yt_dlp/downloader/fragment.py

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