yt_dlp/downloader/fragment.py

   1 import concurrent.futures
   2 import contextlib
   3 import json
   4 import math
   5 import os
   6 import struct
   7 import time
   8
   9 from .common import FileDownloader
  10 from .http import HttpFD
  11 from ..aes import aes_cbc_decrypt_bytes, unpad_pkcs7
  12 from ..compat import compat_os_name
  13 from ..networking import Request
  14 from ..networking.exceptions import HTTPError, IncompleteRead
  15 from ..utils import DownloadError, RetryManager, encodeFilename, traverse_obj
  16 from ..utils.networking import HTTPHeaderDict
  17
  18
  19 class HttpQuietDownloader(HttpFD):
  20     def to_screen(self, *args, **kargs):
  21         pass
  22
  23     to_console_title = to_screen
  24
  25
  26 class FragmentFD(FileDownloader):
  27     """
  28     A base file downloader class for fragmented media (e.g. f4m/m3u8 manifests).
  29
  30     Available options:
  31
  32     fragment_retries:   Number of times to retry a fragment for HTTP error
  33                         (DASH and hlsnative only). Default is 0 for API, but 10 for CLI
  34     skip_unavailable_fragments:
  35                         Skip unavailable fragments (DASH and hlsnative only)
  36     keep_fragments:     Keep downloaded fragments on disk after downloading is
  37                         finished
  38     concurrent_fragment_downloads:  The number of threads to use for native hls and dash downloads
  39     _no_ytdl_file:      Don't use .ytdl file
  40
  41     For each incomplete fragment download yt-dlp keeps on disk a special
  42     bookkeeping file with download state and metadata (in future such files will
  43     be used for any incomplete download handled by yt-dlp). This file is
  44     used to properly handle resuming, check download file consistency and detect
  45     potential errors. The file has a .ytdl extension and represents a standard
  46     JSON file of the following format:
  47
  48     extractor:
  49         Dictionary of extractor related data. TBD.
  50
  51     downloader:
  52         Dictionary of downloader related data. May contain following data:
  53             current_fragment:
  54                 Dictionary with current (being downloaded) fragment data:
  55                 index:  0-based index of current fragment among all fragments
  56             fragment_count:
  57                 Total count of fragments
  58
  59     This feature is experimental and file format may change in future.
  60     """
  61
  62     def report_retry_fragment(self, err, frag_index, count, retries):
  63         self.deprecation_warning('yt_dlp.downloader.FragmentFD.report_retry_fragment is deprecated. '
  64                                  'Use yt_dlp.downloader.FileDownloader.report_retry instead')
  65         return self.report_retry(err, count, retries, frag_index)
  66
  67     def report_skip_fragment(self, frag_index, err=None):
  68         err = f' {err};' if err else ''
  69         self.to_screen(f'[download]{err} Skipping fragment {frag_index:d} ...')
  70
  71     def _prepare_url(self, info_dict, url):
  72         headers = info_dict.get('http_headers')
  73         return Request(url, None, headers) if headers else url
  74
  75     def _prepare_and_start_frag_download(self, ctx, info_dict):
  76         self._prepare_frag_download(ctx)
  77         self._start_frag_download(ctx, info_dict)
  78
  79     def __do_ytdl_file(self, ctx):
  80         return ctx['live'] is not True and ctx['tmpfilename'] != '-' and not self.params.get('_no_ytdl_file')
  81
  82     def _read_ytdl_file(self, ctx):
  83         assert 'ytdl_corrupt' not in ctx
  84         stream, _ = self.sanitize_open(self.ytdl_filename(ctx['filename']), 'r')
  85         try:
  86             ytdl_data = json.loads(stream.read())
  87             ctx['fragment_index'] = ytdl_data['downloader']['current_fragment']['index']
  88             if 'extra_state' in ytdl_data['downloader']:
  89                 ctx['extra_state'] = ytdl_data['downloader']['extra_state']
  90         except Exception:
  91             ctx['ytdl_corrupt'] = True
  92         finally:
  93             stream.close()
  94
  95     def _write_ytdl_file(self, ctx):
  96         frag_index_stream, _ = self.sanitize_open(self.ytdl_filename(ctx['filename']), 'w')
  97         try:
  98             downloader = {
  99                 'current_fragment': {
 100                     'index': ctx['fragment_index'],
 101                 },
 102             }
 103             if 'extra_state' in ctx:
 104                 downloader['extra_state'] = ctx['extra_state']
 105             if ctx.get('fragment_count') is not None:
 106                 downloader['fragment_count'] = ctx['fragment_count']
 107             frag_index_stream.write(json.dumps({'downloader': downloader}))
 108         finally:
 109             frag_index_stream.close()
 110
 111     def _download_fragment(self, ctx, frag_url, info_dict, headers=None, request_data=None):
 112         fragment_filename = '%s-Frag%d' % (ctx['tmpfilename'], ctx['fragment_index'])
 113         fragment_info_dict = {
 114             'url': frag_url,
 115             'http_headers': headers or info_dict.get('http_headers'),
 116             'request_data': request_data,
 117             'ctx_id': ctx.get('ctx_id'),
 118         }
 119         frag_resume_len = 0
 120         if ctx['dl'].params.get('continuedl', True):
 121             frag_resume_len = self.filesize_or_none(self.temp_name(fragment_filename))
 122         fragment_info_dict['frag_resume_len'] = ctx['frag_resume_len'] = frag_resume_len
 123
 124         success, _ = ctx['dl'].download(fragment_filename, fragment_info_dict)
 125         if not success:
 126             return False
 127         if fragment_info_dict.get('filetime'):
 128             ctx['fragment_filetime'] = fragment_info_dict.get('filetime')
 129         ctx['fragment_filename_sanitized'] = fragment_filename
 130         return True
 131
 132     def _read_fragment(self, ctx):
 133         if not ctx.get('fragment_filename_sanitized'):
 134             return None
 135         try:
 136             down, frag_sanitized = self.sanitize_open(ctx['fragment_filename_sanitized'], 'rb')
 137         except FileNotFoundError:
 138             if ctx.get('live'):
 139                 return None
 140             raise
 141         ctx['fragment_filename_sanitized'] = frag_sanitized
 142         frag_content = down.read()
 143         down.close()
 144         return frag_content
 145
 146     def _append_fragment(self, ctx, frag_content):
 147         try:
 148             ctx['dest_stream'].write(frag_content)
 149             ctx['dest_stream'].flush()
 150         finally:
 151             if self.__do_ytdl_file(ctx):
 152                 self._write_ytdl_file(ctx)
 153             if not self.params.get('keep_fragments', False):
 154                 self.try_remove(encodeFilename(ctx['fragment_filename_sanitized']))
 155             del ctx['fragment_filename_sanitized']
 156
 157     def _prepare_frag_download(self, ctx):
 158         if not ctx.setdefault('live', False):
 159             total_frags_str = '%d' % ctx['total_frags']
 160             ad_frags = ctx.get('ad_frags', 0)
 161             if ad_frags:
 162                 total_frags_str += ' (not including %d ad)' % ad_frags
 163         else:
 164             total_frags_str = 'unknown (live)'
 165         self.to_screen(f'[{self.FD_NAME}] Total fragments: {total_frags_str}')
 166         self.report_destination(ctx['filename'])
 167         dl = HttpQuietDownloader(self.ydl, {
 168             **self.params,
 169             'noprogress': True,
 170             'test': False,
 171             'sleep_interval': 0,
 172             'max_sleep_interval': 0,
 173             'sleep_interval_subtitles': 0,
 174         })
 175         tmpfilename = self.temp_name(ctx['filename'])
 176         open_mode = 'wb'
 177
 178         # Establish possible resume length
 179         resume_len = self.filesize_or_none(tmpfilename)
 180         if resume_len > 0:
 181             open_mode = 'ab'
 182
 183         # Should be initialized before ytdl file check
 184         ctx.update({
 185             'tmpfilename': tmpfilename,
 186             'fragment_index': 0,
 187         })
 188
 189         if self.__do_ytdl_file(ctx):
 190             ytdl_file_exists = os.path.isfile(encodeFilename(self.ytdl_filename(ctx['filename'])))
 191             continuedl = self.params.get('continuedl', True)
 192             if continuedl and ytdl_file_exists:
 193                 self._read_ytdl_file(ctx)
 194                 is_corrupt = ctx.get('ytdl_corrupt') is True
 195                 is_inconsistent = ctx['fragment_index'] > 0 and resume_len == 0
 196                 if is_corrupt or is_inconsistent:
 197                     message = (
 198                         '.ytdl file is corrupt' if is_corrupt else
 199                         'Inconsistent state of incomplete fragment download')
 200                     self.report_warning(
 201                         '%s. Restarting from the beginning ...' % message)
 202                     ctx['fragment_index'] = resume_len = 0
 203                     if 'ytdl_corrupt' in ctx:
 204                         del ctx['ytdl_corrupt']
 205                     self._write_ytdl_file(ctx)
 206
 207             else:
 208                 if not continuedl:
 209                     if ytdl_file_exists:
 210                         self._read_ytdl_file(ctx)
 211                     ctx['fragment_index'] = resume_len = 0
 212                 self._write_ytdl_file(ctx)
 213                 assert ctx['fragment_index'] == 0
 214
 215         dest_stream, tmpfilename = self.sanitize_open(tmpfilename, open_mode)
 216
 217         ctx.update({
 218             'dl': dl,
 219             'dest_stream': dest_stream,
 220             'tmpfilename': tmpfilename,
 221             # Total complete fragments downloaded so far in bytes
 222             'complete_frags_downloaded_bytes': resume_len,
 223         })
 224
 225     def _start_frag_download(self, ctx, info_dict):
 226         resume_len = ctx['complete_frags_downloaded_bytes']
 227         total_frags = ctx['total_frags']
 228         ctx_id = ctx.get('ctx_id')
 229         # This dict stores the download progress, it's updated by the progress
 230         # hook
 231         state = {
 232             'status': 'downloading',
 233             'downloaded_bytes': resume_len,
 234             'fragment_index': ctx['fragment_index'],
 235             'fragment_count': total_frags,
 236             'filename': ctx['filename'],
 237             'tmpfilename': ctx['tmpfilename'],
 238         }
 239
 240         start = time.time()
 241         ctx.update({
 242             'started': start,
 243             'fragment_started': start,
 244             # Amount of fragment's bytes downloaded by the time of the previous
 245             # frag progress hook invocation
 246             'prev_frag_downloaded_bytes': 0,
 247         })
 248
 249         def frag_progress_hook(s):
 250             if s['status'] not in ('downloading', 'finished'):
 251                 return
 252
 253             if not total_frags and ctx.get('fragment_count'):
 254                 state['fragment_count'] = ctx['fragment_count']
 255
 256             if ctx_id is not None and s.get('ctx_id') != ctx_id:
 257                 return
 258
 259             state['max_progress'] = ctx.get('max_progress')
 260             state['progress_idx'] = ctx.get('progress_idx')
 261
 262             time_now = time.time()
 263             state['elapsed'] = time_now - start
 264             frag_total_bytes = s.get('total_bytes') or 0
 265             s['fragment_info_dict'] = s.pop('info_dict', {})
 266             if not ctx['live']:
 267                 estimated_size = (
 268                     (ctx['complete_frags_downloaded_bytes'] + frag_total_bytes)
 269                     / (state['fragment_index'] + 1) * total_frags)
 270                 state['total_bytes_estimate'] = estimated_size
 271
 272             if s['status'] == 'finished':
 273                 state['fragment_index'] += 1
 274                 ctx['fragment_index'] = state['fragment_index']
 275                 state['downloaded_bytes'] += frag_total_bytes - ctx['prev_frag_downloaded_bytes']
 276                 ctx['complete_frags_downloaded_bytes'] = state['downloaded_bytes']
 277                 ctx['speed'] = state['speed'] = self.calc_speed(
 278                     ctx['fragment_started'], time_now, frag_total_bytes)
 279                 ctx['fragment_started'] = time.time()
 280                 ctx['prev_frag_downloaded_bytes'] = 0
 281             else:
 282                 frag_downloaded_bytes = s['downloaded_bytes']
 283                 state['downloaded_bytes'] += frag_downloaded_bytes - ctx['prev_frag_downloaded_bytes']
 284                 ctx['speed'] = state['speed'] = self.calc_speed(
 285                     ctx['fragment_started'], time_now, frag_downloaded_bytes - ctx.get('frag_resume_len', 0))
 286                 if not ctx['live']:
 287                     state['eta'] = self.calc_eta(state['speed'], estimated_size - state['downloaded_bytes'])
 288                 ctx['prev_frag_downloaded_bytes'] = frag_downloaded_bytes
 289             self._hook_progress(state, info_dict)
 290
 291         ctx['dl'].add_progress_hook(frag_progress_hook)
 292
 293         return start
 294
 295     def _finish_frag_download(self, ctx, info_dict):
 296         ctx['dest_stream'].close()
 297         if self.__do_ytdl_file(ctx):
 298             self.try_remove(self.ytdl_filename(ctx['filename']))
 299         elapsed = time.time() - ctx['started']
 300
 301         to_file = ctx['tmpfilename'] != '-'
 302         if to_file:
 303             downloaded_bytes = self.filesize_or_none(ctx['tmpfilename'])
 304         else:
 305             downloaded_bytes = ctx['complete_frags_downloaded_bytes']
 306
 307         if not downloaded_bytes:
 308             if to_file:
 309                 self.try_remove(ctx['tmpfilename'])
 310             self.report_error('The downloaded file is empty')
 311             return False
 312         elif to_file:
 313             self.try_rename(ctx['tmpfilename'], ctx['filename'])
 314             filetime = ctx.get('fragment_filetime')
 315             if self.params.get('updatetime', True) and filetime:
 316                 with contextlib.suppress(Exception):
 317                     os.utime(ctx['filename'], (time.time(), filetime))
 318
 319         self._hook_progress({
 320             'downloaded_bytes': downloaded_bytes,
 321             'total_bytes': downloaded_bytes,
 322             'filename': ctx['filename'],
 323             'status': 'finished',
 324             'elapsed': elapsed,
 325             'ctx_id': ctx.get('ctx_id'),
 326             'max_progress': ctx.get('max_progress'),
 327             'progress_idx': ctx.get('progress_idx'),
 328         }, info_dict)
 329         return True
 330
 331     def _prepare_external_frag_download(self, ctx):
 332         if 'live' not in ctx:
 333             ctx['live'] = False
 334         if not ctx['live']:
 335             total_frags_str = '%d' % ctx['total_frags']
 336             ad_frags = ctx.get('ad_frags', 0)
 337             if ad_frags:
 338                 total_frags_str += ' (not including %d ad)' % ad_frags
 339         else:
 340             total_frags_str = 'unknown (live)'
 341         self.to_screen(f'[{self.FD_NAME}] Total fragments: {total_frags_str}')
 342
 343         tmpfilename = self.temp_name(ctx['filename'])
 344
 345         # Should be initialized before ytdl file check
 346         ctx.update({
 347             'tmpfilename': tmpfilename,
 348             'fragment_index': 0,
 349         })
 350
 351     def decrypter(self, info_dict):
 352         _key_cache = {}
 353
 354         def _get_key(url):
 355             if url not in _key_cache:
 356                 _key_cache[url] = self.ydl.urlopen(self._prepare_url(info_dict, url)).read()
 357             return _key_cache[url]
 358
 359         def decrypt_fragment(fragment, frag_content):
 360             if frag_content is None:
 361                 return
 362             decrypt_info = fragment.get('decrypt_info')
 363             if not decrypt_info or decrypt_info['METHOD'] != 'AES-128':
 364                 return frag_content
 365             iv = decrypt_info.get('IV') or struct.pack('>8xq', fragment['media_sequence'])
 366             decrypt_info['KEY'] = (decrypt_info.get('KEY')
 367                                    or _get_key(traverse_obj(info_dict, ('hls_aes', 'uri')) or decrypt_info['URI']))
 368             # Don't decrypt the content in tests since the data is explicitly truncated and it's not to a valid block
 369             # size (see https://github.com/ytdl-org/youtube-dl/pull/27660). Tests only care that the correct data downloaded,
 370             # not what it decrypts to.
 371             if self.params.get('test', False):
 372                 return frag_content
 373             return unpad_pkcs7(aes_cbc_decrypt_bytes(frag_content, decrypt_info['KEY'], iv))
 374
 375         return decrypt_fragment
 376
 377     def download_and_append_fragments_multiple(self, *args, **kwargs):
 378         '''
 379         @params (ctx1, fragments1, info_dict1), (ctx2, fragments2, info_dict2), ...
 380                 all args must be either tuple or list
 381         '''
 382         interrupt_trigger = [True]
 383         max_progress = len(args)
 384         if max_progress == 1:
 385             return self.download_and_append_fragments(*args[0], **kwargs)
 386         max_workers = self.params.get('concurrent_fragment_downloads', 1)
 387         if max_progress > 1:
 388             self._prepare_multiline_status(max_progress)
 389         is_live = any(traverse_obj(args, (..., 2, 'is_live')))
 390
 391         def thread_func(idx, ctx, fragments, info_dict, tpe):
 392             ctx['max_progress'] = max_progress
 393             ctx['progress_idx'] = idx
 394             return self.download_and_append_fragments(
 395                 ctx, fragments, info_dict, **kwargs, tpe=tpe, interrupt_trigger=interrupt_trigger)
 396
 397         class FTPE(concurrent.futures.ThreadPoolExecutor):
 398             # has to stop this or it's going to wait on the worker thread itself
 399             def __exit__(self, exc_type, exc_val, exc_tb):
 400                 pass
 401
 402         if compat_os_name == 'nt':
 403             def future_result(future):
 404                 while True:
 405                     try:
 406                         return future.result(0.1)
 407                     except KeyboardInterrupt:
 408                         raise
 409                     except concurrent.futures.TimeoutError:
 410                         continue
 411         else:
 412             def future_result(future):
 413                 return future.result()
 414
 415         def interrupt_trigger_iter(fg):
 416             for f in fg:
 417                 if not interrupt_trigger[0]:
 418                     break
 419                 yield f
 420
 421         spins = []
 422         for idx, (ctx, fragments, info_dict) in enumerate(args):
 423             tpe = FTPE(math.ceil(max_workers / max_progress))
 424             job = tpe.submit(thread_func, idx, ctx, interrupt_trigger_iter(fragments), info_dict, tpe)
 425             spins.append((tpe, job))
 426
 427         result = True
 428         for tpe, job in spins:
 429             try:
 430                 result = result and future_result(job)
 431             except KeyboardInterrupt:
 432                 interrupt_trigger[0] = False
 433             finally:
 434                 tpe.shutdown(wait=True)
 435         if not interrupt_trigger[0] and not is_live:
 436             raise KeyboardInterrupt()
 437         # we expect the user wants to stop and DO WANT the preceding postprocessors to run;
 438         # so returning a intermediate result here instead of KeyboardInterrupt on live
 439         return result
 440
 441     def download_and_append_fragments(
 442             self, ctx, fragments, info_dict, *, is_fatal=(lambda idx: False),
 443             pack_func=(lambda content, idx: content), finish_func=None,
 444             tpe=None, interrupt_trigger=(True, )):
 445
 446         if not self.params.get('skip_unavailable_fragments', True):
 447             is_fatal = lambda _: True
 448
 449         def download_fragment(fragment, ctx):
 450             if not interrupt_trigger[0]:
 451                 return
 452
 453             frag_index = ctx['fragment_index'] = fragment['frag_index']
 454             ctx['last_error'] = None
 455             headers = HTTPHeaderDict(info_dict.get('http_headers'))
 456             byte_range = fragment.get('byte_range')
 457             if byte_range:
 458                 headers['Range'] = 'bytes=%d-%d' % (byte_range['start'], byte_range['end'] - 1)
 459
 460             # Never skip the first fragment
 461             fatal = is_fatal(fragment.get('index') or (frag_index - 1))
 462
 463             def error_callback(err, count, retries):
 464                 if fatal and count > retries:
 465                     ctx['dest_stream'].close()
 466                 self.report_retry(err, count, retries, frag_index, fatal)
 467                 ctx['last_error'] = err
 468
 469             for retry in RetryManager(self.params.get('fragment_retries'), error_callback):
 470                 try:
 471                     ctx['fragment_count'] = fragment.get('fragment_count')
 472                     if not self._download_fragment(
 473                             ctx, fragment['url'], info_dict, headers, info_dict.get('request_data')):
 474                         return
 475                 except (HTTPError, IncompleteRead) as err:
 476                     retry.error = err
 477                     continue
 478                 except DownloadError:  # has own retry settings
 479                     if fatal:
 480                         raise
 481
 482         def append_fragment(frag_content, frag_index, ctx):
 483             if frag_content:
 484                 self._append_fragment(ctx, pack_func(frag_content, frag_index))
 485             elif not is_fatal(frag_index - 1):
 486                 self.report_skip_fragment(frag_index, 'fragment not found')
 487             else:
 488                 ctx['dest_stream'].close()
 489                 self.report_error(f'fragment {frag_index} not found, unable to continue')
 490                 return False
 491             return True
 492
 493         decrypt_fragment = self.decrypter(info_dict)
 494
 495         max_workers = math.ceil(
 496             self.params.get('concurrent_fragment_downloads', 1) / ctx.get('max_progress', 1))
 497         if max_workers > 1:
 498             def _download_fragment(fragment):
 499                 ctx_copy = ctx.copy()
 500                 download_fragment(fragment, ctx_copy)
 501                 return fragment, fragment['frag_index'], ctx_copy.get('fragment_filename_sanitized')
 502
 503             self.report_warning('The download speed shown is only of one thread. This is a known issue')
 504             with tpe or concurrent.futures.ThreadPoolExecutor(max_workers) as pool:
 505                 try:
 506                     for fragment, frag_index, frag_filename in pool.map(_download_fragment, fragments):
 507                         ctx.update({
 508                             'fragment_filename_sanitized': frag_filename,
 509                             'fragment_index': frag_index,
 510                         })
 511                         if not append_fragment(decrypt_fragment(fragment, self._read_fragment(ctx)), frag_index, ctx):
 512                             return False
 513                 except KeyboardInterrupt:
 514                     self._finish_multiline_status()
 515                     self.report_error(
 516                         'Interrupted by user. Waiting for all threads to shutdown...', is_error=False, tb=False)
 517                     pool.shutdown(wait=False)
 518                     raise
 519         else:
 520             for fragment in fragments:
 521                 if not interrupt_trigger[0]:
 522                     break
 523                 try:
 524                     download_fragment(fragment, ctx)
 525                     result = append_fragment(
 526                         decrypt_fragment(fragment, self._read_fragment(ctx)), fragment['frag_index'], ctx)
 527                 except KeyboardInterrupt:
 528                     if info_dict.get('is_live'):
 529                         break
 530                     raise
 531                 if not result:
 532                     return False
 533
 534         if finish_func is not None:
 535             ctx['dest_stream'].write(finish_func())
 536             ctx['dest_stream'].flush()
 537         return self._finish_frag_download(ctx, info_dict)