __init__(self, target, args=(), daemon=False):
self.target = target
self.args = args
self.daemon = daemon
self.started = False
def start(self):
self.started = True
self.target(*self.args)
def join(self, timeout=None):
pass
def is_alive(self):
return False
class Response:
def __init__(self, chunks=(b'replacement media',)):
self.chunks = chunks
self.closed = False
self.reads = 0
def raise_for_status(self):
pass
def iter_content(self, chunk_size=None):
for chunk in self.chunks:
self.reads += 1
yield chunk
def close(self):
self.closed = True
class HandoverTests(unittest.TestCase):
def setUp(self):
self.env = patch.dict(os.environ, {'SOURCE_STALL_TIMEOUT': '0'})
self.env.start()
self.addCleanup(self.env.stop)
self.sender = Sender('/bin/true', lambda *a: None)
self.p = Mock()
self.p.poll.return_value = None
self.p.stdin = io.BytesIO()
self.old_pump = Mock()
self.old_pump.is_alive.return_value = False
self.old_response = Response()
self.new_response = Response()
self.stop = threading.Event()
self.sess = {'last': time.monotonic() - 4, 'max_seconds': 180,
'force_stop': False, 'timeouts': 0, 'error': '', 'ended': True}
self.bridge = Mock()
self.bridge.feed.side_effect = lambda chunk: chunk
self.refresh = Mock(return_value={'url': 'https://example.invalid/source.flv', 'codec': 'h264', 'label': 'hd'})
def attach(self, refresh=None):
with patch('relay.requests.get', return_value=self.new_response), patch('relay.threading.Thread', InlineThread):
return self.sender._try_reattach('alice', self.p, self.bridge, self.sess,
self.stop, refresh or self.refresh, 10, 3, self.old_response, self.old_pump)
def test_new_pump_reads_media_even_when_scheduled_immediately(self):
attached = self.attach()
self.assertIsNotNone(attached)
self.assertEqual(self.new_response.reads, 1)
self.assertEqual(self.p.stdin.getvalue(), b'replacement media')
self.assertEqual(self.sender.snapshot()['source_bytes'], len(b'replacement media'))
self.assertFalse(self.sess['swap_source'])
def test_old_pump_must_stop_before_bridge_is_reused(self):
self.old_pump.is_alive.return_value = True
self.assertIsNone(self.attach())
self.bridge.reconnect.assert_not_called()
self.refresh.assert_not_called()
self.assertEqual(self.new_response.reads, 0)
def test_stop_during_resolution_does_not_resurrect_the_reader(self):
def stop_then_refresh():
self.stop.set()
return self.refresh()
self.assertIsNone(self.attach(stop_then_refresh))
self.bridge.reconnect.assert_not_called()
self.assertEqual(self.new_response.reads, 0)
def test_refresh_does_not_fabricate_source_heartbeat(self):
self.new_response = Response(chunks=())
previous = self.sess['last']
self.assertIsNotNone(self.attach())
self.assertEqual(self.sess['last'], previous)
self.assertEqual(self.sender.snapshot()['source_bytes'], 0)
def test_opened_source_has_bounded_first_byte_grace(self):
previous = self.sess['last']
now = time.monotonic()
self.new_response = Response(chunks=())
self.attach()
self.assertGreater(self.sess['reattach_not_before'], now)
self.assertLessEqual(self.sess['reattach_not_before'], time.monotonic() + 5)
self.assertEqual(self.sess['last'], previous)
def test_cancelled_reader_cannot_deliver_a_late_chunk(self):
cancel = threading.Event()
response = Response()
def late_chunk(_size=None):
cancel.set()
yield b'old reader late packet'
response.iter_content = late_chunk
with patch('relay.threading.Thread', InlineThread):
self.sender._pump_source(response, self.p, self.bridge, self.stop,
self.sess, 10, 3, cancel)
self.assertEqual(self.p.stdin.getvalue(), b'')
self.assertEqual(self.sender.snapshot()['source_bytes'], 0)
def test_no_source_is_a_failed_handover_not_a_fake_success(self):
self.refresh.return_value = None
self.assertIsNone(self.attach())
self.bridge.reconnect.assert_not_called()
class CanvasTests(unittest.TestCase):
def test_placeholder_16x16_does_not_pin_the_live_canvas(self):
sender = Sender('/bin/true', lambda *a: None)
self.assertEqual(sender.lock_gold({'width': 16, 'height': 16}), (0, 0))
self.assertEqual(sender.lock_gold({'width': 720, 'height': 1440}), (720, 1280))
def test_unknown_or_placeholder_matchgop_uses_safe_constant_canvas(self):
for candidate in ({}, {'width': 16, 'height': 16}):
sender = Sender('/bin/true', lambda *a: None)
sender.mode_override = 'matchgop'
args = sender.command('test.flv', candidate)
vf = args[args.index('-vf') + 1]
self.assertIn('scale=w=720:h=1280:', vf)
# A later real configuration does not mutate the encoder canvas.
sender.lock_gold({'width': 960, 'height': 1920})
self.assertEqual((sender._gop_w, sender._gop_h), (720, 1280))
def test_normal_and_real_1080_source_dimensions_remain_supported(self):
for size in ((160, 120), (320, 240), (1080, 1920)):
sender = Sender('/bin/true', lambda *a: None)
self.assertEqual(sender.lock_gold({'width': size[0], 'height': size[1]}), size)
class CachedSourceTests(unittest.TestCase):
def setUp(self):
self.tmp = tempfile.TemporaryDirectory()
self.addCleanup(self.tmp.cleanup)
self.manager = ManagedYouTube('/bin/true', lambda u, force=False: None,
lambda *a: None, lambda: ['alice'], pathlib.Path(self.tmp.name) / 'journal.json')
self.manager.video_sources = lambda sources: list(sources)
self.hd = {'url': 'https://example.invalid/hd.flv', 'media': 'flv', 'codec': 'h264',
'quality_name': 'hd', 'width': 720, 'height': 1280, 'label': 'hd'}
self.sd = dict(self.hd, url='https://example.invalid/sd.flv', quality_name='sd',
width=480, height=854, label='sd')
self.obs = {'status': 'live', 'room_id': 'same-room', 'sources': [self.hd, self.sd]}
self.manager._observe = Mock(return_value=self.obs)
def test_recent_compatible_source_is_reused_without_blocking_metadata_lookup(self):
self.manager._held['alice'] = (time.monotonic(), self.obs)
chosen = self.manager._refresh_flv('alice', self.hd)
self.assertEqual(chosen['url'], self.sd['url'])
self.manager._observe.assert_not_called()
def test_stale_source_snapshot_requires_fresh_lookup(self):
self.manager._held['alice'] = (time.monotonic() - 60, self.obs)
self.assertEqual(self.manager._refresh_flv('alice', self.hd)['url'], self.sd['url'])
self.manager._observe.assert_called_once_with('alice', force=True, bypass_throttle=True, allow_hold=False)
def test_cache_with_wrong_codec_cannot_be_spliced_into_open_publisher(self):
self.manager._held['alice'] = (time.monotonic(), dict(self.obs, sources=[dict(self.sd, codec='h265')]))
chosen = self.manager._refresh_flv('alice', self.hd)
self.assertEqual(chosen['codec'], 'h264')
self.manager._observe.assert_called_once()
@unittest.skipUnless(shutil.which('ffmpeg') and shutil.which('ffprobe'), 'FFmpeg required')
class RealHTTPContinuityTests(unittest.TestCase):
def run_continuity(self, mode):
with tempfile.TemporaryDirectory(prefix='source-handover-media-') as tmp:
root = pathlib.Path(tmp)
source = root / 'generated.flv'
output = root / 'output.flv'
subprocess.run([shutil.which('ffmpeg'), '-v', 'error', '-y', '-f', 'lavfi',
'-i', 'testsrc2=size=160x120:rate=20', '-f', 'lavfi', '-i', 'sine=sample_rate=48000',
'-t', '2', '-c:v', 'libx264', '-preset', 'ultrafast', '-pix_fmt', 'yuv420p',
'-g', '20', '-c:a', 'aac', '-f', 'flv', str(source)], check=True, capture_output=True)
media = source.read_bytes()
calls = []
class Handler(http.server.BaseHTTPRequestHandler):
def log_message(self, *a):
pass
def do_GET(self):
calls.append(time.monotonic())
self.send_response(200)
self.send_header('Content-Type', 'video/x-flv')
self.send_header('Content-Length', str(len(media)))
self.end_headers()
# Header arrives immediately; replacements wait before their first byte.
# Without startup grace the supervisor closes them before this media arrives.
if len(calls) > 1:
time.sleep(1)
try:
self.wfile.write(media)
self.wfile.flush()
except (OSError, BrokenPipeError):
pass
server = http.server.ThreadingHTTPServer(('127.0.0.1', 0), Handler)
server.daemon_threads = True
server_thread = threading.Thread(target=server.serve_forever, daemon=True)
server_thread.start()
stop = threading.Event()
errors = []
candidate = {'url': 'http://127.0.0.1:%d/source.flv' % server.server_port,
'media': 'flv', 'codec': 'h264', 'label': 'fixture', 'width': 160, 'height': 120}
sender = Sender(shutil.which('ffmpeg'), lambda *a: None)
sender.mode_override = mode
spawned = []
from relay import subprocess as relay_subprocess
real_popen = relay_subprocess.Popen
def track_process(*a, **kw):
child = real_popen(*a, **kw)
spawned.append(child)
return child
def send():
try:
sender.send('fixture', candidate, str(output), stop, refresh=lambda: dict(candidate))
except Exception as exc:
errors.append(exc)
env = {'SOURCE_STALL_TIMEOUT': '3', 'YOUTUBE_REATTACH_GRACE_SECONDS': '2',
'YOUTUBE_SOURCE_READ_TIMEOUT': '5', 'YOUTUBE_PUMP_JOIN_SECONDS': '2',
'YOUTUBE_SEND_MODE': mode, 'YOUTUBE_VIDEO_MODE': 'original',
'YOUTUBE_AUTO_TRANSCODE': '0', 'YOUTUBE_X264_PRESET': 'ultrafast'}
thread = None
try:
with patch.dict(os.environ, env), patch('relay.subprocess.Popen', side_effect=track_process):
thread = threading.Thread(target=send, daemon=True)
thread.start()
time.sleep(7.5)
alive = thread.is_alive() and sender.alive()
snapshot = sender.snapshot()
stop.set()
thread.join(6)
self.assertTrue(alive, 'Publisher ended instead of reusing the delayed replacement source: %r' % errors)
self.assertFalse(errors, repr(errors))
self.assertEqual(len(spawned), 1, 'Source reconnect must keep the same FFmpeg process')
self.assertGreaterEqual(len(calls), 3)
self.assertGreater(snapshot.get('output_seconds', 0), 4)
self.assertFalse(thread.is_alive(), 'Sender did not stop cleanly')
probe = subprocess.run([shutil.which('ffprobe'), '-v', 'error', '-show_entries',
'format=duration', '-of', 'json', str(output)], check=True, capture_output=True, text=True)
self.assertGreater(float(json.loads(probe.stdout)['format']['duration']), 4)
finally:
stop.set()
sender.terminate()
if thread:
thread.join(5)
server.shutdown()
server.server_close()
def test_eof_and_delayed_replacements_keep_same_copy_publisher(self):
self.run_continuity('copy')
def test_eof_and_delayed_replacements_keep_same_matchgop_publisher(self):
self.run_continuity('matchgop')
if __name__ == '__main__':
unittest.main(argv=[sys.argv[0]], verbosity=2)