yt_dlp/downloader/fragment.py

   1 import contextlib
   2 import http.client
   3 import json
   4 import math
   5 import os
   6 import time
   7
   8 try:
   9     import concurrent.futures
  10     can_threaded_download = True
  11 except ImportError:
  12     can_threaded_download = False
  13
  14 from .common import FileDownloader
  15 from .http import HttpFD
  16 from ..aes import aes_cbc_decrypt_bytes, unpad_pkcs7
  17 from ..compat import compat_os_name, compat_struct_pack, compat_urllib_error
  18 from ..utils import (
  19     DownloadError,
  20     encodeFilename,
  21     error_to_compat_str,
  22     sanitized_Request,
  23     traverse_obj,
  24 )
  25
  26
  27 class HttpQuietDownloader(HttpFD):
  28     def to_screen(self, *args, **kargs):
  29         pass
  30
  31     def report_retry(self, err, count, retries):
  32         super().to_screen(
  33             f'[download] Got server HTTP error: {err}. Retrying (attempt {count} of {self.format_retries(retries)}) ...')
  34
  35
  36 class FragmentFD(FileDownloader):
  37     """
  38     A base file downloader class for fragmented media (e.g. f4m/m3u8 manifests).
  39
  40     Available options:
  41
  42     fragment_retries:   Number of times to retry a fragment for HTTP error (DASH
  43                         and hlsnative only)
  44     skip_unavailable_fragments:
  45                         Skip unavailable fragments (DASH and hlsnative only)
  46     keep_fragments:     Keep downloaded fragments on disk after downloading is
  47                         finished
  48     concurrent_fragment_downloads:  The number of threads to use for native hls and dash downloads
  49     _no_ytdl_file:      Don't use .ytdl file
  50
  51     For each incomplete fragment download yt-dlp keeps on disk a special
  52     bookkeeping file with download state and metadata (in future such files will
  53     be used for any incomplete download handled by yt-dlp). This file is
  54     used to properly handle resuming, check download file consistency and detect
  55     potential errors. The file has a .ytdl extension and represents a standard
  56     JSON file of the following format:
  57
  58     extractor:
  59         Dictionary of extractor related data. TBD.
  60
  61     downloader:
  62         Dictionary of downloader related data. May contain following data:
  63             current_fragment:
  64                 Dictionary with current (being downloaded) fragment data:
  65                 index:  0-based index of current fragment among all fragments
  66             fragment_count:
  67                 Total count of fragments
  68
  69     This feature is experimental and file format may change in future.
  70     """
  71
  72     def report_retry_fragment(self, err, frag_index, count, retries):
  73         self.to_screen(
  74             '\r[download] Got server HTTP error: %s. Retrying fragment %d (attempt %d of %s) ...'
  75             % (error_to_compat_str(err), frag_index, count, self.format_retries(retries)))
  76
  77     def report_skip_fragment(self, frag_index, err=None):
  78         err = f' {err};' if err else ''
  79         self.to_screen(f'[download]{err} Skipping fragment {frag_index:d} ...')
  80
  81     def _prepare_url(self, info_dict, url):
  82         headers = info_dict.get('http_headers')
  83         return sanitized_Request(url, None, headers) if headers else url
  84
  85     def _prepare_and_start_frag_download(self, ctx, info_dict):
  86         self._prepare_frag_download(ctx)
  87         self._start_frag_download(ctx, info_dict)
  88
  89     def __do_ytdl_file(self, ctx):
  90         return ctx['live'] is not True and ctx['tmpfilename'] != '-' and not self.params.get('_no_ytdl_file')
  91
  92     def _read_ytdl_file(self, ctx):
  93         assert 'ytdl_corrupt' not in ctx
  94         stream, _ = self.sanitize_open(self.ytdl_filename(ctx['filename']), 'r')
  95         try:
  96             ytdl_data = json.loads(stream.read())
  97             ctx['fragment_index'] = ytdl_data['downloader']['current_fragment']['index']
  98             if 'extra_state' in ytdl_data['downloader']:
  99                 ctx['extra_state'] = ytdl_data['downloader']['extra_state']
 100         except Exception:
 101             ctx['ytdl_corrupt'] = True
 102         finally:
 103             stream.close()
 104
 105     def _write_ytdl_file(self, ctx):
 106         frag_index_stream, _ = self.sanitize_open(self.ytdl_filename(ctx['filename']), 'w')
 107         try:
 108             downloader = {
 109                 'current_fragment': {
 110                     'index': ctx['fragment_index'],
 111                 },
 112             }
 113             if 'extra_state' in ctx:
 114                 downloader['extra_state'] = ctx['extra_state']
 115             if ctx.get('fragment_count') is not None:
 116                 downloader['fragment_count'] = ctx['fragment_count']
 117             frag_index_stream.write(json.dumps({'downloader': downloader}))
 118         finally:
 119             frag_index_stream.close()
 120
 121     def _download_fragment(self, ctx, frag_url, info_dict, headers=None, request_data=None):
 122         fragment_filename = '%s-Frag%d' % (ctx['tmpfilename'], ctx['fragment_index'])
 123         fragment_info_dict = {
 124             'url': frag_url,
 125             'http_headers': headers or info_dict.get('http_headers'),
 126             'request_data': request_data,
 127             'ctx_id': ctx.get('ctx_id'),
 128         }
 129         success = ctx['dl'].download(fragment_filename, fragment_info_dict)
 130         if not success:
 131             return False
 132         if fragment_info_dict.get('filetime'):
 133             ctx['fragment_filetime'] = fragment_info_dict.get('filetime')
 134         ctx['fragment_filename_sanitized'] = fragment_filename
 135         return True
 136
 137     def _read_fragment(self, ctx):
 138         if not ctx.get('fragment_filename_sanitized'):
 139             return None
 140         try:
 141             down, frag_sanitized = self.sanitize_open(ctx['fragment_filename_sanitized'], 'rb')
 142         except FileNotFoundError:
 143             if ctx.get('live'):
 144                 return None
 145             raise
 146         ctx['fragment_filename_sanitized'] = frag_sanitized
 147         frag_content = down.read()
 148         down.close()
 149         return frag_content
 150
 151     def _append_fragment(self, ctx, frag_content):
 152         try:
 153             ctx['dest_stream'].write(frag_content)
 154             ctx['dest_stream'].flush()
 155         finally:
 156             if self.__do_ytdl_file(ctx):
 157                 self._write_ytdl_file(ctx)
 158             if not self.params.get('keep_fragments', False):
 159                 self.try_remove(encodeFilename(ctx['fragment_filename_sanitized']))
 160             del ctx['fragment_filename_sanitized']
 161
 162     def _prepare_frag_download(self, ctx):
 163         if 'live' not in ctx:
 164             ctx['live'] = False
 165         if not ctx['live']:
 166             total_frags_str = '%d' % ctx['total_frags']
 167             ad_frags = ctx.get('ad_frags', 0)
 168             if ad_frags:
 169                 total_frags_str += ' (not including %d ad)' % ad_frags
 170         else:
 171             total_frags_str = 'unknown (live)'
 172         self.to_screen(f'[{self.FD_NAME}] Total fragments: {total_frags_str}')
 173         self.report_destination(ctx['filename'])
 174         dl = HttpQuietDownloader(
 175             self.ydl,
 176             {
 177                 'continuedl': self.params.get('continuedl', True),
 178                 'quiet': self.params.get('quiet'),
 179                 'noprogress': True,
 180                 'ratelimit': self.params.get('ratelimit'),
 181                 'retries': self.params.get('retries', 0),
 182                 'nopart': self.params.get('nopart', False),
 183                 'test': False,
 184             }
 185         )
 186         tmpfilename = self.temp_name(ctx['filename'])
 187         open_mode = 'wb'
 188         resume_len = 0
 189
 190         # Establish possible resume length
 191         if os.path.isfile(encodeFilename(tmpfilename)):
 192             open_mode = 'ab'
 193             resume_len = os.path.getsize(encodeFilename(tmpfilename))
 194
 195         # Should be initialized before ytdl file check
 196         ctx.update({
 197             'tmpfilename': tmpfilename,
 198             'fragment_index': 0,
 199         })
 200
 201         if self.__do_ytdl_file(ctx):
 202             if os.path.isfile(encodeFilename(self.ytdl_filename(ctx['filename']))):
 203                 self._read_ytdl_file(ctx)
 204                 is_corrupt = ctx.get('ytdl_corrupt') is True
 205                 is_inconsistent = ctx['fragment_index'] > 0 and resume_len == 0
 206                 if is_corrupt or is_inconsistent:
 207                     message = (
 208                         '.ytdl file is corrupt' if is_corrupt else
 209                         'Inconsistent state of incomplete fragment download')
 210                     self.report_warning(
 211                         '%s. Restarting from the beginning ...' % message)
 212                     ctx['fragment_index'] = resume_len = 0
 213                     if 'ytdl_corrupt' in ctx:
 214                         del ctx['ytdl_corrupt']
 215                     self._write_ytdl_file(ctx)
 216             else:
 217                 self._write_ytdl_file(ctx)
 218                 assert ctx['fragment_index'] == 0
 219
 220         dest_stream, tmpfilename = self.sanitize_open(tmpfilename, open_mode)
 221
 222         ctx.update({
 223             'dl': dl,
 224             'dest_stream': dest_stream,
 225             'tmpfilename': tmpfilename,
 226             # Total complete fragments downloaded so far in bytes
 227             'complete_frags_downloaded_bytes': resume_len,
 228         })
 229
 230     def _start_frag_download(self, ctx, info_dict):
 231         resume_len = ctx['complete_frags_downloaded_bytes']
 232         total_frags = ctx['total_frags']
 233         ctx_id = ctx.get('ctx_id')
 234         # This dict stores the download progress, it's updated by the progress
 235         # hook
 236         state = {
 237             'status': 'downloading',
 238             'downloaded_bytes': resume_len,
 239             'fragment_index': ctx['fragment_index'],
 240             'fragment_count': total_frags,
 241             'filename': ctx['filename'],
 242             'tmpfilename': ctx['tmpfilename'],
 243         }
 244
 245         start = time.time()
 246         ctx.update({
 247             'started': start,
 248             'fragment_started': start,
 249             # Amount of fragment's bytes downloaded by the time of the previous
 250             # frag progress hook invocation
 251             'prev_frag_downloaded_bytes': 0,
 252         })
 253
 254         def frag_progress_hook(s):
 255             if s['status'] not in ('downloading', 'finished'):
 256                 return
 257
 258             if ctx_id is not None and s.get('ctx_id') != ctx_id:
 259                 return
 260
 261             state['max_progress'] = ctx.get('max_progress')
 262             state['progress_idx'] = ctx.get('progress_idx')
 263
 264             time_now = time.time()
 265             state['elapsed'] = time_now - start
 266             frag_total_bytes = s.get('total_bytes') or 0
 267             s['fragment_info_dict'] = s.pop('info_dict', {})
 268             if not ctx['live']:
 269                 estimated_size = (
 270                     (ctx['complete_frags_downloaded_bytes'] + frag_total_bytes)
 271                     / (state['fragment_index'] + 1) * total_frags)
 272                 state['total_bytes_estimate'] = estimated_size
 273
 274             if s['status'] == 'finished':
 275                 state['fragment_index'] += 1
 276                 ctx['fragment_index'] = state['fragment_index']
 277                 state['downloaded_bytes'] += frag_total_bytes - ctx['prev_frag_downloaded_bytes']
 278                 ctx['complete_frags_downloaded_bytes'] = state['downloaded_bytes']
 279                 ctx['speed'] = state['speed'] = self.calc_speed(
 280                     ctx['fragment_started'], time_now, frag_total_bytes)
 281                 ctx['fragment_started'] = time.time()
 282                 ctx['prev_frag_downloaded_bytes'] = 0
 283             else:
 284                 frag_downloaded_bytes = s['downloaded_bytes']
 285                 state['downloaded_bytes'] += frag_downloaded_bytes - ctx['prev_frag_downloaded_bytes']
 286                 if not ctx['live']:
 287                     state['eta'] = self.calc_eta(
 288                         start, time_now, estimated_size - resume_len,
 289                         state['downloaded_bytes'] - resume_len)
 290                 ctx['speed'] = state['speed'] = self.calc_speed(
 291                     ctx['fragment_started'], time_now, frag_downloaded_bytes)
 292                 ctx['prev_frag_downloaded_bytes'] = frag_downloaded_bytes
 293             self._hook_progress(state, info_dict)
 294
 295         ctx['dl'].add_progress_hook(frag_progress_hook)
 296
 297         return start
 298
 299     def _finish_frag_download(self, ctx, info_dict):
 300         ctx['dest_stream'].close()
 301         if self.__do_ytdl_file(ctx):
 302             ytdl_filename = encodeFilename(self.ytdl_filename(ctx['filename']))
 303             if os.path.isfile(ytdl_filename):
 304                 self.try_remove(ytdl_filename)
 305         elapsed = time.time() - ctx['started']
 306
 307         if ctx['tmpfilename'] == '-':
 308             downloaded_bytes = ctx['complete_frags_downloaded_bytes']
 309         else:
 310             self.try_rename(ctx['tmpfilename'], ctx['filename'])
 311             if self.params.get('updatetime', True):
 312                 filetime = ctx.get('fragment_filetime')
 313                 if filetime:
 314                     with contextlib.suppress(Exception):
 315                         os.utime(ctx['filename'], (time.time(), filetime))
 316             downloaded_bytes = os.path.getsize(encodeFilename(ctx['filename']))
 317
 318         self._hook_progress({
 319             'downloaded_bytes': downloaded_bytes,
 320             'total_bytes': downloaded_bytes,
 321             'filename': ctx['filename'],
 322             'status': 'finished',
 323             'elapsed': elapsed,
 324             'ctx_id': ctx.get('ctx_id'),
 325             'max_progress': ctx.get('max_progress'),
 326             'progress_idx': ctx.get('progress_idx'),
 327         }, info_dict)
 328
 329     def _prepare_external_frag_download(self, ctx):
 330         if 'live' not in ctx:
 331             ctx['live'] = False
 332         if not ctx['live']:
 333             total_frags_str = '%d' % ctx['total_frags']
 334             ad_frags = ctx.get('ad_frags', 0)
 335             if ad_frags:
 336                 total_frags_str += ' (not including %d ad)' % ad_frags
 337         else:
 338             total_frags_str = 'unknown (live)'
 339         self.to_screen(f'[{self.FD_NAME}] Total fragments: {total_frags_str}')
 340
 341         tmpfilename = self.temp_name(ctx['filename'])
 342
 343         # Should be initialized before ytdl file check
 344         ctx.update({
 345             'tmpfilename': tmpfilename,
 346             'fragment_index': 0,
 347         })
 348
 349     def decrypter(self, info_dict):
 350         _key_cache = {}
 351
 352         def _get_key(url):
 353             if url not in _key_cache:
 354                 _key_cache[url] = self.ydl.urlopen(self._prepare_url(info_dict, url)).read()
 355             return _key_cache[url]
 356
 357         def decrypt_fragment(fragment, frag_content):
 358             decrypt_info = fragment.get('decrypt_info')
 359             if not decrypt_info or decrypt_info['METHOD'] != 'AES-128':
 360                 return frag_content
 361             iv = decrypt_info.get('IV') or compat_struct_pack('>8xq', fragment['media_sequence'])
 362             decrypt_info['KEY'] = decrypt_info.get('KEY') or _get_key(info_dict.get('_decryption_key_url') or decrypt_info['URI'])
 363             # Don't decrypt the content in tests since the data is explicitly truncated and it's not to a valid block
 364             # size (see https://github.com/ytdl-org/youtube-dl/pull/27660). Tests only care that the correct data downloaded,
 365             # not what it decrypts to.
 366             if self.params.get('test', False):
 367                 return frag_content
 368             return unpad_pkcs7(aes_cbc_decrypt_bytes(frag_content, decrypt_info['KEY'], iv))
 369
 370         return decrypt_fragment
 371
 372     def download_and_append_fragments_multiple(self, *args, pack_func=None, finish_func=None):
 373         '''
 374         @params (ctx1, fragments1, info_dict1), (ctx2, fragments2, info_dict2), ...
 375                 all args must be either tuple or list
 376         '''
 377         interrupt_trigger = [True]
 378         max_progress = len(args)
 379         if max_progress == 1:
 380             return self.download_and_append_fragments(*args[0], pack_func=pack_func, finish_func=finish_func)
 381         max_workers = self.params.get('concurrent_fragment_downloads', 1)
 382         if max_progress > 1:
 383             self._prepare_multiline_status(max_progress)
 384         is_live = any(traverse_obj(args, (..., 2, 'is_live'), default=[]))
 385
 386         def thread_func(idx, ctx, fragments, info_dict, tpe):
 387             ctx['max_progress'] = max_progress
 388             ctx['progress_idx'] = idx
 389             return self.download_and_append_fragments(
 390                 ctx, fragments, info_dict, pack_func=pack_func, finish_func=finish_func,
 391                 tpe=tpe, interrupt_trigger=interrupt_trigger)
 392
 393         class FTPE(concurrent.futures.ThreadPoolExecutor):
 394             # has to stop this or it's going to wait on the worker thread itself
 395             def __exit__(self, exc_type, exc_val, exc_tb):
 396                 pass
 397
 398         if compat_os_name == 'nt':
 399             def future_result(future):
 400                 while True:
 401                     try:
 402                         return future.result(0.1)
 403                     except KeyboardInterrupt:
 404                         raise
 405                     except concurrent.futures.TimeoutError:
 406                         continue
 407         else:
 408             def future_result(future):
 409                 return future.result()
 410
 411         def interrupt_trigger_iter(fg):
 412             for f in fg:
 413                 if not interrupt_trigger[0]:
 414                     break
 415                 yield f
 416
 417         spins = []
 418         for idx, (ctx, fragments, info_dict) in enumerate(args):
 419             tpe = FTPE(math.ceil(max_workers / max_progress))
 420             job = tpe.submit(thread_func, idx, ctx, interrupt_trigger_iter(fragments), info_dict, tpe)
 421             spins.append((tpe, job))
 422
 423         result = True
 424         for tpe, job in spins:
 425             try:
 426                 result = result and future_result(job)
 427             except KeyboardInterrupt:
 428                 interrupt_trigger[0] = False
 429             finally:
 430                 tpe.shutdown(wait=True)
 431         if not interrupt_trigger[0] and not is_live:
 432             raise KeyboardInterrupt()
 433         # we expect the user wants to stop and DO WANT the preceding postprocessors to run;
 434         # so returning a intermediate result here instead of KeyboardInterrupt on live
 435         return result
 436
 437     def download_and_append_fragments(
 438             self, ctx, fragments, info_dict, *, pack_func=None, finish_func=None,
 439             tpe=None, interrupt_trigger=None):
 440         if not interrupt_trigger:
 441             interrupt_trigger = (True, )
 442
 443         fragment_retries = self.params.get('fragment_retries', 0)
 444         is_fatal = (
 445             ((lambda _: False) if info_dict.get('is_live') else (lambda idx: idx == 0))
 446             if self.params.get('skip_unavailable_fragments', True) else (lambda _: True))
 447
 448         if not pack_func:
 449             pack_func = lambda frag_content, _: frag_content
 450
 451         def download_fragment(fragment, ctx):
 452             if not interrupt_trigger[0]:
 453                 return
 454
 455             frag_index = ctx['fragment_index'] = fragment['frag_index']
 456             ctx['last_error'] = None
 457             headers = info_dict.get('http_headers', {}).copy()
 458             byte_range = fragment.get('byte_range')
 459             if byte_range:
 460                 headers['Range'] = 'bytes=%d-%d' % (byte_range['start'], byte_range['end'] - 1)
 461
 462             # Never skip the first fragment
 463             fatal, count = is_fatal(fragment.get('index') or (frag_index - 1)), 0
 464             while count <= fragment_retries:
 465                 try:
 466                     if self._download_fragment(ctx, fragment['url'], info_dict, headers):
 467                         break
 468                     return
 469                 except (compat_urllib_error.HTTPError, http.client.IncompleteRead) as err:
 470                     # Unavailable (possibly temporary) fragments may be served.
 471                     # First we try to retry then either skip or abort.
 472                     # See https://github.com/ytdl-org/youtube-dl/issues/10165,
 473                     # https://github.com/ytdl-org/youtube-dl/issues/10448).
 474                     count += 1
 475                     ctx['last_error'] = err
 476                     if count <= fragment_retries:
 477                         self.report_retry_fragment(err, frag_index, count, fragment_retries)
 478                 except DownloadError:
 479                     # Don't retry fragment if error occurred during HTTP downloading
 480                     # itself since it has own retry settings
 481                     if not fatal:
 482                         break
 483                     raise
 484
 485             if count > fragment_retries and fatal:
 486                 ctx['dest_stream'].close()
 487                 self.report_error('Giving up after %s fragment retries' % fragment_retries)
 488
 489         def append_fragment(frag_content, frag_index, ctx):
 490             if frag_content:
 491                 self._append_fragment(ctx, pack_func(frag_content, frag_index))
 492             elif not is_fatal(frag_index - 1):
 493                 self.report_skip_fragment(frag_index, 'fragment not found')
 494             else:
 495                 ctx['dest_stream'].close()
 496                 self.report_error(f'fragment {frag_index} not found, unable to continue')
 497                 return False
 498             return True
 499
 500         decrypt_fragment = self.decrypter(info_dict)
 501
 502         max_workers = math.ceil(
 503             self.params.get('concurrent_fragment_downloads', 1) / ctx.get('max_progress', 1))
 504         if can_threaded_download and max_workers > 1:
 505
 506             def _download_fragment(fragment):
 507                 ctx_copy = ctx.copy()
 508                 download_fragment(fragment, ctx_copy)
 509                 return fragment, fragment['frag_index'], ctx_copy.get('fragment_filename_sanitized')
 510
 511             self.report_warning('The download speed shown is only of one thread. This is a known issue and patches are welcome')
 512             with tpe or concurrent.futures.ThreadPoolExecutor(max_workers) as pool:
 513                 for fragment, frag_index, frag_filename in pool.map(_download_fragment, fragments):
 514                     ctx['fragment_filename_sanitized'] = frag_filename
 515                     ctx['fragment_index'] = frag_index
 516                     result = append_fragment(decrypt_fragment(fragment, self._read_fragment(ctx)), frag_index, ctx)
 517                     if not result:
 518                         return False
 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         self._finish_frag_download(ctx, info_dict)
 538         return True