Native concurrent downloading of fragments (#166)

* Option `--concurrent-fragments` (`-N`) to set the number of threads

Related: #165

Known issues:
* When receiving Ctrl+C, the process will exit only after finishing the currently downloading fragments
* The download progress shows the speed of only one thread

Authored by shirt-dev

.gitignore

@@ -60,6 +60,7 @@ yt-dlp.zip
 *.mkv
 *.swf
 *.part
+*.part-*
 *.ytdl
 *.dump
 *.frag

README.md

@@ -297,6 +297,8 @@ Then simply run `make`. You can also run `make yt-dlp` instead to compile only t
     --no-include-ads                 Do not download advertisements (default)
 
 ## Download Options:
+    -N, --concurrent-fragments N     Number of fragments to download
+                                     concurrently (default is 1)
     -r, --limit-rate RATE            Maximum download rate in bytes per second
                                      (e.g. 50K or 4.2M)
     -R, --retries RETRIES            Number of retries (default is 10), or

yt_dlp/__init__.py

@@ -180,6 +180,8 @@ def _real_main(argv=None):
     if opts.overwrites:
         # --yes-overwrites implies --no-continue
         opts.continue_dl = False
+    if opts.concurrent_fragment_downloads <= 0:
+        raise ValueError('Concurrent fragments must be positive')
 
     def parse_retries(retries, name=''):
         if retries in ('inf', 'infinite'):
@@ -463,6 +465,7 @@ def _real_main(argv=None):
         'extractor_retries': opts.extractor_retries,
         'skip_unavailable_fragments': opts.skip_unavailable_fragments,
         'keep_fragments': opts.keep_fragments,
+        'concurrent_fragment_downloads': opts.concurrent_fragment_downloads,
         'buffersize': opts.buffersize,
         'noresizebuffer': opts.noresizebuffer,
         'http_chunk_size': opts.http_chunk_size,

yt_dlp/downloader/dash.py

@@ -1,11 +1,18 @@
 from __future__ import unicode_literals
 
+try:
+    import concurrent.futures
+    can_threaded_download = True
+except ImportError:
+    can_threaded_download = False
+
 from ..downloader import _get_real_downloader
 from .fragment import FragmentFD
 
 from ..compat import compat_urllib_error
 from ..utils import (
     DownloadError,
+    sanitize_open,
     urljoin,
 )
 
@@ -49,11 +56,29 @@ class DashSegmentsFD(FragmentFD):
                 assert fragment_base_url
                 fragment_url = urljoin(fragment_base_url, fragment['path'])
 
-            if real_downloader:
             fragments_to_download.append({
+                'frag_index': frag_index,
+                'index': i,
                 'url': fragment_url,
             })
-                continue
+
+        if real_downloader:
+            info_copy = info_dict.copy()
+            info_copy['fragments'] = fragments_to_download
+            fd = real_downloader(self.ydl, self.params)
+            # TODO: Make progress updates work without hooking twice
+            # for ph in self._progress_hooks:
+            #     fd.add_progress_hook(ph)
+            success = fd.real_download(filename, info_copy)
+            if not success:
+                return False
+        else:
+            def download_fragment(fragment):
+                i = fragment['index']
+                frag_index = fragment['frag_index']
+                fragment_url = fragment['url']
+
+                ctx['fragment_index'] = frag_index
 
                 # In DASH, the first segment contains necessary headers to
                 # generate a valid MP4 file, so always abort for the first segment
@@ -63,8 +88,7 @@ class DashSegmentsFD(FragmentFD):
                     try:
                         success, frag_content = self._download_fragment(ctx, fragment_url, info_dict)
                         if not success:
-                        return False
+                            return False, frag_index
-                    self._append_fragment(ctx, frag_content)
                         break
                     except compat_urllib_error.HTTPError as err:
                         # YouTube may often return 404 HTTP error for a fragment causing the
@@ -80,27 +104,73 @@ class DashSegmentsFD(FragmentFD):
                         # Don't retry fragment if error occurred during HTTP downloading
                         # itself since it has own retry settings
                         if not fatal:
-                        self.report_skip_fragment(frag_index)
                             break
                         raise
 
                 if count > fragment_retries:
                     if not fatal:
-                    self.report_skip_fragment(frag_index)
+                        return False, frag_index
-                    continue
                     self.report_error('giving up after %s fragment retries' % fragment_retries)
+                    return False, frag_index
+
+                return frag_content, frag_index
+
+            def append_fragment(frag_content, frag_index):
+                if frag_content:
+                    fragment_filename = '%s-Frag%d' % (ctx['tmpfilename'], frag_index)
+                    try:
+                        file, frag_sanitized = sanitize_open(fragment_filename, 'rb')
+                        ctx['fragment_filename_sanitized'] = frag_sanitized
+                        file.close()
+                        self._append_fragment(ctx, frag_content)
+                        return True
+                    except FileNotFoundError:
+                        if skip_unavailable_fragments:
+                            self.report_skip_fragment(frag_index)
+                            return True
+                        else:
+                            self.report_error(
+                                'fragment %s not found, unable to continue' % frag_index)
+                            return False
+                else:
+                    if skip_unavailable_fragments:
+                        self.report_skip_fragment(frag_index)
+                        return True
+                    else:
+                        self.report_error(
+                            'fragment %s not found, unable to continue' % frag_index)
                         return False
 
-        if real_downloader:
+            max_workers = self.params.get('concurrent_fragment_downloads', 1)
-            info_copy = info_dict.copy()
+            if can_threaded_download and max_workers > 1:
-            info_copy['fragments'] = fragments_to_download
+                self.report_warning('The download speed shown is only of one thread. This is a known issue')
-            fd = real_downloader(self.ydl, self.params)
+                with concurrent.futures.ThreadPoolExecutor(max_workers) as pool:
-            # TODO: Make progress updates work without hooking twice
+                    futures = [pool.submit(download_fragment, fragment) for fragment in fragments_to_download]
-            # for ph in self._progress_hooks:
+                    # timeout must be 0 to return instantly
-            #     fd.add_progress_hook(ph)
+                    done, not_done = concurrent.futures.wait(futures, timeout=0)
-            success = fd.real_download(filename, info_copy)
+                    try:
-            if not success:
+                        while not_done:
+                            # Check every 1 second for KeyboardInterrupt
+                            freshly_done, not_done = concurrent.futures.wait(not_done, timeout=1)
+                            done |= freshly_done
+                    except KeyboardInterrupt:
+                        for future in not_done:
+                            future.cancel()
+                        # timeout must be none to cancel
+                        concurrent.futures.wait(not_done, timeout=None)
+                        raise KeyboardInterrupt
+                results = [future.result() for future in futures]
+
+                for frag_content, frag_index in results:
+                    result = append_fragment(frag_content, frag_index)
+                    if not result:
                         return False
             else:
+                for fragment in fragments_to_download:
+                    frag_content, frag_index = download_fragment(fragment)
+                    result = append_fragment(frag_content, frag_index)
+                    if not result:
+                        return False
+
             self._finish_frag_download(ctx)
         return True

yt_dlp/downloader/external.py

@@ -126,7 +126,7 @@ class ExternalFD(FileDownloader):
             file_list = []
             dest, _ = sanitize_open(tmpfilename, 'wb')
             for i, fragment in enumerate(info_dict['fragments']):
-                file = '%s_%s.frag' % (tmpfilename, i)
+                file = '%s-Frag%d' % (tmpfilename, i)
                 decrypt_info = fragment.get('decrypt_info')
                 src, _ = sanitize_open(file, 'rb')
                 if decrypt_info:
@@ -274,7 +274,7 @@ class Aria2cFD(ExternalFD):
             url_list_file = '%s.frag.urls' % tmpfilename
             url_list = []
             for i, fragment in enumerate(info_dict['fragments']):
-                tmpsegmentname = '%s_%s.frag' % (os.path.basename(tmpfilename), i)
+                tmpsegmentname = '%s-Frag%d' % (os.path.basename(tmpfilename), i)
                 url_list.append('%s\n\tout=%s' % (fragment['url'], tmpsegmentname))
             stream, _ = sanitize_open(url_list_file, 'wb')
             stream.write('\n'.join(url_list).encode('utf-8'))

yt_dlp/downloader/hls.py

@@ -7,6 +7,11 @@ try:
     can_decrypt_frag = True
 except ImportError:
     can_decrypt_frag = False
+try:
+    import concurrent.futures
+    can_threaded_download = True
+except ImportError:
+    can_threaded_download = False
 
 from ..downloader import _get_real_downloader
 from .fragment import FragmentFD
@@ -19,6 +24,7 @@ from ..compat import (
 )
 from ..utils import (
     parse_m3u8_attributes,
+    sanitize_open,
     update_url_query,
 )
 
@@ -151,7 +157,6 @@ class HlsFD(FragmentFD):
         ad_frag_next = False
         for line in s.splitlines():
             line = line.strip()
-            download_frag = False
             if line:
                 if not line.startswith('#'):
                     if format_index and discontinuity_count != format_index:
@@ -168,13 +173,13 @@ class HlsFD(FragmentFD):
                     if extra_query:
                         frag_url = update_url_query(frag_url, extra_query)
 
-                    if real_downloader:
                     fragments.append({
+                        'frag_index': frag_index,
                         'url': frag_url,
                         'decrypt_info': decrypt_info,
+                        'byte_range': byte_range,
+                        'media_sequence': media_sequence,
                     })
-                        continue
-                    download_frag = True
 
                 elif line.startswith('#EXT-X-MAP'):
                     if format_index and discontinuity_count != format_index:
@@ -191,12 +196,14 @@ class HlsFD(FragmentFD):
                         else compat_urlparse.urljoin(man_url, map_info.get('URI')))
                     if extra_query:
                         frag_url = update_url_query(frag_url, extra_query)
-                    if real_downloader:
+
                     fragments.append({
+                        'frag_index': frag_index,
                         'url': frag_url,
                         'decrypt_info': decrypt_info,
+                        'byte_range': byte_range,
+                        'media_sequence': media_sequence
                     })
-                        continue
 
                     if map_info.get('BYTERANGE'):
                         splitted_byte_range = map_info.get('BYTERANGE').split('@')
@@ -205,7 +212,6 @@ class HlsFD(FragmentFD):
                             'start': sub_range_start,
                             'end': sub_range_start + int(splitted_byte_range[0]),
                         }
-                    download_frag = True
 
                 elif line.startswith('#EXT-X-KEY'):
                     decrypt_url = decrypt_info.get('URI')
@@ -236,8 +242,33 @@ class HlsFD(FragmentFD):
                     ad_frag_next = False
                 elif line.startswith('#EXT-X-DISCONTINUITY'):
                     discontinuity_count += 1
+                i += 1
+                media_sequence += 1
+
+        # We only download the first fragment during the test
+        if test:
+            fragments = [fragments[0] if fragments else None]
+
+        if real_downloader:
+            info_copy = info_dict.copy()
+            info_copy['fragments'] = fragments
+            fd = real_downloader(self.ydl, self.params)
+            # TODO: Make progress updates work without hooking twice
+            # for ph in self._progress_hooks:
+            #     fd.add_progress_hook(ph)
+            success = fd.real_download(filename, info_copy)
+            if not success:
+                return False
+        else:
+            def download_fragment(fragment):
+                frag_index = fragment['frag_index']
+                frag_url = fragment['url']
+                decrypt_info = fragment['decrypt_info']
+                byte_range = fragment['byte_range']
+                media_sequence = fragment['media_sequence']
+
+                ctx['fragment_index'] = frag_index
 
-                if download_frag:
                 count = 0
                 headers = info_dict.get('http_headers', {})
                 if byte_range:
@@ -247,7 +278,7 @@ class HlsFD(FragmentFD):
                         success, frag_content = self._download_fragment(
                             ctx, frag_url, info_dict, headers)
                         if not success:
-                                return False
+                            return False, frag_index
                         break
                     except compat_urllib_error.HTTPError as err:
                         # Unavailable (possibly temporary) fragments may be served.
@@ -258,14 +289,7 @@ class HlsFD(FragmentFD):
                         if count <= fragment_retries:
                             self.report_retry_fragment(err, frag_index, count, fragment_retries)
                 if count > fragment_retries:
-                        if skip_unavailable_fragments:
+                    return False, frag_index
-                            i += 1
-                            media_sequence += 1
-                            self.report_skip_fragment(frag_index)
-                            continue
-                        self.report_error(
-                            'giving up after %s fragment retries' % fragment_retries)
-                        return False
 
                 if decrypt_info['METHOD'] == 'AES-128':
                     iv = decrypt_info.get('IV') or compat_struct_pack('>8xq', media_sequence)
@@ -277,23 +301,65 @@ class HlsFD(FragmentFD):
                     if not test:
                         frag_content = AES.new(
                             decrypt_info['KEY'], AES.MODE_CBC, iv).decrypt(frag_content)
+
+                return frag_content, frag_index
+
+            def append_fragment(frag_content, frag_index):
+                if frag_content:
+                    fragment_filename = '%s-Frag%d' % (ctx['tmpfilename'], frag_index)
+                    try:
+                        file, frag_sanitized = sanitize_open(fragment_filename, 'rb')
+                        ctx['fragment_filename_sanitized'] = frag_sanitized
+                        file.close()
                         self._append_fragment(ctx, frag_content)
-                    # We only download the first fragment during the test
+                        return True
-                    if test:
+                    except FileNotFoundError:
-                        break
+                        if skip_unavailable_fragments:
-                    i += 1
+                            self.report_skip_fragment(frag_index)
-                    media_sequence += 1
+                            return True
+                        else:
+                            self.report_error(
+                                'fragment %s not found, unable to continue' % frag_index)
+                            return False
+                else:
+                    if skip_unavailable_fragments:
+                        self.report_skip_fragment(frag_index)
+                        return True
+                    else:
+                        self.report_error(
+                            'fragment %s not found, unable to continue' % frag_index)
+                        return False
 
-        if real_downloader:
+            max_workers = self.params.get('concurrent_fragment_downloads', 1)
-            info_copy = info_dict.copy()
+            if can_threaded_download and max_workers > 1:
-            info_copy['fragments'] = fragments
+                self.report_warning('The download speed shown is only of one thread. This is a known issue')
-            fd = real_downloader(self.ydl, self.params)
+                with concurrent.futures.ThreadPoolExecutor(max_workers) as pool:
-            # TODO: Make progress updates work without hooking twice
+                    futures = [pool.submit(download_fragment, fragment) for fragment in fragments]
-            # for ph in self._progress_hooks:
+                    # timeout must be 0 to return instantly
-            #     fd.add_progress_hook(ph)
+                    done, not_done = concurrent.futures.wait(futures, timeout=0)
-            success = fd.real_download(filename, info_copy)
+                    try:
-            if not success:
+                        while not_done:
+                            # Check every 1 second for KeyboardInterrupt
+                            freshly_done, not_done = concurrent.futures.wait(not_done, timeout=1)
+                            done |= freshly_done
+                    except KeyboardInterrupt:
+                        for future in not_done:
+                            future.cancel()
+                        # timeout must be none to cancel
+                        concurrent.futures.wait(not_done, timeout=None)
+                        raise KeyboardInterrupt
+                results = [future.result() for future in futures]
+
+                for frag_content, frag_index in results:
+                    result = append_fragment(frag_content, frag_index)
+                    if not result:
                         return False
             else:
+                for fragment in fragments:
+                    frag_content, frag_index = download_fragment(fragment)
+                    result = append_fragment(frag_content, frag_index)
+                    if not result:
+                        return False
+
             self._finish_frag_download(ctx)
         return True

yt_dlp/options.py

@@ -558,6 +558,10 @@ def parseOpts(overrideArguments=None):
         help='Languages of the subtitles to download (optional) separated by commas, use --list-subs for available language tags')
 
     downloader = optparse.OptionGroup(parser, 'Download Options')
+    downloader.add_option(
+        '-N', '--concurrent-fragments',
+        dest='concurrent_fragment_downloads', metavar='N', default=1, type=int,
+        help='Number of fragments to download concurrently (default is %default)')
     downloader.add_option(
         '-r', '--limit-rate', '--rate-limit',
         dest='ratelimit', metavar='RATE',