|
| 1 | +"""End-to-end Test for simulcast video quality layers. |
| 2 | +
|
| 3 | +The test publishes a 1280x720 simulcast video track (rolling colored bars) using |
| 4 | +both VP8 and H264 codecs, and on the receiver side verifies that subscribing to |
| 5 | +each simulcast quality layer (HIGH=f, MEDIUM=h, LOW=q) yields frames of the |
| 6 | +expected resolution. |
| 7 | +
|
| 8 | +Requires the following environment variables to run: |
| 9 | + LIVEKIT_URL |
| 10 | + LIVEKIT_API_KEY |
| 11 | + LIVEKIT_API_SECRET |
| 12 | +""" |
| 13 | + |
| 14 | +from __future__ import annotations |
| 15 | + |
| 16 | +import asyncio |
| 17 | +import os |
| 18 | +import time |
| 19 | +import uuid |
| 20 | +from typing import Callable, Optional, Tuple |
| 21 | + |
| 22 | +import numpy as np |
| 23 | +import pytest |
| 24 | + |
| 25 | +from livekit import api, rtc |
| 26 | +from livekit.rtc._proto.track_publication_pb2 import VideoQuality |
| 27 | +from livekit.rtc.room import EventTypes |
| 28 | + |
| 29 | + |
| 30 | +WAIT_TIMEOUT = 30.0 |
| 31 | +WAIT_INTERVAL = 0.1 |
| 32 | +PUBLISH_WIDTH = 1280 |
| 33 | +PUBLISH_HEIGHT = 720 |
| 34 | +PUBLISH_FPS = 15 |
| 35 | + |
| 36 | +# Default simulcast layer dimensions for a 720p source publication. |
| 37 | +LAYER_DIMENSIONS = { |
| 38 | + "f": (1280, 720), |
| 39 | + "h": (640, 360), |
| 40 | + "q": (320, 180), |
| 41 | +} |
| 42 | + |
| 43 | +QUALITY_SEQUENCE = [ |
| 44 | + (VideoQuality.VIDEO_QUALITY_HIGH, "f"), |
| 45 | + (VideoQuality.VIDEO_QUALITY_MEDIUM, "h"), |
| 46 | + (VideoQuality.VIDEO_QUALITY_LOW, "q"), |
| 47 | +] |
| 48 | + |
| 49 | + |
| 50 | +def skip_if_no_credentials(): |
| 51 | + required_vars = ["LIVEKIT_URL", "LIVEKIT_API_KEY", "LIVEKIT_API_SECRET"] |
| 52 | + missing = [var for var in required_vars if not os.getenv(var)] |
| 53 | + return pytest.mark.skipif( |
| 54 | + bool(missing), reason=f"Missing environment variables: {', '.join(missing)}" |
| 55 | + ) |
| 56 | + |
| 57 | + |
| 58 | +def create_token(identity: str, room_name: str) -> str: |
| 59 | + return ( |
| 60 | + api.AccessToken() |
| 61 | + .with_identity(identity) |
| 62 | + .with_name(identity) |
| 63 | + .with_grants( |
| 64 | + api.VideoGrants( |
| 65 | + room_join=True, |
| 66 | + room=room_name, |
| 67 | + ) |
| 68 | + ) |
| 69 | + .to_jwt() |
| 70 | + ) |
| 71 | + |
| 72 | + |
| 73 | +def unique_room_name(base: str) -> str: |
| 74 | + return f"{base}-{uuid.uuid4().hex[:8]}" |
| 75 | + |
| 76 | + |
| 77 | +async def _wait_until( |
| 78 | + predicate: Callable[[], bool], |
| 79 | + *, |
| 80 | + timeout: float = WAIT_TIMEOUT, |
| 81 | + interval: float = WAIT_INTERVAL, |
| 82 | + message: str = "condition not met", |
| 83 | +) -> None: |
| 84 | + loop = asyncio.get_event_loop() |
| 85 | + deadline = loop.time() + timeout |
| 86 | + while loop.time() < deadline: |
| 87 | + if predicate(): |
| 88 | + return |
| 89 | + await asyncio.sleep(interval) |
| 90 | + raise AssertionError(f"timeout waiting: {message}") |
| 91 | + |
| 92 | + |
| 93 | +async def _ensure_all_connected(rooms: list[rtc.Room]) -> None: |
| 94 | + await _wait_until( |
| 95 | + lambda: all(r.connection_state == rtc.ConnectionState.CONN_CONNECTED for r in rooms), |
| 96 | + message="not all rooms reached CONN_CONNECTED", |
| 97 | + ) |
| 98 | + |
| 99 | + |
| 100 | +async def _ensure_track_subscribed(room: rtc.Room, track_sid: str) -> rtc.RemoteTrackPublication: |
| 101 | + holder: dict[str, rtc.RemoteTrackPublication] = {} |
| 102 | + |
| 103 | + def _has_subscribed() -> bool: |
| 104 | + for participant in room.remote_participants.values(): |
| 105 | + pub = participant.track_publications.get(track_sid) |
| 106 | + if pub is not None and pub.subscribed and pub.track is not None: |
| 107 | + holder["pub"] = pub |
| 108 | + return True |
| 109 | + return False |
| 110 | + |
| 111 | + await _wait_until( |
| 112 | + _has_subscribed, |
| 113 | + message=f"room did not subscribe to track {track_sid}", |
| 114 | + ) |
| 115 | + return holder["pub"] |
| 116 | + |
| 117 | + |
| 118 | +def _expect_event( |
| 119 | + room: rtc.Room, |
| 120 | + event: EventTypes, |
| 121 | + predicate: Optional[Callable[..., bool]] = None, |
| 122 | +) -> asyncio.Future: |
| 123 | + loop = asyncio.get_event_loop() |
| 124 | + fut: asyncio.Future = loop.create_future() |
| 125 | + |
| 126 | + def _on_event(*args, **kwargs) -> None: |
| 127 | + if fut.done(): |
| 128 | + return |
| 129 | + if predicate is None or predicate(*args, **kwargs): |
| 130 | + fut.set_result(args) |
| 131 | + |
| 132 | + room.on(event, _on_event) |
| 133 | + return fut |
| 134 | + |
| 135 | + |
| 136 | +async def _await_event(fut: asyncio.Future, timeout: float = WAIT_TIMEOUT) -> None: |
| 137 | + try: |
| 138 | + await asyncio.wait_for(fut, timeout=timeout) |
| 139 | + except asyncio.TimeoutError as e: |
| 140 | + raise AssertionError("timed out waiting for event") from e |
| 141 | + |
| 142 | + |
| 143 | +def _make_rolling_i420(width: int, height: int, t: float) -> rtc.VideoFrame: |
| 144 | + """Build a 1280x720 I420 frame containing 8 vertical color bars that scroll |
| 145 | + horizontally over time, so the encoder always sees motion.""" |
| 146 | + bar_w = max(width // 8, 1) |
| 147 | + offset = int(t * 240) % bar_w |
| 148 | + |
| 149 | + cols_y = np.arange(width, dtype=np.int32) |
| 150 | + bar_idx_y = ((cols_y + offset) // bar_w) % 8 |
| 151 | + y_row = (bar_idx_y * 28 + 32).astype(np.uint8) |
| 152 | + |
| 153 | + cw = width // 2 |
| 154 | + cols_c = np.arange(cw, dtype=np.int32) |
| 155 | + bar_idx_c = (((cols_c * 2) + offset) // bar_w) % 8 |
| 156 | + u_row = (bar_idx_c * 18 + 80).astype(np.uint8) |
| 157 | + v_row = (220 - bar_idx_c * 18).astype(np.uint8) |
| 158 | + |
| 159 | + y_plane = np.tile(y_row, (height, 1)) |
| 160 | + u_plane = np.tile(u_row, (height // 2, 1)) |
| 161 | + v_plane = np.tile(v_row, (height // 2, 1)) |
| 162 | + |
| 163 | + data = np.concatenate([y_plane.ravel(), u_plane.ravel(), v_plane.ravel()]) |
| 164 | + return rtc.VideoFrame(width, height, rtc.VideoBufferType.I420, data.tobytes()) |
| 165 | + |
| 166 | + |
| 167 | +async def _publish_loop(source: rtc.VideoSource, stop: asyncio.Event) -> None: |
| 168 | + interval = 1.0 / PUBLISH_FPS |
| 169 | + start = time.monotonic() |
| 170 | + while not stop.is_set(): |
| 171 | + t = time.monotonic() - start |
| 172 | + frame = _make_rolling_i420(PUBLISH_WIDTH, PUBLISH_HEIGHT, t) |
| 173 | + source.capture_frame(frame) |
| 174 | + try: |
| 175 | + await asyncio.wait_for(stop.wait(), timeout=interval) |
| 176 | + except asyncio.TimeoutError: |
| 177 | + pass |
| 178 | + |
| 179 | + |
| 180 | +async def _wait_for_layer( |
| 181 | + stream: rtc.VideoStream, |
| 182 | + expected_w: int, |
| 183 | + expected_h: int, |
| 184 | + *, |
| 185 | + timeout: float = 20.0, |
| 186 | + samples: int = 5, |
| 187 | + tolerance: float = 0.20, |
| 188 | +) -> Tuple[int, int]: |
| 189 | + """Drain frames until we observe `samples` consecutive frames whose |
| 190 | + dimensions match the expected layer (within `tolerance`).""" |
| 191 | + deadline = asyncio.get_event_loop().time() + timeout |
| 192 | + matches = 0 |
| 193 | + last: Optional[Tuple[int, int]] = None |
| 194 | + iterator = stream.__aiter__() |
| 195 | + while asyncio.get_event_loop().time() < deadline: |
| 196 | + try: |
| 197 | + ev = await asyncio.wait_for(iterator.__anext__(), timeout=2.0) |
| 198 | + except asyncio.TimeoutError: |
| 199 | + continue |
| 200 | + except StopAsyncIteration: |
| 201 | + break |
| 202 | + w, h = ev.frame.width, ev.frame.height |
| 203 | + last = (w, h) |
| 204 | + if ( |
| 205 | + abs(w - expected_w) / expected_w <= tolerance |
| 206 | + and abs(h - expected_h) / expected_h <= tolerance |
| 207 | + ): |
| 208 | + matches += 1 |
| 209 | + if matches >= samples: |
| 210 | + return last |
| 211 | + else: |
| 212 | + matches = 0 |
| 213 | + raise AssertionError( |
| 214 | + f"timed out waiting for ~{expected_w}x{expected_h}, last seen={last}" |
| 215 | + ) |
| 216 | + |
| 217 | + |
| 218 | +@skip_if_no_credentials() |
| 219 | +@pytest.mark.asyncio |
| 220 | +@pytest.mark.parametrize( |
| 221 | + "video_codec, codec_name", |
| 222 | + [ |
| 223 | + (rtc.VideoCodec.VP8, "vp8"), |
| 224 | + (rtc.VideoCodec.H264, "h264"), |
| 225 | + ], |
| 226 | +) |
| 227 | +async def test_simulcast_quality_layers( |
| 228 | + video_codec: rtc.VideoCodec.ValueType, codec_name: str |
| 229 | +) -> None: |
| 230 | + room_name = unique_room_name(f"py-simulcast-{codec_name}") |
| 231 | + url = os.environ["LIVEKIT_URL"] |
| 232 | + |
| 233 | + sender, receiver = rtc.Room(), rtc.Room() |
| 234 | + await sender.connect(url, create_token("sender", room_name)) |
| 235 | + await receiver.connect(url, create_token("receiver", room_name)) |
| 236 | + await _ensure_all_connected([sender, receiver]) |
| 237 | + |
| 238 | + source = rtc.VideoSource(PUBLISH_WIDTH, PUBLISH_HEIGHT) |
| 239 | + track = rtc.LocalVideoTrack.create_video_track(f"simulcast-{codec_name}", source) |
| 240 | + options = rtc.TrackPublishOptions( |
| 241 | + source=rtc.TrackSource.SOURCE_CAMERA, |
| 242 | + simulcast=True, |
| 243 | + video_codec=video_codec, |
| 244 | + video_encoding=rtc.VideoEncoding(max_bitrate=3_000_000, max_framerate=PUBLISH_FPS), |
| 245 | + ) |
| 246 | + |
| 247 | + stop = asyncio.Event() |
| 248 | + pub_task = asyncio.create_task(_publish_loop(source, stop)) |
| 249 | + |
| 250 | + stream: Optional[rtc.VideoStream] = None |
| 251 | + try: |
| 252 | + track_published = _expect_event( |
| 253 | + receiver, |
| 254 | + "track_published", |
| 255 | + predicate=lambda pub, _p: pub.kind == rtc.TrackKind.KIND_VIDEO, |
| 256 | + ) |
| 257 | + local_pub = await sender.local_participant.publish_track(track, options) |
| 258 | + await _await_event(track_published) |
| 259 | + |
| 260 | + print( |
| 261 | + f"[{codec_name}] local_pub: sid={local_pub.sid} " |
| 262 | + f"simulcasted={local_pub.simulcasted} " |
| 263 | + f"mime_type={local_pub.mime_type} " |
| 264 | + f"{local_pub.width}x{local_pub.height}" |
| 265 | + ) |
| 266 | + remote_pub = await _ensure_track_subscribed(receiver, local_pub.sid) |
| 267 | + assert remote_pub.track is not None |
| 268 | + |
| 269 | + # Give the SFU a moment to propagate simulcast layer metadata and |
| 270 | + # let the encoder/bandwidth estimator ramp up to all layers before |
| 271 | + # we start switching qualities. |
| 272 | + await asyncio.sleep(5.0) |
| 273 | + print( |
| 274 | + f"[{codec_name}] remote_pub: sid={remote_pub.sid} " |
| 275 | + f"simulcasted={remote_pub.simulcasted} " |
| 276 | + f"mime_type={remote_pub.mime_type} " |
| 277 | + f"{remote_pub.width}x{remote_pub.height}" |
| 278 | + ) |
| 279 | + |
| 280 | + stream = rtc.VideoStream.from_track(track=remote_pub.track) |
| 281 | + |
| 282 | + for quality, layer in QUALITY_SEQUENCE: |
| 283 | + remote_pub.set_video_quality(quality) |
| 284 | + ew, eh = LAYER_DIMENSIONS[layer] |
| 285 | + actual = await _wait_for_layer(stream, ew, eh, timeout=20.0) |
| 286 | + print( |
| 287 | + f"[{codec_name}] layer={layer} expected~{ew}x{eh} got={actual[0]}x{actual[1]}" |
| 288 | + ) |
| 289 | + finally: |
| 290 | + stop.set() |
| 291 | + try: |
| 292 | + await pub_task |
| 293 | + except Exception: |
| 294 | + pass |
| 295 | + if stream is not None: |
| 296 | + await stream.aclose() |
| 297 | + await sender.disconnect() |
| 298 | + await receiver.disconnect() |
0 commit comments