aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorSergey M․ <dstftw@gmail.com>2021-02-17 04:03:54 +0700
committerSergey M․ <dstftw@gmail.com>2021-02-17 04:03:54 +0700
commita363fb5d28da7c1b651e6de98b9e799544a4df73 (patch)
tree255d5c7beb7e695cd9d537a033d3842099d1e506
parent646052e416577cc805b7ba169c49158716541570 (diff)
[yandexmusic:playlist] Request missing tracks in chunks (closes #27355, closes #28184)
-rw-r--r--youtube_dl/extractor/yandexmusic.py35
1 files changed, 23 insertions, 12 deletions
diff --git a/youtube_dl/extractor/yandexmusic.py b/youtube_dl/extractor/yandexmusic.py
index 7893f363e..84969f8e1 100644
--- a/youtube_dl/extractor/yandexmusic.py
+++ b/youtube_dl/extractor/yandexmusic.py
@@ -1,8 +1,9 @@
# coding: utf-8
from __future__ import unicode_literals
-import re
import hashlib
+import itertools
+import re
from .common import InfoExtractor
from ..compat import compat_str
@@ -209,17 +210,27 @@ class YandexMusicPlaylistBaseIE(YandexMusicBaseIE):
missing_track_ids = [
track_id for track_id in track_ids
if track_id not in present_track_ids]
- missing_tracks = self._call_api(
- 'track-entries', tld, url, item_id,
- 'Downloading missing tracks JSON', {
- 'entries': ','.join(missing_track_ids),
- 'lang': tld,
- 'external-domain': 'music.yandex.%s' % tld,
- 'overembed': 'false',
- 'strict': 'true',
- })
- if missing_tracks:
- tracks.extend(missing_tracks)
+ # Request missing tracks in chunks to avoid exceeding max HTTP header size,
+ # see https://github.com/ytdl-org/youtube-dl/issues/27355
+ _TRACKS_PER_CHUNK = 250
+ for chunk_num in itertools.count(0):
+ start = chunk_num * _TRACKS_PER_CHUNK
+ end = start + _TRACKS_PER_CHUNK
+ missing_track_ids_req = missing_track_ids[start:end]
+ assert missing_track_ids_req
+ missing_tracks = self._call_api(
+ 'track-entries', tld, url, item_id,
+ 'Downloading missing tracks JSON chunk %d' % (chunk_num + 1), {
+ 'entries': ','.join(missing_track_ids_req),
+ 'lang': tld,
+ 'external-domain': 'music.yandex.%s' % tld,
+ 'overembed': 'false',
+ 'strict': 'true',
+ })
+ if missing_tracks:
+ tracks.extend(missing_tracks)
+ if end >= len(missing_track_ids):
+ break
return tracks